RE: [agi] SOTA
No, and it's a damn good thing it isn't. If it was we would be sentencing it to a mindless job with no time off, only to be disposed of when a better model comes out. We only want our AI's to be a smart as necessary to accomplish their jobs just as our cells and organs are. Limited conciousness or self-reflectivity may only be necessary in highly complex systems like computers where we may want them to recognize that they have a virus and take steps like searching for a digital vaccine to eliminate it without the owner even knowing it was there. Even in these cases we are only giving the system conciousness over one specific aspect of it's being. I would say that until we have software that can learn new free format information as we do and modify it's goal stack based upon that new information then we do not have a truly concious computer. _ From: Bob Mottram [mailto:[EMAIL PROTECTED] Sent: Friday, January 12, 2007 9:45 AM To: agi@v2.listbox.com Subject: Re: [agi] SOTA Ah, but is a thermostat conscious ? :-) On 12/01/07, [EMAIL PROTECTED] [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: http://www.thermostatshop.com/ Not sure what you've been Googling on but here they are. There's even one you can call on the telephone If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Well, there's no reason to stop at a purely utilitarian level. A high level of consciousness may be necessary for performing certain kinds of task, such as imagining someone's reaction to a particular event. On 14/01/07, Gary Miller [EMAIL PROTECTED] wrote: No, and it's a damn good thing it isn't. If it was we would be sentencing it to a mindless job with no time off, only to be disposed of when a better model comes out. We only want our AI's to be a smart as necessary to accomplish their jobs just as our cells and organs are. Limited conciousness or self-reflectivity may only be necessary in highly complex systems like computers where we may want them to recognize that they have a virus and take steps like searching for a digital vaccine to eliminate it without the owner even knowing it was there. Even in these cases we are only giving the system conciousness over one specific aspect of it's being. I would say that until we have software that can learn new free format information as we do and modify it's goal stack based upon that new information then we do not have a truly concious computer. -- *From:* Bob Mottram [mailto:[EMAIL PROTECTED] *Sent:* Friday, January 12, 2007 9:45 AM *To:* agi@v2.listbox.com *Subject:* Re: [agi] SOTA Ah, but is a thermostat conscious ? :-) On 12/01/07, [EMAIL PROTECTED] [EMAIL PROTECTED] wrote: http://www.thermostatshop.com/ Not sure what you've been Googling on but here they are. There's even one you can call on the telephone If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. -- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 -- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 -- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Ah, but is a thermostat conscious ? :-) On 12/01/07, [EMAIL PROTECTED] [EMAIL PROTECTED] wrote: http://www.thermostatshop.com/ Not sure what you've been Googling on but here they are. There's even one you can call on the telephone If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. -- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/12/07, [EMAIL PROTECTED] [EMAIL PROTECTED] wrote: http://www.thermostatshop.com/ Not sure what you've been Googling on but here they are. Haven't been googling. But the fact is that I've never actually /seen/ one in the wild. My point is that the market demand for such simple and useful and cheap items is low enough that I've never actually seen one. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Philip Goetz wrote: Haven't been googling. But the fact is that I've never actually /seen/ one in the wild. My point is that the market demand for such simple and useful and cheap items is low enough that I've never actually seen one. Check any hardware store, there's a whole shelf. I bought one for my last apartment. I see them all over the place. They're really not rare. Moral: in AI, the state of the art is often advanced far beyond what people think it is. -- Eliezer S. Yudkowsky http://singinst.org/ Research Fellow, Singularity Institute for Artificial Intelligence - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
--- Bob Mottram [EMAIL PROTECTED] wrote: Ah, but is a thermostat conscious ? :-) Are humans conscious? It depends on your definition of consciousness, which is really hard to define. Does a thermostat want to keep the room at a constant temperature? Or does it just behave as if that is what it wants? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/12/07, Eliezer S. Yudkowsky [EMAIL PROTECTED] wrote: Philip Goetz wrote: Haven't been googling. But the fact is that I've never actually /seen/ one in the wild. My point is that the market demand for such simple and useful and cheap items is low enough that I've never actually seen one. The term for this type of thermostat is a 'set-back' thermostat, and they were originally designed to save energy and heating/cooling bills by having programmable periods. They have become increasingly complex. All my most recent houses had essentially little calendar computers in them. They are extremely common in new construction, as this link shows: http://www.marketresearch.com/product/display.asp?productid=1354170g=1 STUDY HIGHLIGHTS - Honeywell/Magicstat was the top brand of thermostat bought in 2003. - Two-thirds of the thermostats purchased in 2003 were set-back models. - The average price paid for the electronic set-back thermostats was $70. - Thermostats were purchased mostly from builders/contractors and home centers. Check any hardware store, there's a whole shelf. I bought one for my last apartment. I see them all over the place. They're really not rare. Moral: in AI, the state of the art is often advanced far beyond what people think it is. There are really two things being talked about here. One is SOTA, which almost by definition, is beyond what people think it is, and the other is market availability, or practical availability, which is very different than SOTA technology. SOTA AI technology is essentially that which you, knowing the latest theories, build yourself. There is no such thing as a SOTA AI system, the way there are SOTA stereo systems, or SOTA crypto systems available, because the market availability of the technology does not have the same characteristics. -- Justin Corwin [EMAIL PROTECTED] http://outlawpoet.blogspot.com http://www.adaptiveai.com - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 06/01/07, Gary Miller [EMAIL PROTECTED] wrote: I like the idea of the house being the central AI though and communicating to house robots through an wireless encrypted protocol to prevent inadvertant commands from other systems and hacking. This is the way it's going to go in my opinion. In a house or office the robots would really be dumb actuators - puppets - being controlled from a central AI which integrates multiple systems together. That way you can keep the cost and maintenance requirements of the robot to a bare minimum. Such a system also future-proofs the robot in a rapidly changing software world, and allows intelligence to be provided as an internet based service. If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. Huh? I've never lived in a home without such (nor been aware that they were rare). - Original Message - From: Philip Goetz [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, January 11, 2007 1:29 PM Subject: Re: [agi] SOTA On 06/01/07, Gary Miller [EMAIL PROTECTED] wrote: I like the idea of the house being the central AI though and communicating to house robots through an wireless encrypted protocol to prevent inadvertant commands from other systems and hacking. This is the way it's going to go in my opinion. In a house or office the robots would really be dumb actuators - puppets - being controlled from a central AI which integrates multiple systems together. That way you can keep the cost and maintenance requirements of the robot to a bare minimum. Such a system also future-proofs the robot in a rapidly changing software world, and allows intelligence to be provided as an internet based service. If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On Jan 11, 2007, at 1:29 PM, Philip Goetz wrote: On 06/01/07, Gary Miller [EMAIL PROTECTED] wrote: This is the way it's going to go in my opinion. In a house or office the robots would really be dumb actuators - puppets - being controlled from a central AI which integrates multiple systems together. That way you can keep the cost and maintenance requirements of the robot to a bare minimum. Such a system also future-proofs the robot in a rapidly changing software world, and allows intelligence to be provided as an internet based service. If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. http://www.google.com/search?q=programmable+thermostat They're extremely common; there's an entire aisle of such things at my local Home Depot. -- Randall Randall [EMAIL PROTECTED] If you are trying to produce a commercial product in a timely and cost efficient way, it is not good to have somebody's PhD research on your critical path. -- Chip Morningstar - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On Jan 11, 2007, at 10:29 AM, Philip Goetz wrote: If there's a market for this, then why can't I even buy a thermostat with a timer on it to turn the temperature down at night and up in the morning? The most basic home automation, which could have been built cheaply 30 years ago, is still, if available at all, so rare that I've never seen it. What country do you live in? That thermostat technology has been ubiquitous in the US for new construction for many years. Granted, older buildings tend to have the thermostats opportunistically replaced, but it has been a lot of years and a couple places since I didn't live in a place with this type of thermostat. Apparently, the future has arrived and just is not evenly distributed. J. Andrew Rogers - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 06/01/07, Philip Goetz [EMAIL PROTECTED] wrote: I worked for a robotics company called Arctec in the early 1980s. We built a robot called the Gemini. They essentially solved the navigation problem - in an office-space world. You stuck one small reflector on both sides of every door, at intersections, and on its recharging station, and if the world consisted of hallways with doors leading off the hallways, it went where you told it to go, and went back to its docking station and recharged when it had to. This was using a 1MHz 65C02. The navigational code took 12K of RAM. It relied largely on having highly-accurate narrow-beam sonar sensors all around its body, and on the reflectors. Reflectors have been used on AGVs for quite some time. However, even using reflectors the robot has no real idea of what its environment looks like. Most of the time it's flying blind, guessing its way between reflectors, like a moth navigating by the light of the moon. In my opinion to make real progress the machine needs to be able to see the three dimensional structure of its environment at least as well as you or I can. Only then can it make more intelligent decisions about what to do. The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. This is something which I've also considered. When you look at a dumb robot like a Roomba, the technology to build this fairly economically has existed for something like the last 25 years. So why didn't these devices appear much earlier? I think for a robot product to be successfull not only does the technology need to be there, but also the cultural attitude. Even now robot startup companies such as White Box Robotics seem to have little idea of what their machines might actually be used for. The application of last resort is always security, but this is a very poor use for a robot in my opinion. Security is better and more economically done with a dissembodied intelligence using fixed cameras, a la 2001. - Bob - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. Phil, I think the real issue is that no one wants an expensive, stupid, awkward robot... A broadly functional household robot would be very useful, even if it lacked intelligence beyond the human-moron level... For instance, right now, I would like a robot to go into my daughter's room and clean up the rabbit turds that are in the rabbit playpen in there. I would rather not do it. But, a Roomba can't handle this task because it can't climb over the walls of the playpen, nor understand my instructions, nor pick up the turds but leave the legos on the floor alone... Heck, a robot to let the dogs in and out of the house would be nice too... being doggie doorman gets tiring. Of course, this could be solved more easily by installing a doggie door ;-) How about a robot to bring me the cordless phone when it rings, but has been left somewhere else in the house ... ? ;-) How about one to put the dishes in the dishwasher and unload them ... and re-insert the ones that didn't get totally cleaned? The dishwasher is a good invention but it only does half the job The problem **is** technological: it's that current robots really suck ... not that non-sucky robots would be useles... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Stanford scientists plan to make a robot capable of performing everyday tasks, such as unloading the dishwasher. http://news-service.stanford.edu/news/2006/november8/ng-110806.html On 1/6/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. Phil, I think the real issue is that no one wants an expensive, stupid, awkward robot... A broadly functional household robot would be very useful, even if it lacked intelligence beyond the human-moron level... For instance, right now, I would like a robot to go into my daughter's room and clean up the rabbit turds that are in the rabbit playpen in there. I would rather not do it. But, a Roomba can't handle this task because it can't climb over the walls of the playpen, nor understand my instructions, nor pick up the turds but leave the legos on the floor alone... Heck, a robot to let the dogs in and out of the house would be nice too... being doggie doorman gets tiring. Of course, this could be solved more easily by installing a doggie door ;-) How about a robot to bring me the cordless phone when it rings, but has been left somewhere else in the house ... ? ;-) How about one to put the dishes in the dishwasher and unload them ... and re-insert the ones that didn't get totally cleaned? The dishwasher is a good invention but it only does half the job The problem **is** technological: it's that current robots really suck ... not that non-sucky robots would be useles... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Needless to say, I don't consider cleaning up the house a particularly interesting goal for AGI projects. I can well imagine it being done by a narrow AI system with no capability to do anything besides manipulate simple objects, navigate, etc. Being able to understand natural language commands pertaining to cleaning up the house is a whole other kettle of fish, of course. This, as opposed to the actual house-cleaning, appears to be an AGI-hard problem... -- BenG On 1/6/07, Pei Wang [EMAIL PROTECTED] wrote: Stanford scientists plan to make a robot capable of performing everyday tasks, such as unloading the dishwasher. http://news-service.stanford.edu/news/2006/november8/ng-110806.html On 1/6/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. Phil, I think the real issue is that no one wants an expensive, stupid, awkward robot... A broadly functional household robot would be very useful, even if it lacked intelligence beyond the human-moron level... For instance, right now, I would like a robot to go into my daughter's room and clean up the rabbit turds that are in the rabbit playpen in there. I would rather not do it. But, a Roomba can't handle this task because it can't climb over the walls of the playpen, nor understand my instructions, nor pick up the turds but leave the legos on the floor alone... Heck, a robot to let the dogs in and out of the house would be nice too... being doggie doorman gets tiring. Of course, this could be solved more easily by installing a doggie door ;-) How about a robot to bring me the cordless phone when it rings, but has been left somewhere else in the house ... ? ;-) How about one to put the dishes in the dishwasher and unload them ... and re-insert the ones that didn't get totally cleaned? The dishwasher is a good invention but it only does half the job The problem **is** technological: it's that current robots really suck ... not that non-sucky robots would be useles... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303 - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/6/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: Needless to say, I don't consider cleaning up the house a particularly interesting goal for AGI projects. I can well imagine it being done by a narrow AI system with no capability to do anything besides manipulate simple objects, navigate, etc. Being able to understand natural language commands pertaining to cleaning up the house is a whole other kettle of fish, of course. This, as opposed to the actual house-cleaning, appears to be an AGI-hard problem... But if the AGI were built, wouldn't it be the intelligence to pretty much the entire world of human-moron level housecleaning robots? All they'd need is wifi to get instructions from the main brain. But then a real AGI would likely become the main brain for just about ever process control program we use, so the term quickly changes from human-moron level to human-level moron. :) I really want to see a central traffic computer take driving away from all the unqualified (or disinterested) drivers on the roads. I'd really like to see companies get incentives to allow knowledge workers work from home offices to save commute time and fuel resources, but until that happens (yeah, the employer wants to give up their sense of control?) it would be nice to reclaim that time by allowing me to focus on what *I* want rather than on driving. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/6/07, Bob Mottram [EMAIL PROTECTED] wrote: Reflectors have been used on AGVs for quite some time. However, even using reflectors the robot has no real idea of what its environment looks like. Most of the time it's flying blind, guessing its way between reflectors, like a moth navigating by the light of the moon. In my opinion to make real progress the machine needs to be able to see the three dimensional structure of its environment at least as well as you or I can. Only then can it make more intelligent decisions about what to do. The robot navigated successfully around a home or office, avoiding obstacles, at walking speed. What more do you need? I think it performed so much better than the robots developed in academia partly because we didn't have enough processing power to use a camera, so we weren't sucked into that black hole (at the time) of trying to process vision. Even now robot startup companies such as White Box Robotics seem to have little idea of what their machines might actually be used for. The application of last resort is always security, but this is a very poor use for a robot in my opinion. Security is better and more economically done with a dissembodied intelligence using fixed cameras, a la 2001. We had a customer who wanted us to attach a gun to the robot. Probably not a programmer. :P - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
RE: [agi] SOTA
Ben Said:
Being able to understand natural language commands pertaining
to cleaning up the house is a whole other kettle of fish, of
course. This, as opposed to the actual house-cleaning, appears
to be an AGI-hard problem...
