Re: [agi] Symbols in search of meaning
In a message dated 2/26/2003 9:47:58 PM Mountain Standard Time, [EMAIL PROTECTED] writes: Human children will learn that certain sound patterns are associated with patterned human behaviour. So very soon (plus or minus one year) children will start to accumulate awareness of words that they know are important because big people around them use those words - but the child has to expend mental effort to discover the meaning of the words. So, once this meta-behaviour is established, it is possible to download symbols into the young NGI and the youngster then begins the laborious task of attaching meaning to the words (derived from both experiential and taught leaning). Agree. This is at least part of the reason that shameless self plug my Rogue-AI project started out from a mostly cognitive linquistics / semiontics base. Could probably win the Loebner prize... Not interested in entering... Hope to start teaching her to understand her source code by the end of the year the year
Re: [agi] seed AI vs Cyc, where does Novamente fall?
Yep. Novamente contains particular mechanisms for converting between declarative and procedural knowledge... something that is learned procedurally can become declarative and vice versa. In fact, if all goes according to plan (a big if of course ;) Novamente should *eventually* be much better at this than the human brain. I'm glad that you choose to incorporate elements of human cognitive theory into Novamente, even if you are not intent on building a brain. Such commonalities will make the design of NM far more intuitive and accessible to designer and lay-person alike. For instance, humans are not very good at making procedural knowledge declarative -- it takes a rare human to be able to explain and understand how they do something they know how to do well. There is a real algorithmic difficulty here, but even so, I think a lot of the difficulty that humans have in doing this is unnecessary, i.e. a consequence of the particular way the brain is structured rather than a consequence of the (admittedly large) difficulty of the problem involved. I disagree that we have a problem converting procedural to declarative for all domains. As an example, I can retrieve a phone number from procedural memory with 1 retrieval operation (watch my fingers dial it). Admittedly this system isn't as slick as one that would work purely internally, it requires performance of the task, but it works. Grammar is tougher, I can test any given rule by using it out in a sentence and seeing how it sounds. But extrapolating all of the rules I use is a tricky problem, in fact it's one we haven't completely finished solving (the rules of English grammar are similar, but not identical to the rules our brains want to use). And then communicating how to swing a golf club is another matter, but I think the limitation there lies in a lack of communication. Our brains have no good way of transmitting or interpreting such fine grained information. And to be fair to our brains, transcribing a motor memory of how to move 10,000 muscles in a very precise sequence into declarative knowledge is an extremely challenging problem. Particularly because that sequence isn't static, but requires feedback from joint sensors. The information isn't just the sequence of neural impulses, it's the substance of the entire network. That said, Novamente would be far better at it than we. With the ability to understand it's own code, NM could just rattle off the relevant parameters into declarative memory. Making this declarative knowledge useful would require understanding how it functions though. That would be the tricky part. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] seed AI vs Cyc, where does Novamente fall?
Hi, I disagree that we have a problem converting procedural to declarative for all domains. Sure, you're right. Here as in many other areas, the human brain's performance is highly domain-variant. That said, Novamente would be far better at it than we. With the ability to understand it's own code, NM could just rattle off the relevant parameters into declarative memory. Making this declarative knowledge useful would require understanding how it functions though. That would be the tricky part. Right. This doesn't require source-code-analysis either, just an understanding of learned parameters existing within the run-time state of the system. Indeed, making the declarative knowledge derived from rattling off parameters describing procedures useful is a HARD problem... but at least Novamente can get the data, which as you have greed, would seem to give AI systems an in-principle advantage over humans in this area... It's hard to overestimate the intelligence-enhancement potential of a more fluid process of interconversion btw declarative and procedural knowledge ... Ben G --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
Indeed, making the declarative knowledge derived from rattling off parameters describing procedures useful is a HARD problem... but at least Novamente can get the data, which as you have greed, would seem to give AI systems an in-principle advantage over humans in this area... It's hard to overestimate the intelligence-enhancement potential of a more fluid process of interconversion btw declarative and procedural knowledge Yes, getting this data is what the entire field of neurophys is about. Being able to extract it without using surgery, electrodes, amplifiers, and gajillions of manhours would be outstanding. A lack of data is the primary thing holding neuroscience back and to a large degree, the depth of cognitive theory over time mirrors the quality of the acquisition and analysis tools. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] seed AI vs Cyc, where does Novamente fall?
