No, Logan, “go to the kitchen” is NOT to a previously known location. The whole point of the Woz test is that the robot must be able to “go to the kitchen” as any human does – i.e. to ANY kitchen in ANY house – i.e. to UNFAMILIAR/UNKNOWN LOCATIONS. No matter what the house you will be able to go to the kitchen sight unseen And so essentially will any animal – in the sense that you can set an animal to go down any hitherto unknown field or street, and it will make its way down that street round whatever obstacles there are – OR presumably if the animal can smell delectable food from a distant kitchen, it will be able to make its way there through whatever house.
This is what real AGI’s do, this is what narrow AI robots/algos can’t. Living creatures are continually exploring the world, continually going down new paths, into new fields – and doing so in every sense of the terms, at every level of functioning. You clearly think – like pretty everyone else here – that creativity and exploration – going to an unknown location through an unknown field – are pretty exotic. (That’s the effect of narrow AI training) Actually every single real world activity, physical or mental, is creative/exploratory. Going to the toilet is exploratory – you never know what you’ll find there, especially if it’s a new Japanese toilet, – and you may have to find a new way of coping. Reading a text is creative/exploratory because no normal text is formulaically, predictably like any other text – every text presents something new and you may have to find a new way of interpreting it. Ditto any conversation. Ditto writing and reading posts here. The ordinary person, of course, takes all these things for granted - thinks of them as “routine”, potentially algorithmic – but AGI-ers should know better. Real world life/AGI is continuous journeys into more or less new, unknown territories. From: Logan Streondj Sent: Sunday, December 16, 2012 7:37 PM To: AGI Subject: Re: [agi] Internal Representation to be fair the original question was about "going to the kitchen" where word "the" means it's a previously known location. I'm assuming that was answered to your satisfaction. So now you are asking a new question, about how exploration is to occur. Exploration is typically a goal oriented behavior, even if simply to satisfy desire to have new experience. For instance a little bunny may leave the nest for a tiny hop around at first, and then extend it's hops incrementally outside. Once they have matured enough to be independent, they may co-locate in the territory if there are enough resources, or find themselves a new territory. Typically a territory must have a source of water, food, place to make nest or burrow and mating prospects. Most of an animals life is spent within it's own territory as long as all the requirements are met. In some species, either the male or female roams to other territories during mating season. Also there are nomadic species, which have multiple grazing areas which they travel between throughout the year. Anyways so the actual exploration is as simple as enabling a locomotion driver into a direction, typically instincts even in babies have some preventative measures such as avoiding going in directions with large drops, and avoiding walking into walls. If there are specific goals in mind, such as for instance water, then can pattern match to find a good location, such as a lower location, with denser vegetation, and higher humidity. On Sun, Dec 16, 2012 at 2:12 PM, Mike Tintner <[email protected]> wrote: Logan, Yes, as I said, I could see you have a loose sense of what is needed (any sense can only be “loose”). However, your idea that your robot can do this by an algorithm is parallel with Jim’s that complexity is involved. You can only have an algo for a journey if you have already made the journey, and can predict the territory, – and have a route[s]-map. (That incl. “ replotting alternative ways of getting there” after meeting unforeseen obstacles). By extension, you can only have problems of complexity if you already have a known set of possible journeys/route-maps to consider (a la Travelling Salesman Problem).. That’s why algos only work in artificial controlled environments like factories, labs and warehouses, which are fully known and have been artificially structured to be predictable.. To be able to travel in the real world, is to travel through *unpredictable* and *more or less unknown” territories. We want robots which like living creatures can “go through that field” or street or room *blind* - just like you - through an “unstructured field” with “unknown obstacles” - i.e. without ever having seen it before, or encountered anything quite like it, or having any route-map. They will have a rough idea of what such travel may involve – know some of the steps that may be needed to go through a room – but have only partial knowledge of what is needed. The journey can’t be algorithmically plotted in advance – it can only be made up as you go along, and usually requires some innovation – experimenting with the body to take some new or different kinds of steps. Narrow AI/algos are for rational problems, when you already know the way[s] to the solution and the goal – already have the route-map. AGI is about creative problems when