A full Turing complete Natural Language system would not be necessary
for robotic control.
A pattern such as {clean|sweep|vacuum} (the )[RoomName]room( {for|in}
[Number] minutes)
When coupled with a voice recogniton system such as Nuance is marketing
would
increase the usefulness and interactivity of the robot immensely.
[RoomName] and [Number] become variables passed to the robot vacuum and can
have defaults
If omitted from the command.
The robot could come with a set of canned patterns for starters and the
patterns could
by customized by the user and associated with new behaviours.
I like the idea of the house being the central AI though and communicating
to
house robots through an wireless encrypted protocol to prevent inadvertant
commands from other systems and hacking.
If the robots were made to a standard such as Microsoft's robotics toolkit
then a single control and monitoring system could coordinate multiple robots
activities and prevent collisions, coordinate efforts, etc... You'd
probably need
a backup fault tolerent system to prevent loss of critical systems like
security,
fire reporting, and temperature control.
The house AI would be interfaced with the telephone and internet so that you
could
enter remote commands if you think of something while you're away.
-Original Message-
From: Benjamin Goertzel [mailto:[EMAIL PROTECTED]
Sent: Saturday, January 06, 2007 11:16 AM
To: agi@v2.listbox.com
Subject: Re: [agi] SOTA
Needless to say, I don't consider cleaning up the house a particularly
interesting goal for AGI projects. I can well imagine it being done by a
narrow AI system with no capability to do anything besides manipulate simple
objects, navigate, etc.
Being able to understand natural language commands pertaining to cleaning up
the house is a whole other kettle of fish, of course.
This, as opposed to the actual house-cleaning, appears to be an AGI-hard
problem...
-- BenG
On 1/6/07, Pei Wang [EMAIL PROTECTED] wrote:
Stanford scientists plan to make a robot capable of performing
everyday tasks, such as unloading the dishwasher.
http://news-service.stanford.edu/news/2006/november8/ng-110806.html
On 1/6/07, Benjamin Goertzel [EMAIL PROTECTED] wrote:
The problem wasn't technological. It was that nobody had any use
for a robot. We never figured out what people would want the robot
for.
I think that's still the problem.
Phil, I think the real issue is that no one wants an expensive,
stupid, awkward robot...
A broadly functional household robot would be very useful, even if
it lacked intelligence beyond the human-moron level...
For instance, right now, I would like a robot to go into my
daughter's room and clean up the rabbit turds that are in the rabbit
playpen in there. I would rather not do it. But, a Roomba can't
handle this task because it can't climb over the walls of the
playpen, nor understand my instructions, nor pick up the turds but
leave the legos on the floor alone...
Heck, a robot to let the dogs in and out of the house would be nice
too... being doggie doorman gets tiring. Of course, this could be
solved more easily by installing a doggie door ;-)
How about a robot to bring me the cordless phone when it rings, but
has been left somewhere else in the house ... ? ;-)
How about one to put the dishes in the dishwasher and unload them ...
and re-insert the ones that didn't get totally cleaned? The
dishwasher is a good invention but it only does half the job
The problem **is** technological: it's that current robots really
suck ... not that non-sucky robots would be useles...
-- Ben
-
This list is sponsored by AGIRI: http://www.agiri.org/email To
unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-
This list is sponsored by AGIRI: http://www.agiri.org/email To
unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-
This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe
or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 06/01/07, Mike Dougherty [EMAIL PROTECTED] wrote: I really want to see a central traffic computer take driving away from all the unqualified (or disinterested) drivers on the roads. I'd really like to see companies get incentives to allow knowledge workers work from home offices to save commute time and fuel resources, but until that happens (yeah, the employer wants to give up their sense of control?) it would be nice to reclaim that time by allowing me to focus on what *I* want rather than on driving. I agree on the driving thing. One day people will laugh about the times when you once had to get behind a steering wheel and wrestle with it, and how human driving was so unreliable that people had to be insured against the results of their own incompetence. The most optimistic pundits believe that it will be ten years from Stanley to the highway. There could be numerous non-technical hurdles (both legal and insurance related) which might delay progress further. I expect to see the first autonomous road system in some relatively quiet area of the world, where there are few (or no) traffic regulations, and no insurance companies to complain. Maybe some country in Africa or a province of China will be the first to go autonomous. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 06/01/07, Gary Miller [EMAIL PROTECTED] wrote: I like the idea of the house being the central AI though and communicating to house robots through an wireless encrypted protocol to prevent inadvertant commands from other systems and hacking. This is the way it's going to go in my opinion. In a house or office the robots would really be dumb actuators - puppets - being controlled from a central AI which integrates multiple systems together. That way you can keep the cost and maintenance requirements of the robot to a bare minimum. Such a system also future-proofs the robot in a rapidly changing software world, and allows intelligence to be provided as an internet based service. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/6/07, Bob Mottram wrote: This is the way it's going to go in my opinion. In a house or office the robots would really be dumb actuators - puppets - being controlled from a central AI which integrates multiple systems together. That way you can keep the cost and maintenance requirements of the robot to a bare minimum. Such a system also future-proofs the robot in a rapidly changing software world, and allows intelligence to be provided as an internet based service. http://www.pinktentacle.com/2006/12/top-10-robots-selected-for-robot-award-2006/ Robotic building cleaning system (Fuji Heavy Industries/ Sumitomo) - This autonomous robot roams the hallways of buildings, performing cleaning operations along the way. Capable of controlling elevators, the robot can move from floor to floor unsupervised, and it returns to its start location once it has finished cleaning. The robot is currently employed as a janitor at 10 high-rise buildings in Japan, including Harumi Triton Square and Roppongi Hills. BillK - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 10/23/06, Bob Mottram [EMAIL PROTECTED] wrote: Grinding my own axe, I also think that stereo vision systems will bring significant improvements to robotics over the next few years. Being able to build videogame-like 3D models of the environment in real time is now a feasible proposition which I think will happen before the decade is out. With a good model of the environment the robot can rehearse possible scenarios before actually running them, and find important features such as desk or table surfaces. I worked for a robotics company called Arctec in the early 1980s. We built a robot called the Gemini. They essentially solved the navigation problem - in an office-space world. You stuck one small reflector on both sides of every door, at intersections, and on its recharging station, and if the world consisted of hallways with doors leading off the hallways, it went where you told it to go, and went back to its docking station and recharged when it had to. This was using a 1MHz 65C02. The navigational code took 12K of RAM. It relied largely on having highly-accurate narrow-beam sonar sensors all around its body, and on the reflectors. The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
On 1/6/07, Philip Goetz [EMAIL PROTECTED] wrote: The problem wasn't technological. It was that nobody had any use for a robot. We never figured out what people would want the robot for. I think that's still the problem. Well, I for one want a job assistant who can fetch things - what apprentices or surgical nurse-assistanty things are often called to do. Assistant: Please get me a Phillips head screwdriver and half-a-dozen 10mm screws A robot that could 1) Voice recognise instructions 2) Understand simple commands like Get me X, Hold this still, Return this... 3) Manoeuvre from your work space to your tool-store 4) Grab items from an appropriately set-up tool-store etc Would be pretty damn useful, and I see most of this as being feasible with current day tech. Sure, such an assistant would be pretty damn expensive, and less useful than a high-school-dropout apprentice/assistant (who can also run down the street and get you a sandwich), but this is a real, possible application for a robot. -- Olie - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
Re: [agi] SOTA
Another recent development is the CMU telepresence robot, which is quite low cost and would be a good place to start. Since it uses a linux based PC there should be plenty of scope for programming more sophisticated applications than Lego would be able to handle. http://www.terk.ri.cmu.edu/recipes/index.phpAlthough its intended to be used for education I'm sure a ruggedised version of this could have industrial or home uses. Al the software at present is open source. For a more commercial system intelligence would be supplied to the robot via a web based subscription service, and could be purely human, purely AI, or a mixture of the two. Once you have people driving robots around via the internet you can bring your data mining systems to bear and start to automate some of that human intelligence. On 24/10/06, Pei Wang [EMAIL PROTECTED] wrote: Bob and Neil,Thanks for the informative discussion!Several questions for you and others who are familiar with robotics:For people whose interests are mainly in the connection betweensensorimotor and high-level cognition, what kind of API can be expected in a representative robot? Something like Tekkotsu?Any comments on Microsoft Robotics Studio?Recently some people are talking about cognitive robotics, though Ihaven't found any major new idea beyond what robotics has been covering, except the suggestion that high-level cognition should betaken into consideration. Am I missing something important?If I want to start to try some low-budget programmable robot (say, inthe price range of Robosapien V2 and LEGO Mindstorms NXT), which one will you recommend? I won't have high expectation in performance, butwill be interested in testing ideas on the coordination of perception,reasoning, learning, and action.PeiOn 10/24/06, Bob Mottram [EMAIL PROTECTED] wrote: On 23/10/06, Neil H. [EMAIL PROTECTED] wrote: I'm also pretty surprise that they haven't done anything major with their vSLAM tech: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1570091 Evolution really failed to capitalise upon their early success.One of their biggest mistakes was to make their API prohibiively expensive, so that few people have ever used it.There was a command line API supplied by default, but that was dreadful and had major limitations which users of the robots have long complained about. The vSLAM technology is a monocular SLAM method which works to an extent, but when it fails it fails catastrophically.They did experiment with stereo SLAM last year (there is a paper somewhere online about that). Stereo gives much better accuracy and doesn't require the robot to travel for at least one metre before it can localise, but there are some fundamental issues with using things like SIFT features for doing stereo which they probably didn't realise. I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Sony licensed the tech, but I think they only used it so that AIBO could visually recognize pre-printed patterns on cards, which would signal the AIBO to dance, return to the charging station, etc. SIFT is IMHO overkill for that kind of thing, and it's a pity they didn't do anything more interesting with it. It's a shame they ditched AIBO and their other robots in development.AIBO users were rather unhappy about that.Perhaps some other company will buy the rights. Perhaps. To play devil's advocate, how well do you think stereo vision system would actually work for creating a 3D structure of a home environment? It seems that distinctive features in the home tend to be few and far between. Of course, the regions between distinctive features tend to be planar surfaces, so perhaps it isn't too bad. Well this is exactly what I'm (unofficially) working on now.From the results I have at the moment I can say with confidence that it will be possible to navigate a robot around a home environment using a pair of stereo cameras, with the robot remaining within at least a 7cm position tollerance.7cm is just a raw localisation figure, and after kalman filtering and sensor fusion with odometry the accuracy should be much better than that.You might think that there are not many features on walls, but even in environments which people consider to be blank there are often small imperfections or shading gradients which stereo algorithms can pick up.In real life few surfaces are perfectly uniform. With good localisation performance high quality mapping becomes possible.I can run the stereo algorithms at various levels of detail, and use traditional occupancy grid methods (with a few tweaks) to build up evidence in a probablistic fashion.The idea at the moment is to have the localisation algorithms running in real time using low-res grids, and to build a separate high quality model of the environment in a high resolution grid more gradually in a low priority background task.Once you have a good quality grid model its then quite
Re: [agi] SOTA