Yes, getting this data is what the entire field of neurophys is about. Being able to extract it without using surgery, electrodes, amplifiers, and gajillions of manhours would be outstanding. A lack of data is the primary thing holding neuroscience back and to a large degree, the depth of cognitive theory over time mirrors the quality of the acquisition and analysis tools. -Brad That was exactly my impression when I last looked seriously into neuroscience (1995-96). I wanted to understand cognitive dynamics, and I hoped that tech like PET and fMRI would do the trick. But nothing existing giving the combination of temporal and spatial acuity that you'd need to even make a start on the problem I had a PhD student (Graham Zemunik) who tried to make a detailed model of the cognitive dynamics in a cockroach's brain -- and even that was pretty dicey because the data found by different researchers was often inconsistent. From what you're describing, some headway is finally being made on modeling cognitive dynamics in parts of the rat's brain, and that's a great thing. I've enjoyed following Walter Freeman's work on olfaction in rabbits, but, I've also noticed the pattern of bold hypotheses and partial retractions in his work over time, which is due to the fact that the data is not quite rich enough to support the kind of theorizing he wants to do. Fortunately, neuro-analysis technologies are advancing really fast just like computer chips. In another 10-30 years we will have the data to understand our brains, and the computers and algorithms to crunch this data. (And we may have AI's to do the work for us, who knows ;) -- Ben G --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
That was exactly my impression when I last looked seriously into neuroscience (1995-96). I wanted to understand cognitive dynamics, and I hoped that tech like PET and fMRI would do the trick. But nothing existing giving the combination of temporal and spatial acuity that you'd need to even make a start on the problem I had a PhD student (Graham Zemunik) Just FYI, MEG's (Magnetoencephalography) is a good step in providing temporal precision, but is still a long way from discerning individual neurons. It can basically give us EEG measurements from deep inside the brain without using electrodes(which obviously opens alot of doors for human experimentation) who tried to make a detailed model of the cognitive dynamics in a cockroach's brain -- and even that was pretty dicey because the data found by different researchers was often inconsistent. From what you're I'm sure you know this, but for the benefit of others: Insect brains are much easier to study because the neurons are explicitly laid down by the genetic code. They are identifiable neuron by neuron and are roughly identical from insect to insect (within the same species). The fact that even these networks aren't quite yet understood is a shining example of how far we have to go in understanding the human brain. describing, some headway is finally being made on modeling cognitive dynamics in parts of the rat's brain, and that's a great thing. I've enjoyed following Walter Freeman's work on olfaction in rabbits, but, I've also noticed the pattern of bold hypotheses and partial retractions in his work over time, which is due to the fact that the data is not quite rich enough to support the kind of theorizing he wants to do. I support fringe theorists like Freeman as long as they stay in touch with the community and don't sail off to parts unknown. (Edelman tends to do this). Progress takes all types, the careful, methodical data collectors, and the people on the front lines pushing the theories to extents that the data barely supports. Fortunately, neuro-analysis technologies are advancing really fast just like computer chips. In another 10-30 years we will have the data to understand our brains, and the computers and algorithms to crunch this data. (And we may have AI's to do the work for us, who knows ;) Here's hoping. Although I fear they probably said similar things 10-30 years ago. Only nanotech can get us the type of noninvasive, detailed data that we need. The type of electrodes we currently use are never going to suffice. Lucky for us that the brain uses electrically recordable signals from a structure that is so easily accessible. We'd be in dire straits if the brain used entirely chemical mechanisms and was located in an abdominal sack. Thank you evolution for making our jobs as easy as they are :) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
I actually have a big MEG datafile on my hard drive, which I haven't gotten around to playing with. It consists of about 120 time series, each with about 100,000 points in it. It represents the magnetic field of someone's brain, measured through 120 sensors on their skull, while they sit in a chair and perform an experiment of clicking a button when they see a line appear on a screen. (Pretty exciting, huh?) I got the data from my friend Barak Pearlmutter at UNM, who has spent a few years working on signal processing tools (using blind source signal separation methods) designed to clean up the raw data (basically subtracting off for noise caused by repeated reflection of magnetic fields off the inside of the skull). It's actually a very complicated data analysis. You basically have a spherical surface of data (the sensors), and you are trying to reconstruct the sources and sinks in 3d that created the 2d data you are observing. The problem is underconstrained, because many 3d data sets could produce the same 2d data set, but you try to build in some anatomical assumptions (ie: we know the hippocampus is probably a powerful source/sink, so pin that thumbtack there) to constrain the possible results. As you can imagine it's very weak spatially, but far more precise temporally than PET or FMRI, which can only measure blood flow changes occuring 1 second or more after the source activity. I think combined MEG/FMRI(or was it PET/FMRI) is going to be able to get the best of both worlds. Either way, there are plenty of technological obstacles. I guess that MEG can be used, over time, for stuff subtler than clicking buttons when lines appear. But using it to track the dynamics of thoughts seems a long way off Basically, one needs a lot more than 120 sensors!! ... and then one needs to hope the signal processing code scales well (it probably can be made to do so) Well you can use far more complex behavioral tasks than that even with existing MEG technology(have people navigate a maze, do math, word problems, etc). But in order to get a footing with the new MEG technology, they need to start at the basics so that they can map MEG responses with known EEG signatures available from work that's already been done. The first decade of any new neurophys technique is characterized by a whirlwind of very basic, boring results (usually that create pretty pictures generating funding). Only after the tech has matured do you even begin to hit the cool stuff. I'll bet AI's will be required to analyze the data sets will be getting in the next 20 years. -brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] seed AI vs Cyc, where does Novamente fall?