On 23/10/06, Neil H. [EMAIL PROTECTED] wrote: I'm also pretty surprise that they haven't done anything major withtheir vSLAM tech:http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1570091 Evolution really failed to capitalise upon their early success. One of their biggest mistakes was to make their API prohibiively expensive, so that few people have ever used it. There was a command line API supplied by default, but that was dreadful and had major limitations which users of the robots have long complained about. The vSLAM technology is a monocular SLAM method which works to an extent, but when it fails it fails catastrophically. They did experiment with stereo SLAM last year (there is a paper somewhere online about that). Stereo gives much better accuracy and doesn't require the robot to travel for at least one metre before it can localise, but there are some fundamental issues with using things like SIFT features for doing stereo which they probably didn't realise. I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Sony licensed the tech, but I think they only used it so that AIBOcould visually recognize pre-printed patterns on cards, which wouldsignal the AIBO to dance, return to the charging station, etc. SIFT is IMHO overkill for that kind of thing, and it's a pity they didn't doanything more interesting with it.It's a shame they ditched AIBO and their other robots in development. AIBO users were rather unhappy about that. Perhaps some other company will buy the rights. Perhaps. To play devil's advocate, how well do you think stereo vision system would actually work for creating a 3D structure of a homeenvironment? It seems that distinctive features in the home tend to befew and far between. Of course, the regions between distinctivefeatures tend to be planar surfaces, so perhaps it isn't too bad. Well this is exactly what I'm (unofficially) working on now. From the results I have at the moment I can say with confidence that it will be possible to navigate a robot around a home environment using a pair of stereo cameras, with the robot remaining within at least a 7cm position tollerance. 7cm is just a raw localisation figure, and after kalman filtering and sensor fusion with odometry the accuracy should be much better than that. You might think that there are not many features on walls, but even in environments which people consider to be blank there are often small imperfections or shading gradients which stereo algorithms can pick up. In real life few surfaces are perfectly uniform. With good localisation performance high quality mapping becomes possible. I can run the stereo algorithms at various levels of detail, and use traditional occupancy grid methods (with a few tweaks) to build up evidence in a probablistic fashion. The idea at the moment is to have the localisation algorithms running in real time using low-res grids, and to build a separate high quality model of the environment in a high resolution grid more gradually in a low priority background task. Once you have a good quality grid model its then quite straightforward to detect things like walls and furniture, and to simplify the data down to something which is a more efficient representation similar to something you might find in a game or an AGI sim. You can also use the grid model in exactly the same way that 2D background subtraction systems work (except in 3D) in order to detect changes within the environment. This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/20/06, Richard Loosemore wrote: I would *love* to see those IBM folks put a couple of jabbering four-year-old children in front of that translation system, to see how it likes their 'low-intelligence' language. :-) Does anyone have any contacts on the team, so we could ask? I sent an email to Liang Gu on the IBM MASTOR project team (not really expecting a reply) :) and have just received this response. Sounds hopeful. BillK - Bill, Thanks for your interests on MASTOR. And your suggestion of MASTOR for Children is really great! It is definitely much more meaningful if MASTOR can not only help adults but also children communicate with each other around the world using different languages! Although recognizing Children's voice has been proved a very challenging task, the translation and text-to-speech techniques thus involved should be very similar to what we have now. We will seriously investigate the possibility of this approach and will send you a test link if we later developed a pilot system on the web. Regards and thanks again for your enthusiasm about MASTOR, Liang - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/24/06, Bob Mottram [EMAIL PROTECTED] wrote: On 23/10/06, Neil H. [EMAIL PROTECTED] wrote: I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Sony licensed the tech, but I think they only used it so that AIBO could visually recognize pre-printed patterns on cards, which would signal the AIBO to dance, return to the charging station, etc. SIFT is IMHO overkill for that kind of thing, and it's a pity they didn't do anything more interesting with it. It's a shame they ditched AIBO and their other robots in development. AIBO users were rather unhappy about that. Perhaps some other company will buy the rights. Not just end-users, but there were also a number of research labs which used AIBOs as a robotics platform. (I was in such a lab as an undergrad) They were pretty nice for running software on, and it was more-or-less impossible to get a robot with similar capabilities at that $1000 price range. Somewhat surprisingly, the RoboCup four-legged league is still active, with 24 teams qualifying to compete this year: http://www.robocup2006.org/sixcms/detail.php?id=390lang=en I wonder how they keep their AIBOs operational over the years... Perhaps. To play devil's advocate, how well do you think stereo vision system would actually work for creating a 3D structure of a home environment? It seems that distinctive features in the home tend to be few and far between. Of course, the regions between distinctive features tend to be planar surfaces, so perhaps it isn't too bad. Well this is exactly what I'm (unofficially) working on now. From the results I have at the moment I can say with confidence that it will be possible to navigate a robot around a home environment using a pair of stereo cameras, with the robot remaining within at least a 7cm position tollerance. 7cm is just a raw localisation figure, and after kalman filtering and sensor fusion with odometry the accuracy should be much better than that. You might think that there are not many features on walls, but even in environments which people consider to be blank there are often small imperfections or shading gradients which stereo algorithms can pick up. In real life few surfaces are perfectly uniform. With good localisation performance high quality mapping becomes possible. I can run the stereo algorithms at various levels of detail, and use traditional occupancy grid methods (with a few tweaks) to build up evidence in a probablistic fashion. The idea at the moment is to have the localisation algorithms running in real time using low-res grids, and to build a separate high quality model of the environment in a high resolution grid more gradually in a low priority background task. Once you have a good quality grid model its then quite straightforward to detect things like walls and furniture, and to simplify the data down to something which is a more efficient representation similar to something you might find in a game or an AGI sim. You can also use the grid model in exactly the same way that 2D background subtraction systems work (except in 3D) in order to detect changes within the environment. This a pretty interesting approach. I'd love to see more details on this in the future. -- Neil - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Bob and Neil, Thanks for the informative discussion! Several questions for you and others who are familiar with robotics: For people whose interests are mainly in the connection between sensorimotor and high-level cognition, what kind of API can be expected in a representative robot? Something like Tekkotsu? Any comments on Microsoft Robotics Studio? Recently some people are talking about cognitive robotics, though I haven't found any major new idea beyond what robotics has been covering, except the suggestion that high-level cognition should be taken into consideration. Am I missing something important? If I want to start to try some low-budget programmable robot (say, in the price range of Robosapien V2 and LEGO Mindstorms NXT), which one will you recommend? I won't have high expectation in performance, but will be interested in testing ideas on the coordination of perception, reasoning, learning, and action. Pei On 10/24/06, Bob Mottram [EMAIL PROTECTED] wrote: On 23/10/06, Neil H. [EMAIL PROTECTED] wrote: I'm also pretty surprise that they haven't done anything major with their vSLAM tech: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1570091 Evolution really failed to capitalise upon their early success. One of their biggest mistakes was to make their API prohibiively expensive, so that few people have ever used it. There was a command line API supplied by default, but that was dreadful and had major limitations which users of the robots have long complained about. The vSLAM technology is a monocular SLAM method which works to an extent, but when it fails it fails catastrophically. They did experiment with stereo SLAM last year (there is a paper somewhere online about that). Stereo gives much better accuracy and doesn't require the robot to travel for at least one metre before it can localise, but there are some fundamental issues with using things like SIFT features for doing stereo which they probably didn't realise. I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Sony licensed the tech, but I think they only used it so that AIBO could visually recognize pre-printed patterns on cards, which would signal the AIBO to dance, return to the charging station, etc. SIFT is IMHO overkill for that kind of thing, and it's a pity they didn't do anything more interesting with it. It's a shame they ditched AIBO and their other robots in development. AIBO users were rather unhappy about that. Perhaps some other company will buy the rights. Perhaps. To play devil's advocate, how well do you think stereo vision system would actually work for creating a 3D structure of a home environment? It seems that distinctive features in the home tend to be few and far between. Of course, the regions between distinctive features tend to be planar surfaces, so perhaps it isn't too bad. Well this is exactly what I'm (unofficially) working on now. From the results I have at the moment I can say with confidence that it will be possible to navigate a robot around a home environment using a pair of stereo cameras, with the robot remaining within at least a 7cm position tollerance. 7cm is just a raw localisation figure, and after kalman filtering and sensor fusion with odometry the accuracy should be much better than that. You might think that there are not many features on walls, but even in environments which people consider to be blank there are often small imperfections or shading gradients which stereo algorithms can pick up. In real life few surfaces are perfectly uniform. With good localisation performance high quality mapping becomes possible. I can run the stereo algorithms at various levels of detail, and use traditional occupancy grid methods (with a few tweaks) to build up evidence in a probablistic fashion. The idea at the moment is to have the localisation algorithms running in real time using low-res grids, and to build a separate high quality model of the environment in a high resolution grid more gradually in a low priority background task. Once you have a good quality grid model its then quite straightforward to detect things like walls and furniture, and to simplify the data down to something which is a more efficient representation similar to something you might find in a game or an AGI sim. You can also use the grid model in exactly the same way that 2D background subtraction systems work (except in 3D) in order to detect changes within the environment. This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 24/10/06, Pei Wang [EMAIL PROTECTED] wrote: Any comments on Microsoft Robotics Studio?The microsoft robotics studio is quite an unimpressive release. I had expected to see user friendly IDEs and drag-and-drop function block programming, but there's none of that. About the best I can say is that it's at a very early stage and isn't something which is easy to get started with. The only redeeming feature of the microsoft robotics studio is the physics simulation. This might become of interest further down the line, but such simulations are useless unless the robot has a good perception system and is able to construct realistic models of its environment so that it may test out the consequences of possible actions before actually performing them (rather like Ben's conception of free will). Recently some people are talking about cognitive robotics, though I haven't found any major new idea beyond what robotics has beencovering, except the suggestion that high-level cognition should betaken into consideration. Am I missing something important? Well in most robotics applications there is a huge gap between The World and high level cognitive concepts formed about it. That gap is called perception, and at the moment it's only very sparsely filled. We have various low resolution sensors, such as ultrasonics and laser scanners but these only usually give quite an impoverished view of what the world out there is like. One of my main aims over the next few years is to try to fill this perception gap, using vision systems to build high fidelity models. Once you have a good model you then stand a fighting chance of doing all kinds of reasoning and planning within the cognitive realm. If I want to start to try some low-budget programmable robot (say, in the price range of Robosapien V2 and LEGO Mindstorms NXT), which onewill you recommend? I won't have high expectation in performance, butwill be interested in testing ideas on the coordination of perception,reasoning, learning, and action. The LEGO kit is really unbeatable for value. You can also remote control the robot using bluetooth. It doesn't have a camera, but it's not inconceivable that you could attach a wireless webcam and try some image processing on a desktop computer. The types of reasoning which you'll be able to do with a LEGO robot will be very limited, and a long way from anything which might be regarded as high level cognition, but it would be a good place to start. The main excitement will come when PC based robots are available at a reasonable ( $1000) cost, because PCs mean that some serious processing power can be brought to bear and detailed mapping of the environment becomes possible. Whitebox Robotics claim that their aim is to get the cost of their product into this kind of price range, but at the moment it remains as merely an expensive research curiosity. This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: Re: Re: [agi] SOTA
Hi, On 10/24/06, Pei Wang [EMAIL PROTECTED] wrote: Hi Ben, As you know, though I think AGISim is interesting, I'd rather directly try the real thing. ;-) I felt that way too once, and so (in 1996) I did directly try the real thing. Building a mobile robot and experimenting with it was fun, but I quickly learned that one spends all one's time dealing with sensor and actuator issues and never really gets to deal with cognition in any interesting way. Admittedly, robotic tech has advanced a lot since then, but I think the basic point still holds. IMO, given the current state of mobile robot tech, to do robotics-based AI effectively requires at least one dedicated team member fully devoted to the robotics side Much robotics-based AI involves first experimenting in robot simulation software, anyway. AGISim right now is not a robot simulation software package, but it could be tailored into one, which would be interesting Maybe we'll start by making an AGISim Roomba, using the Pyro interface to Roomba ;-) -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: Re: Re: Re: [agi] SOTA
I used to be of the opinion that doing robotics in simulation was a waste of time. The simulations were too perfect. To simplistic compared to the nitty gritty of real world environments. Algorithms developed and optimised for simulated environments would not translate well (or at all) into real robotics applications operating in non trivial environments. Ten years ago that was true, but now I think it's possible to build simulations with graphics and physics which are substantially more realistic, to the point where it might be possible to take algorithms developed within simulation, dump them onto a real robot and expect to see similar performance. You'd need to be careful to simulate sensor uncertainties, but it should be possible. Indeed ... I am interested in seeing AGISim go in this direction, though there is still more basic work to be done on AGISim first... Ironically, good quality simulations for robotics development will themselves assist in the cognitive process, becoming the robots inner theatre of the mind within which it may experiment with possible scenarios before committing to a course of action. Yeah, this is something we have discussed in an AGISim/Novamente context, though we have not done it yet. Basically, giving a Novamente system an internal AGISim theatre in which to experiment with various actions and scenarios, using simulated physics to shortcut the need for extensive cognition when appropriate... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/23/06, Bob Mottram [EMAIL PROTECTED] wrote: Another interesting development is the rise of the use of invariant feature detection algorithms together with geometric hashing for some kinds of object recognition. The most notable successes to date have been using David Lowe's SIFT method, which I think bears some resemblence to earlier methods such as Moravec's interest operator. To an extent these are just old algorithms developed in the 1980s enjoying a new lease of life within a more favourable computational environment. Heh, I know what you mean. In the computer vision/recognition literature, it almost seems like the non-deformable stuff can be split into pre-SIFT and post-SIFT. After Lowe's SIFT paper came out in 1999, it's kind of tricky to find a non-face visual recognition paper that doesn't use SIFT or some derivative. More recently, representations which take advantage of more contextual information (like Belongie's Shape Context or Berg's geometric blur) seem rather interesting, but I'm not sure how much they've proven themselves yet. For those of you who haven't seen SIFT in action, there's a neat downloadable demo available from Evolution Robotics: http://www.evolution.com/core/ViPR/ (registration required) It's been a few years since I've tried the demo myself, but if I recall correctly, you just plug a webcam into your computer, and you can use the program to learn to visually recognize different objects you hold in front of it. -- Neil - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/23/06, Bob Mottram [EMAIL PROTECTED] wrote: It's a shame that Evolution Robotics weren't able to develop that system further. A logical progression would be to extend the geometric hashing to 3D and eventually 4D, although that would require a stereo camera or some other way of measuring distances to the observed features. Even so that demo program of theirs remains as one of the more impressive examples of invariant object recognition. You can present objects at all kinds of different rotations and scales and still have the program locate them, even within a noisy webcam image. I know one guy who bought an Evolution robot some years ago. He took it to a primary school and demonstrated how it recognised different objects. He said that the robot could recognise the objects and speak their names faster than the children could. Is a stereo camera system really necessary if you can move the camera around to get shape-from-motion? Last I heard Evolution was partnering up with WowWee, to load their software onto the next generation of their RoboSapien toy: http://www.pcmag.com/article2/0,1895,1941233,00.asp I find a SIFT-equipped robot in the $200 range to be quite exciting. Hopefully the new generation will be as hack-friendly as the older generations. -- Neil - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
You can get depth information from single camera motion (eg Andrew Davison's MonoSLAM), but this requires an initial size calibration and continuous tracking. If the tracking is lost at any time you need to recalibrate. This makes single camera systems less practical. With a stereo camera the baseline gives the scale calibration. My inside sources tell me that there's little or no software development going on at Evolution Robotics, and that longstanding issues and bugs remain unfixed. They did licence their stuff to WoWee, and also Whitebox Robotics, so its likely we'll see more SIFT enabled robots in the not too distant future. I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Grinding my own axe, I also think that stereo vision systems will bring significant improvements to robotics over the next few years. Being able to build videogame-like 3D models of the environment in real time is now a feasible proposition which I think will happen before the decade is out. With a good model of the environment the robot can rehearse possible scenarios before actually running them, and find important features such as desk or table surfaces. On 23/10/06, Neil H. [EMAIL PROTECTED] wrote: Is a stereo camera system really necessary if you can move the cameraaround to get shape-from-motion?Last I heard Evolution was partnering up with WowWee, to load theirsoftware onto the next generation of their RoboSapien toy: http://www.pcmag.com/article2/0,1895,1941233,00.aspI find a SIFT-equipped robot in the $200 range to be quite exciting.Hopefully the new generation will be as hack-friendly as the older generations.-- Neil-This list is sponsored by AGIRI: http://www.agiri.org/emailTo unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/23/06, Bob Mottram [EMAIL PROTECTED] wrote: My inside sources tell me that there's little or no software development going on at Evolution Robotics, and that longstanding issues and bugs remain unfixed. They did licence their stuff to WoWee, and also Whitebox Robotics, so its likely we'll see more SIFT enabled robots in the not too distant future. I hope so. I visited Evolution last year, and I remember that (what seemed to be) their chief product was a system for automatically identifying groceries hidden at the bottom of shopping carts. I guess that's useful from a business perspective, but it isn't particularly sexy. I'm also pretty surprise that they haven't done anything major with their vSLAM tech: http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1570091 I think their stuff was also licenced to Sony for use on their AIBO, before Sony axed their robotics products. Sony licensed the tech, but I think they only used it so that AIBO could visually recognize pre-printed patterns on cards, which would signal the AIBO to dance, return to the charging station, etc. SIFT is IMHO overkill for that kind of thing, and it's a pity they didn't do anything more interesting with it. Grinding my own axe, I also think that stereo vision systems will bring significant improvements to robotics over the next few years. Being able to build videogame-like 3D models of the environment in real time is now a feasible proposition which I think will happen before the decade is out. With a good model of the environment the robot can rehearse possible scenarios before actually running them, and find important features such as desk or table surfaces. Perhaps. To play devil's advocate, how well do you think stereo vision system would actually work for creating a 3D structure of a home environment? It seems that distinctive features in the home tend to be few and far between. Of course, the regions between distinctive features tend to be planar surfaces, so perhaps it isn't too bad. -- Neil - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On Fri, 20 Oct 2006 22:15:37 -0400, Richard Loosemore wrote Matt Mahoney wrote: From: Pei Wang [EMAIL PROTECTED] On 10/20/06, Matt Mahoney [EMAIL PROTECTED] wrote: It is not that we can't come up with the right algorithms. It's that we don't have the computing power to implement them. Can you give us an example? I hope you don't mean algorithms like exhaustive search. For example, neural networks which perform rudamentary pattern detection and control for vision, speech, language, robotics etc. Most of the theory had been worked out by the 1980's, but applications have been limited by CPU speed, memory, and training data. The basic building blocks were worked out much earlier. There are only two types of learning in animals, classical (association) and operant (reinforcement) conditioning. Hebb's rule for classicical condioning proposed in 1949 is the basis for most neural network learning algorithms today. Models of operant conditioning date back to W. Ross Ashby's 1960 Design for a Brain where he used randomized weight adjustments to stabilize a 4 neuron system build from vacuum tubes and mechanical components. Neural algorithms are not intractable. They run in polynomial time. Neural networks can recognize arbitrarily complex patterns by adding more layers and training them one at a time. This parallels the way people learn complex behavior. We learn simple patterns first, then build on them. I initially wrote a few sentences saying what was wrong with the above, but I chopped it. There is just no point. What you said above is just flat-out wrong from beginning to end. I have done research in that field, and taught postgraduate courses in it, and what you are saying is completely divorced from reality. Richard Loosemore I have simply taken maybe one and say a half (because it seems like every ai survey class has to touch upon neural nets again) graduate classes on the subject, and not taught or done research in the area, but I recognized that most of that was wrong. I at least hold out the possibility that neural nets can be made useful with some greater theory about architectures and much greater computing power. I think it would be worthwhile for you to take the time to list what you think the flaws were, if only to open the possibility for some positive recomendations for research directions. Even thought you may be completely disillusioned, maybe not everyone is. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Andrew, I happen to have a list you asked. Last year I taught a graduate course on NN (http://www.cis.temple.edu/~pwang/525-NC/CIS525.htm), and afterwards wrote a paper (http://nars.wang.googlepages.com/wang.AGI-CNN.pdf) to list its strength and weakness, with respect to AGI. In the paper, I only analyzed NN in its classical form, because it simply has too many variants to be covered in a single paper. However, even taking all the variants I know into consideration, I still cannot agree with Matt that NN has got the right algorithms, and only needs the required hardware to work. Of course, I'm talking about AGI, rather than specific pattern recognition tasks, which NN often does well indeed, and additional computational power will surely further improve its performance there. Pei On 10/21/06, Andrew Babian [EMAIL PROTECTED] wrote: On Fri, 20 Oct 2006 22:15:37 -0400, Richard Loosemore wrote Matt Mahoney wrote: From: Pei Wang [EMAIL PROTECTED] On 10/20/06, Matt Mahoney [EMAIL PROTECTED] wrote: It is not that we can't come up with the right algorithms. It's that we don't have the computing power to implement them. Can you give us an example? I hope you don't mean algorithms like exhaustive search. For example, neural networks which perform rudamentary pattern detection and control for vision, speech, language, robotics etc. Most of the theory had been worked out by the 1980's, but applications have been limited by CPU speed, memory, and training data. The basic building blocks were worked out much earlier. There are only two types of learning in animals, classical (association) and operant (reinforcement) conditioning. Hebb's rule for classicical condioning proposed in 1949 is the basis for most neural network learning algorithms today. Models of operant conditioning date back to W. Ross Ashby's 1960 Design for a Brain where he used randomized weight adjustments to stabilize a 4 neuron system build from vacuum tubes and mechanical components. Neural algorithms are not intractable. They run in polynomial time. Neural networks can recognize arbitrarily complex patterns by adding more layers and training them one at a time. This parallels the way people learn complex behavior. We learn simple patterns first, then build on them. I initially wrote a few sentences saying what was wrong with the above, but I chopped it. There is just no point. What you said above is just flat-out wrong from beginning to end. I have done research in that field, and taught postgraduate courses in it, and what you are saying is completely divorced from reality. Richard Loosemore I have simply taken maybe one and say a half (because it seems like every ai survey class has to touch upon neural nets again) graduate classes on the subject, and not taught or done research in the area, but I recognized that most of that was wrong. I at least hold out the possibility that neural nets can be made useful with some greater theory about architectures and much greater computing power. I think it would be worthwhile for you to take the time to list what you think the flaws were, if only to open the possibility for some positive recomendations for research directions. Even thought you may be completely disillusioned, maybe not everyone is. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
With regard to the computational requirements of AI, there is a very clear relation showing that the quality of a language model improves by adding time and memory, as shown in the following table: http://cs.fit.edu/~mmahoney/compression/text.html And with the size of the training set, as shown in this graph: http://cs.fit.edu/~mmahoney/dissertation/ Before you argue that text compression has nothing to do with AI, please read http://cs.fit.edu/~mmahoney/compression/rationale.html I recognize that language modeling is just one small aspect of AGI. But compression gives us hard numbers to compare the work of over 80 researchers spanning decades. The best performing systems push the hardware to its limits. This, and the evolutionary arguments I gave earlier lead me to believe that AGI will require a lot of computing power. Exactly how much, nobody knows. Whether or not AGI can be accomplished most efficiently with neural networks is an open question. But the one working system we know of is based on it, and we ought to study it. One critical piece of missing knowledge is the density of synapses in the human brain. I think this could be resolved by putting some brain tissue under an electron microscope, but I guess that the number is not important to neurobiologists. I read Pei Wang's paper, http://nars.wang.googlepages.com/wang.AGI-CNN.pdf Some of the shortcomings of neural networks mentioned only apply to classical (feedforward or symmetric) neural networks, not to asymmetric networks with recurrent circuits and time delay elements, as exist in the brain. Such circuits allow for short term stable or oscillating states which overcome some shortcomings such as the inability to train on multiple goals, which could be accomplished by turning parts of the network on or off. Also, it is not true that training has to be offline using multiple passes, as with backpropagation. Human language is structured so that layers can be trained progressively without need to search over hidden units. Word associations like sun-moon or to-from are linear. Some of the top compressors mentioned above (paq8, WinRK) use online, single pass neural networks to combine models, alternating prediction and training. But it is interesting that most of the remaining shortcomings are also shortcomings of human thought, such as the inability to insert or represent structured knowledge accurately. This is evidence that our models are correct. This does not mean they are the best answer. We don't want to duplicate the shortcomings of humans. We do not want to slow down our responses and insert errors in order to pass the Turing test (as in Turing's 1950 example). -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/21/06, Matt Mahoney [EMAIL PROTECTED] wrote: I read Pei Wang's paper, http://nars.wang.googlepages.com/wang.AGI-CNN.pdf Some of the shortcomings of neural networks mentioned only apply to classical (feedforward or symmetric) neural networks, not to asymmetric networks with recurrent circuits and time delay elements, as exist in the brain. I made it clear that the paper is only directly about classical NN, as defined at the beginning of the paper. I guess for each of the shortcomings I mentioned in the paper, someone may design a NN just to overcome it. However, I haven't known any NN that can overcome all of the shortcomings, or even most of them, in a consistent manner. I never mentioned the brain in the paper, because it is a completely different issue. When we say The brain is a neural network, we are using the phrase neural network in a very different sense. Therefore, because the brain can do something, it doesn't mean that what we call neural network in AI research can do the same, even after major extensions and revisions of the existing models. But it is interesting that most of the remaining shortcomings are also shortcomings of human thought, such as the inability to insert or represent structured knowledge accurately. This is evidence that our models are correct. Every shortcoming of NN I mentioned in the paper is selected exactly because the human mind does not have the problem at all, or it is much less a problem there. Furthermore, I mentioned it as a shortcoming of NN because I believe it can be overcome by other AI techniques, though I didn't discuss them in the paper. It makes no sense to criticize NN on failing a task that nobody can accomplish. For example, the human mind and some other AI techniques handle structured knowledge much better than NN does. Pei - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