We need one of the technologies to evolve to the point where it delivers decent spatial AND temporal resolution... That's exactly what I meant actually: combined FMRI and MEG within the same experiment. You get data from each simultaneously and combine them afterwards, using the spatially precise FMRI data to pin down the temporally precise MEG data. It's hard to squeeze a MEG rig into an FMRI machine at the moment, particularly without using ferrous metals (ouch). But I'm sure they'll figure it out in the near future. Fascinating! I didn't know that was possible... But, it's all magnetism, I guess -- the rest is details ;-) ben g --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Symbols in search of meaning - what is the meaning of B31-58-DFT?
Ben, One question is whether it's enough to create general pattern-recognition functionality, and let it deal with seeking meaning for symbols as a subcase of its general behavior. Or does one need to create special heuristics/algorithms/structures just for guiding this particular process? Bit of both I think. Its a bit like there's a search for 'meaning' and a search for 'Meaning'. I think all AGIs need to search for meaning behind patterns to be able to work out useful cause/effect webs. And when AGIs work with symbols this general 'seeking the meaning of patterns' process can be applied as the first level of contemplation. But in the ethical context I think we are after 'Meaning' where this relates to to some notion of the importance of the pattern or symbol for some significant entity - for the AGI, the AGIs mentors, other sentient beings and other life. At the moment you have truth and attention values attached to nodes and links. I'm wondering whether you need to have a third numerical value type relating to 'importance'. Attention has a temporal implication - it's intended to focus significant mental resources on a key issue in the here and now. And truth values indicate the reliability of the data. Neither of these concepts capture the notion of importance. I guess the next question is, what would an AGI do with data on importance. I'm just thinking off the top of my head, but my guess is that if the nodes and links had high importance values but low truth values that this should set up an 'itch' in the system driving the AGI to engage in learning and contemplation that would lift the truth values. Maybe the higher the dissonance between the importance values and the truth values, the more this would stimulate high attention values for the related nodes and links. Then there's the question of what would generate the importance values. I think these values would ultimately be derived from the perceived importance values conveyed by 'significant others' for the AGI and by the AGIs own ethical goal structure. I don't think that preloading symbols and behavior models for something as complex as *ethical issues* is really going to be possible. I think ethical issues and associated behavior models are full of nuances that really need to be learned. Of course ethical issues and associated behavior models are full of nuances that really need to be learned to make much deep sense. Even NGIs like us, with presumably loads of hardwired predisposition to ethical behaviour, can spend their whole life in ethical learning and contemplation! :) So I guess the issues are (a) whether it's worth preload ethical concepts and (b) whether it's possible to do it. I'll start with (b) first and then cosider (a) (since lots of people have a pragmatic tendency not to bother about issues till the means for acting on them are available). (Please bear in mind that I'm not experienced or expert in any of the domains I'm riding rough shod over.everything I say will be intuitive generalist ideas...) Let's take the hardest case first. Let's take the most arcane abstract concept that you can think of or the one that has the most intricate and complex implications/shades of meaning for living. Lets label the concept B31-58-DFT. We create a register in the AGI machinery to store important ethical concepts. We load in the label B31- 58-DFT and we give it a high importance value. We also load in a set of words in quite a few major languages into two other registers - one set of words are considered to have meaning very close to the concept that we have prelabelled as B31-58-DFT. We also load in words that are not the descriptive *meaning* of the B31-58-DFT concept but are often associated with it. We then set the truth value of B31-58-DFT to, say, zero. We also create a GoalNode associated to B31-58-DFT that indicates whether the AGI should link B31-58-DFT to its positive goal structure or to its negative goal structure ie. is B31-58-DFT more of an attractor or a repeller concept? (BTW, most likely there would need to be some system for ensuring that the urge to contemplate concept B31-58-DFT didn't get so strong that the AGI was incapable of doing anything else.) We could also load in some body-language patterns often observed in association with the concept if there are such things in this case eg. smiles on human faces, wagging tails on dogs, purring in cats, etc. (or some other pattern, eg. (1) bared teeth, growling hissing, frowns, red faces; (2) pricked ears, lifted eye brows, quite alterness; and so on). We make sure that the words we load in to the language registers include words that the AGI in the infantile stages of development might most likely associate with concept B31-58-DFT - so that the assocation between the prebuilt info about B31-58-DFT and what the AGI learns early in its life can