- Original Message From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Saturday, October 21, 2006 5:25:13 PM Subject: Re: [agi] SOTA For example, the human mind and some other AI techniques handle structured knowledge much better than NN does. Is this because the brain is representing the knowledge differently than a classical neural network, or because the brain has a lot more memory and can afford to represent structured knowledge inefficiently? I agree with the conclusion of your paper that a classical neural network is not sufficient to solve AGI. The brain is much more complex than that. But I think a neural architecture or a hybrid system that includes neural networks of some type is the right direction. For example, Novamente (if I understand correctly, a weighted hypergraph) has some resemblance to a neural network -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/21/06, Matt Mahoney [EMAIL PROTECTED] wrote: - Original Message From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Saturday, October 21, 2006 5:25:13 PM Subject: Re: [agi] SOTA For example, the human mind and some other AI techniques handle structured knowledge much better than NN does. Is this because the brain is representing the knowledge differently than a classical neural network, or because the brain has a lot more memory and can afford to represent structured knowledge inefficiently? I believe it is mainly the former. I agree with the conclusion of your paper that a classical neural network is not sufficient to solve AGI. The brain is much more complex than that. But I think a neural architecture or a hybrid system that includes neural networks of some type is the right direction. For example, Novamente (if I understand correctly, a weighted hypergraph) has some resemblance to a neural network Well, in that sense NARS also has some resemblance to a neural network, as well as many other AI systems. To me, the problem is that the current NN technique is not rich and powerful enough to support AGI design, though many ideas behind NN are really necessary for AGI, as I argued in my paper --- I hope the paper is not taken by people as a pure criticism to NN, since I also listed its advantages over traditional symbolic AI. Even if an AGI has some similarity with a NN, it is not necessarily a hybrid system with a NN part (though I cannot exclude that possibility), but may take some NN ideas without most of the details, as in the case of NARS. Pei -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Olie, I agree with most of your statements, however, history shows that almost every subfield of AI has enjoyed a rapid progress period accompanied by optimism, then followed by a long period of slow moving accompanied by pessimism. Just think about expert system in the early 1980s and neural network in the late 1980s, or, more recently, the rise of multi-agent system and data mining movements. Even NLP looked promising to some people when statistical methods began to show their power. Because of this, I'm not sure how long robotics can keep its recent improving rate without major progress in AI in general. I wonder if there is anyone in this list who has been actually working in the field of robotics, and I would be very interested in learning the causes of the recent development. Pei On 10/19/06, Olie Lamb [EMAIL PROTECTED] wrote: (Excellent list there, Matt) Although Pei Wang makes a good point that the fragmentation of AI does make it difficult to compare projects, it is interesting+ to note the huge differences in the movements in different narrow-AI fields. As has already been mentioned, it is interesting+ to compare the way that progress is very slow in areas such as NLP and Expert Systems, whereas there is significant, albeit gradual progress in physical interaction systems. For instance, the soccer-bots get better every year, cars can now finish DARPA grand challenge -like events in reasonable time... (I personally think that we're fast approaching a critical point where the technology is just good enough to attract more cash and hence more improvement; although meatbags will be better traffic-drivers for a while yet, physical interaction systems can now perform well enough for many applications) Although the question What is State-of-the-Art? won't attract an incontrivertibly good answer, it prompts a lot of bloody good questions that can be answered usefully. -- Olie This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
BillK wrote: On 10/19/06, Richard Loosemore [EMAIL PROTECTED] wrote: Sorry, but IMO large databases, fast hardware, and cheap memory ain't got nothing to do with it. Anyone who doubts this get a copy of Pim Levelt's Speaking, read and digest the whole thing, and then meditate on the fact that that book is a mere scratch on the surface (IMO a scratch in the wrong direction, too, but that's neither here nor there). I saw a recent talk about an NLP system which left me stupified that so little progress has been made since 20 years ago. Having a clue about just what a complex thing intelligence is, has everything to do with it. Most normal speaking requires relatively little 'intelligence'. Adults who take young children on foreign holidays are amazed at how quickly the children appear to be chattering away to other children in a foreign language. They manage it for several reasons: 1) they don't have the other interests and priorities that adults have. 2) they use simple sentence structures and smallish vocabularies. 3) they discuss simple subjects of interest to children. Bill, with all due respect, this flies in the face of a massive body of work that reveals just how incredibly complex the understanding and production of language actually is. As I said before, try taking a look at the opening pages of Levelt's book. He takes a fragment of conversation consisting of one question and one answer, then he looks at exactly what he thinks was going on when the respondent said (IIRC): I I don't know the way ... play WELL enough, sir. The amount of planning, at numerous levels, that had to be engaged just to come out with that utterance is truly staggering. You don't have to duplicate the faltering prose, but on the other hand this flatering prose reveals that enormously sophisticated processes were at work when the utterance was produced, and the machine should have at least the same level of sophistication. Humans are built in such a way that they can understand (at least implicitly) many levels of message in even as simple a phrase as the one above. To be able to understand humans, machines would need the same degree of subtle understanding. For you to blithely say Most normal speaking requires relatively little 'intelligence' is just mind-boggling. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Matt Mahoney wrote: Sorry, but IMO large databases, fast hardware, and cheap memory ain't got nothing to do with it. Yes it does. The human brain has vastly more computing power, memory and knowledge than all the computers we have been doing AI experiments on. We have been searching for decades for shortcuts to fit our machines and found none. If any existed, then why hasn't human intelligence evolved in insect sized brains? If the brain used its hardware in such a way that (say) a million neurons were required to implement a function that, on a computer, required a few hundred gates, your comparisons would be meaningless. We do not know one way or the other what the equivalence ratio is, but your statement implicitly assumes a particular (and unfavorable) ratio: you simply cannot make that statement without good reasons for assuming a ratio. There are no such reasons, so the statement is meaningless. We have been searching for decades to find shortcuts to fit our machines? When you send a child into her bedroom to search for a missing library book, she will often come back 15 seconds after entering the room with the statement I searched for it and it's not there. Drawing definite conclusions is about equally reliable, in both these cases. It used to be a standing joke in AI that researchers would claim there was nothing wrong with their basic approach, they just needed more computing power to make it work. That was two decades ago: has this lesson been forgotten already? As long as knowledge accumulates exponentially (as it has been doing for centuries) and Moore's law holds (which it has since the 1950s), you can be sure that machines will catch up with human brains. When that happens, a lot of AI problems that have been stagnant for a long time will be solved all at once. A hundred dog's dinners maketh not a feast. Having a clue about just what a complex thing intelligence is, has everything to do with it. Which is why we will not be able to control AI after we produce it. It is not possible, even in theory, for a machine (your brain) to simulate or predict another machine with more states or greater Kolmogorov complexity [1]. The best you can do is duplicate the architecture and learning mechanisms of the brain and feed it data you can't examine because there is too much of it. You will have built AI but you won't know how it works. [1] Legg, Shane, (2006), Is There an Elegant Universal Theory of Prediction?, Technical Report IDSIA-12-06, IDSIA / USI-SUPSI, Dalle Molle Institute for Artificial Intelligence, Galleria 2, 6928 Manno, Switzerland. A completely spurious argument. You would not necessarily *need* to simulate or predict the AI, because the kind of simulation and prediction you are talking about is low-level, exact state prediction (this is inherent in the nature of proofs about Kolmogorov complexity). It is entirely possible to build an AI in such a way that the general course of its behavior is as reliable as the behavior of an Ideal Gas: can't predict the position and momentum of all its particles, but you sure can predict such overall characteristics as temperature, pressure and volume. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
It is entirely possible to build an AI in such a way that the general course of its behavior is as reliable as the behavior of an Ideal Gas: can't predict the position and momentum of all its particles, but you sure can predict such overall characteristics as temperature, pressure and volume. Thank you! FINALLY, someone came up with a clear succint and accurate rebuttal . . . . - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
BillK wrote: On 10/20/06, Richard Loosemore [EMAIL PROTECTED] wrote: For you to blithely say Most normal speaking requires relatively little 'intelligence' is just mind-boggling. I am not trying to say that language skills don't require a human level of intelligence. That's obvious. That is what make humans human. But day-to-day chat can be mastered by children, even in a foreign language. Watch that video I referenced in my previous post, of an American chatting to a Chinese woman via a laptop running MASTOR software. http://www.research.ibm.com/jam/speech_to_speech.mpg Now tell me that that laptop is showing great intelligence to translate at the basic level of normal conversation. Simple subject object predicate stuff. Basic everyday vocabulary. No complex similes, metaphors, etc. There is a big difference between discussing philosophy and saying Where is the toilet? That is what I was trying to point out. I take your point, but I the problem I have with what you are saying is that you are making a claim about the feasibility of a restricted type of performance (not general language understanding, but just translation of very simple sentences), while at the same time trying to push the envelope on how useful and widely applicable that restricted performance actually is. So when you say Most normal speaking requires relatively little 'intelligence', I simply cannot buy the implications of that most. Most language contains subtleties that do require a good deal of intelligence. People consistently underestimate the sophistication of child language, in particular. I would *love* to see those IBM folks put a couple of jabbering four-year-old children in front of that translation system, to see how it likes their 'low-intelligence' language. :-) Does anyone have any contacts on the team, so we could ask? Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
- Original Message From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Friday, October 20, 2006 10:14:09 AM Subject: Re: [agi] SOTA We have been searching for decades to find shortcuts to fit our machines? When you send a child into her bedroom to search for a missing library book, she will often come back 15 seconds after entering the room with the statement I searched for it and it's not there. Drawing definite conclusions is about equally reliable, in both these cases. If you have figured out how to implement AI on a PC, please share it with us. Until then, you will need a more convincing argument that we aren't limited by hardware. A lot of people smarter than you or me have been working on this problem for a lot longer than 15 seconds. James first proposed association models of thought in 1890, about 90 years before connectionist neural models were popular. Hebb proposed a model of classical conditioning in which memory is stored in the synapse in 1949, decades before the phenomena was actually observed in living organisms. By the early 1960s we had programs that could answer natural language queries (the 1959 BASEBALL program), translate Russian to English, prove theorems in geometry, solve arithmetic word problems, and recognize handwritten digits. It is not that we can't come up with the right algorithms. It's that we don't have the computing power to implement them. The most successful AI applications today like Google require vast computing power. If the brain used its hardware in such a way that (say) a million neurons were required to implement a function that, on a computer, required a few hundred gates, your comparisons would be meaningless. I doubt the brain is that inefficient. There are lower animals that crawl with just a couple hundred neurons. In higher animals, neural processing is expensive, so there is evolutionary pressure to compute efficiently. Most of the energy you burn at rest is used by your brain. Humans had to evolve larger bodies than other primates to support our larger brains. In most neural models, it takes only one neuron to implement a logic gate and only one synapse to store a bit of memory. It used to be a standing joke in AI that researchers would claim there was nothing wrong with their basic approach, they just needed more computing power to make it work. That was two decades ago: has this lesson been forgotten already? I don't see why this should not still be true. The problem is we still do not know just how much computing power is needed. There is still no good estimate of the number of synapses in the human brain. We only know it is probably between 10^12 to 10^15 and we aren't even sure of that. So when AI is solved, it will probably be a surprise. -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Bill, Richard, etc, Children don't have a great grasp of language, but they have all the sensory and contextual mechanisms to learn a language by causal interaction with their environment. Semantics are a learned system, just as words are. In current AI we're programming semantic rules into a huge neural database, and asking it to play an big matching game. These two types of learning may give the same result, but it's not the same process by a long shot. Every time we logically code an algorithm, we're only mimicking the logic function of a learned neural process, which doesn't allow the tiered complexity and concept grasping that sensory learning does. Because language uses discrete semantic rules, it's easy to fall into the trap of thinking computers, given enough horsepower, are capable of human thought. Give a computer as many semantic algorithms, metaphor databases, and reaction grading mechanisms as you want, but it takes much deeper and differentiated networks to apply those words and derive a physical meaning beyond grammatical or metaphorical boundaries. This is the difference between a system that resembles intelligence, and an intelligent system. The resembling system is only capable of processing information based on algorithms, and not reworking an algorithm based on the reasoning for executing the function. Whether our AGI is conscious or not, it could still be functionally equivalent to a human mind in terms of output. The recursive bidirectional nature of neurons and their relation to forming a gestalt is something we're barely able to grasp as a concept, let alone code for. The nature of our hardware is going to have to change to accommodate these multidimensional and recursive problems in computing. Josh Treadwell Systems Administrator [EMAIL PROTECTED] direct:480.206.3776 C.R.I.S. Camera Services 250 North 54th Street Chandler, AZ 85226 USA p 480.940.1103 / f 480.940.1329 http://www.criscam.com BillK wrote: On 10/20/06, Richard Loosemore [EMAIL PROTECTED] wrote: For you to blithely say "Most normal speaking requires relatively little 'intelligence'" is just mind-boggling. I am not trying to say that language skills don't require a human level of intelligence. That's obvious. That is what make humans human. But day-to-day chat can be mastered by children, even in a foreign language. Watch that video I referenced in my previous post, of an American chatting to a Chinese woman via a laptop running MASTOR software. http://www.research.ibm.com/jam/speech_to_speech.mpg Now tell me that that laptop is showing great intelligence to translate at the basic level of normal conversation. Simple subject object predicate stuff. Basic everyday vocabulary. No complex similes, metaphors, etc. There is a big difference between discussing philosophy and saying "Where is the toilet?" That is what I was trying to point out. Billk - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
Re: [agi] SOTA
I want to strongly agree with Richard on several points here, and expand on them a bit in light of later discussion. On 10/20/06, Richard Loosemore [EMAIL PROTECTED] wrote: It used to be a standing joke in AI that researchers would claim there was nothing wrong with their basic approach, they just needed more computing power to make it work. That was two decades ago: has this lesson been forgotten already? This is very true then, and continues to be now. For those who use the explanation of insufficient computing power, I would question what approaches you would expect to be viable at higher computing power? How do they scale? Why would they work better with more computation? Relatedly, very very few AI research programmes operate in strict real time. Many use batch processes, or virtual worlds, or automated interaction scripts. It would be trivial to modify these systems to behave as if they had 10 times as much computational power, or a thousand times. Even if it took 1,000,000 seconds(11 1/2 days) for every second of intelligent behavior with currently available computing power, the results would be worth it, and unmistakeable, if true. I suspect that this would not work, as simply increasing computing power would not validate current AI systems. A completely spurious argument. You would not necessarily *need* to simulate or predict the AI, because the kind of simulation and prediction you are talking about is low-level, exact state prediction (this is inherent in the nature of proofs about Kolmogorov complexity). This very important, and I strongly agree that analysis of this kind is unhelpful. It's easy to show that heat engines and turbines and all sorts of things are so insanely complex that they can't possibly be modeled in the general case. But we needn't do so. We are interested in the behavior of certain parameters of such systems, and we can reduce the space of the systems we investigate(very few people build turbines with disconnected parts, or assymetrical rotation, for example). It is entirely possible to build an AI in such a way that the general course of its behavior is as reliable as the behavior of an Ideal Gas: can't predict the position and momentum of all its particles, but you sure can predict such overall characteristics as temperature, pressure and volume. This is the only claim in this message I have any disagreement with (which must be some sort of record given my poor history with Richard). I agree that its true in principle that AIs can be made this way, but I'm not yet convinced that it's possible in practice. It may be that the goals of and motivations from such artificial systems are not one of those characteristics that lies on the surface of such boiling complexity, but within it. I have the same disagreement with Eliezer, about the certainty he places on the future characteristics of AIs, given that no one here is describing the behavior of a specific AI system, such conclusions strike me as premature, but perhaps not unwarrented. -- Justin Corwin [EMAIL PROTECTED] http://outlawpoet.blogspot.com http://www.adaptiveai.com - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/19/06, Olie Lamb [EMAIL PROTECTED] wrote: For instance, the soccer-bots get better every year, cars can now finish DARPA grand challenge -like events in reasonable time... (I personally think that we're fast approaching a critical point where the technology is just good enough to attract more cash and hence more improvement; although meatbags will be better traffic-drivers for a while yet, physical interaction systems can now perform well enough for many applications) I used to work some with NASA on Free Flight issues. Free Flight is a grand plan to let pilots choose their own routes. Part of it involves computer-controlled conflict resolution, meaning that instead of having a guy with a radar screen giving instructions to pilots to keep them from crashing into each other, their onboard computers do it. Computers have been able to handle every aspect of flying, from takeoff, piloting, collision-avoidance, routing, and landing, better than people, for a long time now. But nobody will let computers do it. Part of the issue is reliability - you need to have less than a one in a million chance of error to get away with zero errors per year, in the US. Although air traffic controllers in the US today commit tens of thousands of errors per year; most of them have no consequences because airspace is big. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/20/06, Josh Treadwell [EMAIL PROTECTED] wrote: The resembling system is only capable of processing information based on algorithms, and not reworking an algorithm based on the reasoning for executing the function. This appears to be the same argument Spock made in an old Star Trek episode, that the computer chess-player could never beat the person who programmed it. Note to the world: It is wrong. Please stop using this argument. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/19/06, Peter Voss [EMAIL PROTECTED] wrote: I'm often asked about state-of-the-art in AI, and would like to get some opinions. What do you regard, or what is generally regarded as SOTA in the various AI aspects that may be, or may be seen to be relevant to AGI? - NLP components such as parsers, translators, grammar-checkers AFAIK, our group here at NIH has the state-of-the-art in extracting knowledge from text. We have a system that reads medical abstracts - not fake or hand-chosen texts, but horribly messy run-on sentences with generally correct but reprehensible grammar - and extracts predicates, like so: Qinghaosu and its derivatives are rapidly effective antimalarial drugs derived from a Chinese plant. - Qinghaosu treats malaria Kava as an anticraving agent: preliminary data. - Kava treats craving Leukotriene receptor antagonists and synthesis inhibitors are a new class of anti-inflammatory drugs that have clinical efficacy in the management of asthma, allergic rhinitis and inflammatory bowel disease. - receptor antagonists treat asthma receptor antogonists treat allergic rhinitis receptor antogonists treat inflammatory bowel disease synthesis inhibitors treat asthma ... Unfortunately, the secret to getting such good performance is to have lots and lots of domain knowledge, and lots of linguists studying lots of different grammatical constructions, and lots of coders writing code to parse each construction, and do this for many years. This project is in its 15th year. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
- Original Message From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Friday, October 20, 2006 3:35:57 PM Subject: Re: [agi] SOTA On 10/20/06, Matt Mahoney [EMAIL PROTECTED] wrote: It is not that we can't come up with the right algorithms. It's that we don't have the computing power to implement them. Can you give us an example? I hope you don't mean algorithms like exhaustive search. For example, neural networks which perform rudamentary pattern detection and control for vision, speech, language, robotics etc. Most of the theory had been worked out by the 1980's, but applications have been limited by CPU speed, memory, and training data. The basic building blocks were worked out much earlier. There are only two types of learning in animals, classical (association) and operant (reinforcement) conditioning. Hebb's rule for classicical condioning proposed in 1949 is the basis for most neural network learning algorithms today. Models of operant conditioning date back to W. Ross Ashby's 1960 Design for a Brain where he used randomized weight adjustments to stabilize a 4 neuron system build from vacuum tubes and mechanical components. Neural algorithms are not intractable. They run in polynomial time. Neural networks can recognize arbitrarily complex patterns by adding more layers and training them one at a time. This parallels the way people learn complex behavior. We learn simple patterns first, then build on them. The most successful AI applications today like Google require vast computing power. In what sense do you call Google an AI application? Google does pretty well with natural language questions like how many days until xmas? even though they don't advertise it that way (like Ask Jeeves did) and most people don't use it that way. Of course you might say that Google isn't doing AI, it is just matching query terms to documents. But it is always that way. Once you solve the problem, it's not AI any more. Deep Blue isn't AI. It just implements a chess playing algorithm in fast hardware. Suppose we decide the easiest way to build a huge neural network is to use real neurons and some genetic engineering. Is that AI? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Philip Goetz wrote: On 10/20/06, Josh Treadwell [EMAIL PROTECTED] wrote: The resembling system is only capable of processing information based on algorithms, and not reworking an algorithm based on the reasoning for executing the function. This appears to be the same argument Spock made in an old Star Trek episode, that the computer chess-player could never beat the person who programmed it. Note to the world: It is wrong. Please stop using this argument. It's not the same. A chess program is merely comparing outcomes and percentages, while adapting algorithmically to play styles. It's a discrete system within which logically written functions are executed. Yes, it adapts to moves and keeps a track of which moves are going on, but there is no higher order AI that is thinking "out of the box" about the problem. It simply approaches, computes based on a database of moves, and weighs it's advantages and disadvantages. A chess program never reworks it's strategy based on it's own reasoning of why it's playing. It just does, and does well. Yes it could beat us, but it's akin to saying a calculator is faster at math than we are. Josh Treadwell Systems Administrator [EMAIL PROTECTED] direct:480.206.3776 C.R.I.S. Camera Services 250 North 54th Street Chandler, AZ 85226 USA p 480.940.1103 / f 480.940.1329 http://www.criscam.com This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
Re: [agi] SOTA
Matt Mahoney wrote: From: Pei Wang [EMAIL PROTECTED] On 10/20/06, Matt Mahoney [EMAIL PROTECTED] wrote: It is not that we can't come up with the right algorithms. It's that we don't have the computing power to implement them. Can you give us an example? I hope you don't mean algorithms like exhaustive search. For example, neural networks which perform rudamentary pattern detection and control for vision, speech, language, robotics etc. Most of the theory had been worked out by the 1980's, but applications have been limited by CPU speed, memory, and training data. The basic building blocks were worked out much earlier. There are only two types of learning in animals, classical (association) and operant (reinforcement) conditioning. Hebb's rule for classicical condioning proposed in 1949 is the basis for most neural network learning algorithms today. Models of operant conditioning date back to W. Ross Ashby's 1960 Design for a Brain where he used randomized weight adjustments to stabilize a 4 neuron system build from vacuum tubes and mechanical components. Neural algorithms are not intractable. They run in polynomial time. Neural networks can recognize arbitrarily complex patterns by adding more layers and training them one at a time. This parallels the way people learn complex behavior. We learn simple patterns first, then build on them. I initially wrote a few sentences saying what was wrong with the above, but I chopped it. There is just no point. What you said above is just flat-out wrong from beginning to end. I have done research in that field, and taught postgraduate courses in it, and what you are saying is completely divorced from reality. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
justin corwin wrote: It is entirely possible to build an AI in such a way that the general course of its behavior is as reliable as the behavior of an Ideal Gas: can't predict the position and momentum of all its particles, but you sure can predict such overall characteristics as temperature, pressure and volume. This is the only claim in this message I have any disagreement with (which must be some sort of record given my poor history with Richard). I agree that its true in principle that AIs can be made this way, but I'm not yet convinced that it's possible in practice. [Heck, yeah: I had to check to see if maybe you were some other justin corwin ;-)] I agree with your caution about that claim. For what its worth, though, what I had in mind was a motivational system that acts like a point moving around in an unconstrained way in a multidimensional space (hence my thermodynamics metaphor), BUT with the shape of that space heavily distorted so that it has one enormously deep basin of attraction. To be stable and safe, it must stay in the basin. Can it get out? Yes, in principle. Mean time between escapes from the basin? Greater than lifetime of the universe by a huge factor. Is it therefore guaranteed to be safe? No. Is it ever likely to get out? Not a chance. (I haven't said how to do this. What I just gave was a summary of overall dynamics of a particular system). In the same way, all the molecules of an ideal gas could, in theory, just happen to divide into two equal chunks and head toward opposite ends of the box at full speed. They have the freedom to do so. We just don't have to worry about it happening, for obvious reasons. The beauty of the approach is that the system can know perfectly well that it has been engineered that way by us. But it does not care: if it starts out being empathic to humanity, it will not deliberately or accidentally change itself to contradict that initial goal. It will have plenty of freedom to do so in principle, but in practice it will know that the consequences could be disastrous. I realise that I have been tempted to explain an idea in partial, cryptic terms (laying myself open to requests for more detail, or scorn), so apologies if the above seems opaque. More when I get the time. Richard Loosemore. It may be that the goals of and motivations from such artificial systems are not one of those characteristics that lies on the surface of such boiling complexity, but within it. I have the same disagreement with Eliezer, about the certainty he places on the future characteristics of AIs, given that no one here is describing the behavior of a specific AI system, such conclusions strike me as premature, but perhaps not unwarrented. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Peter, I'm afraid that your question cannot be answered as it is. AI is highly fragmented, which not only means that few project is aiming at the whole field, but also that few is even covering a subfield as you listed. Instead, each project usually aims at a special problem under a set of special assumptions. Consequently, it is not always meaningful to compare them in functionality. For example, many people may agree that Stanley the Volkswagen represents the SOTA in robot car, but is it SOTA in Interactive robotics systems? Is it ahead of Cog? When common-sense KB is mentioned, people will think about Cyc, but is it SOTA? If it is not, which one is? How can we compare an inference engine based on first-order predicate calculus to one on Bayesian net? Of course, in each field, there are projects that are more typical, more influential, or more interesting than the rest, but they are not really SOTA in the sense that it is ahead of the others in functionality, since the others are usually running to different directions. In your list, NLP may be an exception to what I said above. Since I'm not an expert in that field, I won't try to answer. By definition, Integrated intelligent systems should be comparable, but clearly there is no consensus on this topic yet. ;-) Pei On 10/19/06, Peter Voss [EMAIL PROTECTED] wrote: I'm often asked about state-of-the-art in AI, and would like to get some opinions. What do you regard, or what is generally regarded as SOTA in the various AI aspects that may be, or may be seen to be relevant to AGI? For example: - Comprehensive (common-sense) knowledge-bases and/or ontologies - Inference engines, etc. - Adaptive expert systems - Question answering systems - NLP components such as parsers, translators, grammar-checkers - Interactive robotics systems (sensing/ actuation) - physical or virtual - Vision, voice, pattern recognition, etc. - Interactive learning systems - Integrated intelligent systems ... whatever ... I'm looking for the best functionality -- irrespective of proprietary, open-source, or academic. Peter - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
- Comprehensive (common-sense) knowledge-bases and/or ontologies Cyc/OpenCyc, Wordnet, etc. but there seems to be no good way for applications to use this information and no good alternative to hand coding knowledge. - Inference engines, etc. - Adaptive expert systems A dead end. There has been little progress since the 1970's. - Question answering systems Google. - NLP components such as parsers, translators, grammar-checkers Parsing is unsolved. Translators like Babelfish have progressed little since the 1959 Russian-English project. Microsoft Word's grammar checker catches some mistakes but is clearly not AI. - Interactive robotics systems (sensing/ actuation) - physical or virtual The Mars Rovers and the DARPA Grand Challenge (robotic auto race) are impressive but we clearly have a long way to go before your car drives itself. - Vision, voice, pattern recognition, etc. It is difficult to say about face recognition systems, because of their use in security, accuracy rates are secret. I believe they have been oversold. Voice recognition is limited to words and short phrases until we develop better language models with AI behind them. A keyboard is still faster than a microphone. - Interactive learning systems - Integrated intelligent systems Lots of theoretical results, but no real applications. -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/19/06, Matt Mahoney wrote: - NLP components such as parsers, translators, grammar-checkers Parsing is unsolved. Translators like Babelfish have progressed little since the 1959 Russian-English project. Microsoft Word's grammar checker catches some mistakes but is clearly not AI. http://www.charlotte.com/mld/charlotte/news/nation/15783022.htm American soldiers bound for Iraq equipped with laptop translators Called the Two Way Speech-to-Speech Program, it's a translator that uses a computer to convert spoken English to Iraqi Arabic and vice versa. - If it is life-or-death, it must work pretty well. :) I believe this is based on the IBM MASTOR project. http://domino.watson.ibm.com/comm/research.nsf/pages/r.uit.innovation.html MASTOR's innovations include: methods that automatically extract the most likely meaning of the spoken utterance, store it in a tree structured set of concepts like actions and needs, methods that take the tree-based output of a statistical semantic parser and transform the semantic concepts in the tree to express the same set of concepts in a way appropriate for another language; methods for statistical natural language generation that take the resultant set of transformed concepts and generate a sentence for the target language; generation of proper inflections by filtering hypotheses with an n-gram statistical language model; etc BillK - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
Hi Peter, I think in all of the categories you listed, thereshould be a lot ofprogress, but they will hit a ceiling because of the lack of an AGI architecture. It is very clear that vision requires AGI to be complete. So does NLP. In vision, many objects require reasoning to recognize.NLP also requires reasoning to interpret metaphors, which are beyond the scope of current parsers. So thegoal is for vision/NLP researchers to work within some AGI framework.Unfortunately a standard framework isunavailable now. We may start such a framework;lying out the common knowledge representation would be most important. This also shows theneed formodularity and divide-and-conquer. AGI sub-problems like vision and NLP are themselves pretty big projects. So it maybeunwise to try to solve them all alone. I think other candidates that have the potential tobecome AGI are: Cyc, Soar, ACT-R, andother less known cognitive architectures. YKY This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
- Original Message From: BillK [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, October 19, 2006 11:43:46 AM Subject: Re: [agi] SOTA On 10/19/06, Matt Mahoney wrote: - NLP components such as parsers, translators, grammar-checkers Parsing is unsolved. Translators like Babelfish have progressed little since the 1959 Russian-English project. Microsoft Word's grammar checker catches some mistakes but is clearly not AI. http://www.charlotte.com/mld/charlotte/news/nation/15783022.htm I think the problem will eventually be solved. There was a long period of stagnation since the 1959 Russian-English project but I think this period will soon end thanks to better language models due to the recent availability of large text databases, fast hardware, and cheap memory. Once we solve the language modeling problem, we will remove the main barrier to many NLP problems such as speech recognition, translation, OCR, handwriting recognition, and question answering. Google has made good progress in this area using statistical modeling methods and was top ranked in a recent competition. Google has access to terabytes of text in many languages and a custom operating system for running programs in parallel on thousands of PCs. Here is Google's translation of the above article into Arabic and back to English. But as you can see, the job isn't finished. American soldiers heading to Iraq with a laptop translators from Stephanie Hinatz daily newspapers (Newport News,va. (ethnic)نورفولكVa. army-star trip now using similar instrument in Iraq to help the forces of language training without contact with Iraqi civilians and the training of the country's emerging police and military forces. the name of a double discourse to address Albernamjoho translator, which uses computers to convert spoken English Iraqi pwmound and vice versa. while the program is still technically in the research and development stage,Norfolk-based U.S. Joint Forces Command,in conjunction with the Defense Advanced Research projects Agency,some models has been sent to Iraq, 70 troops is used in tactical environments to evaluate its effectiveness. and so far is fine and said Wayne Richards,Commander leadership in the implementation section. the need for such a device for the first time in April 2004 when the joint forces command received an urgent request from commanders on the ground in Abragherichards. soldiers on the ground needed to improve communication with the Iraqi people. But because of the shortage of linguists and translators throughout the Department of Defense do not come from the difficult,even some of the forces of the so-called most important work in Iraq today in Iraq, the training of police and military forces. get those troops trained and capable of maintaining the security of the country itself is a reminder of return for service members to continue der inside and outside the war zone. experts are trying to develop this kind of technical translation for 10 years,He said that Richards. today, in its current form,The translator is the rugged laptop with the plugs are two or loudspeakers and Alsmaatrichards pointing to a model and convert. It is also easy to use Talking on the phone,as evidenced shortly after the Norfolk demonstration Tuesday. I tell you, an Iraqi withdrawal on a computer. you put the microphone up to your mouth. when he said :We are here to provide food and water for your family, You held by the E key to security in a painting keys. you,I wrote to you the text of what we discussed to delight on the screen. you wipe the words to make sure you get exactly. If you can change it manually. when you are convinced you to the t key to the interpretation and sentence looming on the screen once Achrihzh time in Arab Iraq. the computer also says his loud speakers through. the process is the same Balanceof those who did not talk to you. I repeat what you have and the Arab computer will spit on you, the words in the English language. as do translator rights,the program assumes some meanings. not 100% Richards. when I ask,For example,Can the newspaper today, the Arab-language Alanklizihaltrgmeh direct Can the newspaper today. because in any act made in every conversation with the translator is taken. any translation is not due to the past program. Defense Language Institute in California also true of all the translations and Richards. now,because of its size,the best place to use the translator is at the center of command and control or a classroom. It is unlikely that the average Navy will be overseeing the cart with 100 pounds of equipment to implement that attacks in Baghdad, in Sadr City. We hope if the days will be small enough that the sergeant to be implemented in a skirt. Think about it and Richards. sergeant beating on the door of the house formulateseen in Fallujah. a woman answers the door. The soldier's weapon. because it is afraid. the soldier immediately to the effects translator
Re: [agi] SOTA
Matt Mahoney wrote: From: BillK [EMAIL PROTECTED] Parsing is unsolved. Translators like Babelfish have progressed little since the 1959 Russian-English project. Microsoft Word's grammar checker catches some mistakes but is clearly not AI. I think the problem will eventually be solved. There was a long period of stagnation since the 1959 Russian-English project but I think this period will soon end thanks to better language models due to the recent availability of large text databases, fast hardware, and cheap memory. Once we solve the language modeling problem, we will remove the main barrier to many NLP problems such as speech recognition, translation, OCR, handwriting recognition, and question answering. Sorry, but IMO large databases, fast hardware, and cheap memory ain't got nothing to do with it. Anyone who doubts this get a copy of Pim Levelt's Speaking, read and digest the whole thing, and then meditate on the fact that that book is a mere scratch on the surface (IMO a scratch in the wrong direction, too, but that's neither here nor there). I saw a recent talk about an NLP system which left me stupified that so little progress has been made since 20 years ago. Having a clue about just what a complex thing intelligence is, has everything to do with it. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
On 10/19/06, Richard Loosemore [EMAIL PROTECTED] wrote: Sorry, but IMO large databases, fast hardware, and cheap memory ain't got nothing to do with it. Anyone who doubts this get a copy of Pim Levelt's Speaking, read and digest the whole thing, and then meditate on the fact that that book is a mere scratch on the surface (IMO a scratch in the wrong direction, too, but that's neither here nor there). I saw a recent talk about an NLP system which left me stupified that so little progress has been made since 20 years ago. Having a clue about just what a complex thing intelligence is, has everything to do with it. Most normal speaking requires relatively little 'intelligence'. Adults who take young children on foreign holidays are amazed at how quickly the children appear to be chattering away to other children in a foreign language. They manage it for several reasons: 1) they don't have the other interests and priorities that adults have. 2) they use simple sentence structures and smallish vocabularies. 3) they discuss simple subjects of interest to children. The new IBM MASTOR system seems to be better than Babelfish. IBM are just starting on widespread commercial marketing of the system. Aiming at business travellers, apparently. MASTOR project description http://domino.watson.ibm.com/comm/research.nsf/pages/r.uit.innovation.html Here is a pdf file describing the MASTOR system in more detail http://acl.ldc.upenn.edu/W/W06/W06-3711.pdf Here is a 12MB mpg download of the system in use. Simple speech, but impressive. http://www.research.ibm.com/jam/speech_to_speech.mpg BillK - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA
(Excellent list there, Matt)Although Pei Wang makes a good point that the fragmentation of AI does make it difficult to compare projects, it is interesting+ to note the huge differences in the movements in different narrow-AI fields. As has already been mentioned, it is interesting+ to compare the way that progress is very slow in areas such as NLP and Expert Systems, whereas there is significant, albeit gradual progress in physical interaction systems. For instance, the soccer-bots get better every year, cars can now finish DARPA grand challenge -like events in reasonable time... (I personally think that we're fast approaching a critical point where the technology is just good enough to attract more cash and hence more improvement; although meatbags will be better traffic-drivers for a while yet, physical interaction systems can now perform well enough for many applications) Although the question What is State-of-the-Art? won't attract an incontrivertibly good answer, it prompts a lot of bloody good questions that can be answered usefully.-- Olie This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] SOTA
I'm often asked about state-of-the-art in AI, and would like to get some opinions. What do you regard, or what is generally regarded as SOTA in the various AI aspects that may be, or may be seen to be relevant to AGI? For example: - Comprehensive (common-sense) knowledge-bases and/or ontologies - Inference engines, etc. - Adaptive expert systems - Question answering systems - NLP components such as parsers, translators, grammar-checkers - Interactive robotics systems (sensing/ actuation) - physical or virtual - Vision, voice, pattern recognition, etc. - Interactive learning systems - Integrated intelligent systems ... whatever ... I'm looking for the best functionality -- irrespective of proprietary, open-source, or academic. Peter - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
AMD demonstrates the first x86 dual-core processor http://www.digitimes.com/news/a20040831PR200.html Confirms it will re-use the current Opteron 940-pin socket -- Brian Atkins Singularity Institute for Artificial Intelligence http://www.singinst.org/ --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
Opteron system are definitely the sweet spot currently, and for the near future. Rumors are that major server companies are working on 32-way systems to be released soon. Also of course Cray bought that OctigaBay company and now has this: http://www.cray.com/products/systems/xd1/ Also rumor has it that when the dual-core Opterons come out in 2H 2005 you will be able to drop them into most existing motherboards/servers that you can buy right now. So upgradability looks potentially excellent. -- Brian Atkins Singularity Institute for Artificial Intelligence http://www.singinst.org/ --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
David Hart wrote: Because the memory controller resides on the CPU in Opteron systems, all 8 CPUs must be populated, but this can be achieved with the slowest/cheapest model, the Opteron 840 (1.4 GHz). I would second using the cheapest CPU part available, which currently is the 1.6GHz part, you'll save pocket change by going with the 1.4GHz part (for the 2xx series, the difference is about $10). The low clock speed is deceptive. If you use one of the AMD64 optimized compilers (e.g. http://www.pathscale.com) rather than GCC, the real performance is stunning even on slow CPU parts -- it is as fast or faster than pretty much anything else in the general case. For SMP codes, the only thing comparable is the Unix Big Iron (e.g. IBM's Power series boxen) in terms of how it scales across multiple processors. Of all the different architectures I touch, the Opteron is my favorite. Top-notch Big Iron performance at commodity prices. Itanium is a bit better for floating point (PPC970 only for DSP codes), but not much and it is worse at a lot of other codes and expensive. It is worth noting that the Intel AMD64-compatible chips will be ISA compatible, but missing all the features that make the Opterons scalable. And as Brian noted, it has an aggressive looking roadmap. There is a new version of HyperTransport coming out relatively soon (maybe first part of next year?) which will increase the scalability even more, and the multi-core CPUs should be with us shortly. As the big server vendors put more and more money into big Opteron boxes, Linux being the default OS, this looks like the bang/buck champion for the foreseeable future. 64-bit Windows may be viable, but I have no idea if it will run well on the bigger systems as a practical matter. The Windows VM continues to be funky from what I understand, though I have no personal experience. j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
Shane wrote: As for more indirect solutions like RAM disks... I think you would loose at least a factor of ten in speed compared to simply accessing system memory directly as you would need to go through the file system and out to an external device with RAM in it pretending to be a disk. If I remember correctly the fastest disk interfaces you can get for a PC are in the order of 100 MB per second while system RAM is more like a few GB per second. So if you need really high speed then RAM is perhaps your only option I think. For RAM, as with disk, it is more about latency than bandwidth. Even with contrived codes, you can't drive RAM at 100% bandwidth even on architectures with extremely low latency (e.g. Opterons), and on more average architectures (e.g. PPC970) it is not even remotely close to theoretical. For big memory systems a la ccNUMA, the performance is less. A big part of the reason is that the latency limits the number of requests to core that you can make per second regardless of the theoretical bandwidth. The disk may have less bandwidth, about 250MB per channel on a normal high-end HBA, but it can drive that close 100% and it is trivial to run multiple channels in parallel which will get you in the same region of real RAM bandwidth-wise. The big downside is that it gets this performance because it works on big chunks of data compared to real RAM, so the number of discrete fetches per second will probably be an order of magnitude less than system core even under ideal circumstances. I think the biggest argument against using a big RAM disk array is that the HBAs and drivers are optimized for very different types of things than a RAM extension. When you get into big ccNUMA systems, the real memory latencies start to creep within spitting distance of network DMA, arguably the most interesting low-cost alternative. j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
Flash memories could not be used as memory - they withstand no more than 100.000 rewrites. (I do not recall this properly it could be even less then 10.000) Milon Lukasz Kaiser wrote: Hi. Given that your core system is C# this could be a bit of a problem. Just to put my 2c, if you should have the idea to try .NET under linux, better first do some tests. In my experience the linux .NET runtime (mono), although almost fully compatible, is about 5-10 time slower than that on Windows and does not handle over 1GB of memory even if it is there (on 32 bit system at least). Does anyone know how flash drives perform in such setting, can they be used as a slower alternative to RAM ? - lk --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] SOTA 50 GB memory for Windows
Visual Studio (beta) with 64-bit (c#) compilation is available now: http://lab.msdn.microsoft.com/vs2005/productinfo/productline/ as is Windows XP 64-bit for testing. That's all one needs for development. Peter -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Shane Sent: Friday, August 20, 2004 9:57 PM To: [EMAIL PROTECTED] Subject: Re: [agi] SOTA 50 GB memory for Windows Well I guess I have become skeptical about when they will release such a thing as they have been saying that they will put out a 64 bit version for Intel for years now but then always pushing back the release date. No doubt it's related to the difficulties Intel has been having with Itanium. If Intel deliver their AMD compatible 64 bit chips reasonably soon then surely a 64 bit Windows release can't be too far away. The other thing is that when something as big as this changes in the OS it can take a while for various things to straighten themselves out. Things like development tools, devices drivers and so on. I guess for you the key thing is when they will deliver a 64 bit version of C# and associated tools. Curiously, you could get 64 bit Windows for Alpha CPUs about 8 years ago! A friend of mind used to develop things for it way back then, but that version of Windows was eventually killed off by Microsoft. Shane Peter Voss wrote: Microsoft Updates 64-bit Windows XP Preview Editions http://www.eweek.com/article2/0,1759,1637471,00.asp Peter --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- Incoming mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.742 / Virus Database: 495 - Release Date: 8/19/2004 --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.742 / Virus Database: 495 - Release Date: 8/19/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] SOTA 50 GB memory for Windows
What are the best options for large amounts of fast memory for Windows-based systems? I'm looking for price performance (access time) for: 1) Cached RAID 2) RAM disks 3) Internal RAM (using 64 bit architecture?) 4) other Thanks for any info. Peter --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.740 / Virus Database: 494 - Release Date: 8/16/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
I'm looking for price performance (access time) for: 1) Cached RAID This will be useless for runtime VM or pseudo-VM purposes. RAID cache isolates the application from write burst bottlenecks when syncing disks (e.g. checkpointing transaction logs), but that's about it. For flatter I/O patterns, you'll lose 3-4 orders of magnitude access time over non-cached main memory and it won't be appreciably faster than raw spindle. Wrong tool for the application. 2) RAM disks Functionally workable, but very expensive. It is much cheaper per GB to buy the biggest RAM chips you can find and put them on the motherboard. The primary advantage is that you can scale it to very large sizes while only losing somewhere around an order of magnitude versus main core if done well. 3) Internal RAM (using 64 bit architecture?) The best performing, and relatively cheap too. You can slap 32 GB of RAM in an off-the-shelf Opteron system for not much money. The biggest problem is finding motherboards with loads of memory slots and the fact that there is a hard upper bound on how much memory a given system will support. 4) other Nothing I can think of that will work with Windows. There are other performant and cost-effective options for Linux/Unix systems. A compromise might be to max out system RAM within reason (e.g. using 2GB DIMMs), and then using RAM disks on a fast HBA to get the rest of your capacity. All of this will require a 64-bit OS to be efficient. j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] SOTA 50 GB memory for Windows
Thanks Andrew. I didn't realize that RAID cache doesn't help on reads (like RAM disks do). Just how expensive is a high-performance 50GB RAM disk system? Off hand, anyone know progress/ETA on Intel EM64T for .net laguages (c#) ? Also, what Windows compatible machines offer the most RAM ? (Dell seems to max out at 8Gb) Peter -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of J. Andrew Rogers Sent: Friday, August 20, 2004 2:12 PM To: [EMAIL PROTECTED] Subject: Re: [agi] SOTA 50 GB memory for Windows I'm looking for price performance (access time) for: 1) Cached RAID This will be useless for runtime VM or pseudo-VM purposes. RAID cache isolates the application from write burst bottlenecks when syncing disks (e.g. checkpointing transaction logs), but that's about it. For flatter I/O patterns, you'll lose 3-4 orders of magnitude access time over non-cached main memory and it won't be appreciably faster than raw spindle. Wrong tool for the application. 2) RAM disks Functionally workable, but very expensive. It is much cheaper per GB to buy the biggest RAM chips you can find and put them on the motherboard. The primary advantage is that you can scale it to very large sizes while only losing somewhere around an order of magnitude versus main core if done well. 3) Internal RAM (using 64 bit architecture?) The best performing, and relatively cheap too. You can slap 32 GB of RAM in an off-the-shelf Opteron system for not much money. The biggest problem is finding motherboards with loads of memory slots and the fact that there is a hard upper bound on how much memory a given system will support. 4) other Nothing I can think of that will work with Windows. There are other performant and cost-effective options for Linux/Unix systems. A compromise might be to max out system RAM within reason (e.g. using 2GB DIMMs), and then using RAM disks on a fast HBA to get the rest of your capacity. All of this will require a 64-bit OS to be efficient. j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- Incoming mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.740 / Virus Database: 494 - Release Date: 8/16/2004 --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.740 / Virus Database: 494 - Release Date: 8/16/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] SOTA 50 GB memory for Windows
I didn't realize that RAID cache doesn't help on reads (like RAM disks do). Yeah, a lot of people have never really thought about it much. I've worked with database servers for years though, where we actually tuned that type of hardware. The main difference is that a write doesn't return a block to the application, so it can return immediately without physically writing to disk, just putting the blocks in the RAM cache until it has some spare iops to burn or the cache becomes full. For reads though, you have to block until you have physically pulled the block off the disk so that you have something to return. RAID controllers do support predictive read-ahead caching, but that only makes a difference if you have sequential access patterns e.g. streaming large files. Otherwise, it has to block until it pulls data off the spindle every time because it doesn't know what you'll ask for next (unless you get very lucky and the data is in cache). So it is primarily good for streaming large files (e.g. video editing) or buffering write bursts (e.g. database servers). Just how expensive is a high-performance 50GB RAM disk system? Expect to pay ~$2 per MB on the cheap end of things. Or for 50GB, about $100k. Using cheap machines maxed with RAM and RDMA fabrics or similar is a cheaper way to do this for roughly equivalent performance, but you won't be able to do this on Windows -- this is one of those arenas where Linux excels. cheers, j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
Peter, In terms of hardware as far as I know the biggest PC style hardware you can buy supports 32 GB of RAM. For example you can by PCs this big from people like www.penguincomputing.com However there isn't a 64 bit version of Windows on the market nor will there be for some time. Thus your only option is to run something like Linux if you want to have all this data being accessed by code directly in RAM. Given that your core system is C# this could be a bit of a problem. I think this is perhaps the first issue you would need to sort out before thinking about hardware --- the hardware exists alright but with Windows you can't make use of it, at least in terms of having it directly in RAM. As for more indirect solutions like RAM disks... I think you would loose at least a factor of ten in speed compared to simply accessing system memory directly as you would need to go through the file system and out to an external device with RAM in it pretending to be a disk. If I remember correctly the fastest disk interfaces you can get for a PC are in the order of 100 MB per second while system RAM is more like a few GB per second. So if you need really high speed then RAM is perhaps your only option I think. Shane Peter Voss wrote: What are the best options for large amounts of fast memory for Windows-based systems? I'm looking for price performance (access time) for: 1) Cached RAID 2) RAM disks 3) Internal RAM (using 64 bit architecture?) 4) other Thanks for any info. Peter --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.740 / Virus Database: 494 - Release Date: 8/16/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] SOTA 50 GB memory for Windows
Microsoft Updates 64-bit Windows XP Preview Editions http://www.eweek.com/article2/0,1759,1637471,00.asp Peter -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Shane . However there isn't a 64 bit version of Windows on the market nor will there be for some time. Thus your only option is to run something like Linux if you want to have all this data being accessed by code directly in RAM --- Outgoing mail is certified Virus Free. Checked by AVG anti-virus system (http://www.grisoft.com). Version: 6.0.740 / Virus Database: 494 - Release Date: 8/16/2004 --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] SOTA 50 GB memory for Windows
Hi. Given that your core system is C# this could be a bit of a problem. Just to put my 2c, if you should have the idea to try .NET under linux, better first do some tests. In my experience the linux .NET runtime (mono), although almost fully compatible, is about 5-10 time slower than that on Windows and does not handle over 1GB of memory even if it is there (on 32 bit system at least). Does anyone know how flash drives perform in such setting, can they be used as a slower alternative to RAM ? - lk --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