you don’t know the way[s], don’t have a route map – and have to make it up as you go along. All creative problems whatsoever are like this. If you think different, try it. What’s the algo for going through “that room”? Oh no, you don’t get to know what the room is in advance. You have to write your algo [sight of room unseen] in advance.Try it. And only then will I tell you what the room is and what’s in it. ... You can’t plan for the real world. You’ve got to “take it as it comes.” From: Logan Streondj Sent: Sunday, December 16, 2012 6:04 PM To: AGI Subject: Re: [agi] Internal Representation In terms of actual movement, that is done quite similarly, to how we have an interpreter with input and output on a screen, simply the outputs go to motors, and inputs are sensors. For the actual "drivers" I was intending to have the AI algorithms available in HSPL by that point, so that the GI could evolve drivers to work with the hardware it has available, and potentially work around hardware failures. So crawling would be defined as a form of locomotion using arms and legs with torso parallel to the floor. Based on a similarly simple description, it could then evolve a process involving the sensors and actuators it has available. Similar to snake robots it would be able to adapt to changes in it's configuration, and layout of objects. In terms of "fluidity" that is part of the process, as the course itself is simply a rough outline of the journey, the walking algorithm would evolve to take into account obstacles and such as part of the evolution of the walking driver. If for instance there is some unforseen obstacle, the plot could be replotted, for alternative potential ways of getting there. This is exactly how humans, do it, likely most animals as well, for instance they figure out where they are going to jump before jumping there. And if they are going to get water, they figure out a good path to take to get there, before actually setting out. On Sun, Dec 16, 2012 at 12:43 PM, Mike Tintner <[email protected]> wrote: PM: You must say why it cannot be done to be helpful, which you haven't. I would have thought a man with your imagination wouldn’t need to be told – the principal reason you cannot have a complex set here, is that you cannot define or calculate a) the trajectories or lines of action a body must take to “go across a room”” – esp. given b) the infinite diversity of starting points and positions of the body, and c) the infinite diversity of obstacles that may lie in the way, and d) the infinite natures and configurations of room terrains, and e) the endless combinations of limbs and body positions that may be necessary to move from the infinite diversity of starting points. The solution lies in the direction of having truly “fluid concepts” of lines of action, with which the body can be fluidly aligned, and proceeding not by prior consideration of a set of trajectories (which is absolutely impossible) but adventurously “putting your best foot forward” towards the goal, one step at a time, and seeing what happens. Logan’s formulations lie intuitively in that direction – but they look like purely symboiic formulations – whereas internal representations must be literally linear in nature. Concepts provide literally “lines of action” - and involve literally “thinking along these lines”. From: Piaget Modeler Sent: Sunday, December 16, 2012 4:54 PM To: AGI Subject: RE: [agi] Internal Representation According to Confucius, Mike Tinter, you are correct. "The person who says it can't be done, and the person who says it can be done are both usually correct." ~ Confucius. If you want to use some imagination, and figure out how it can be done, then by all means. If you just want to nay say and say it can't be done, then that's not helpful. You must say why it cannot be done to be helpful, which you haven't. I think it is already being done, and you simply may not know about it, because certain people aren't revealing their results just yet. ~PM. ---------------------------------------------------------------------------- From: [email protected] To: [email protected] Subject: Re: [agi] Internal Representation Date: Sun, 16 Dec 2012 16:06:49 +0000 “Complexity” is a substitute for thought here. You couldn’t begin to specify what are the complex elements here. The reality is that living systems can translate these desires/goals, however internally represented, into initial actions in a second, and extended courses of action in just a few seconds. The idea that there is some systematic consideration of sets of sets of alternative courses of action and environments – “sets of sets” because the infant/animal could be in an infinite diversity of situations – is quite, quite mad, i.e. divorced from any reality whatsoever other than narrow AI programs which are incapable of this kind of intelligence and action. From: Jim Bromer Sent: Sunday, December 16, 2012 3:09 PM To: AGI Subject: Re: [agi] Internal Representation On Sun, Dec 16, 2012 at 4:43 AM, Mike Tintner <[email protected]> wrote: How do the words translate into a physical course of action? How do you get from “go” to its first and subsequent movements of limbs? Bear in mind, that the same wish – the same formulation – could apply to the infant in vastly diverse physical situations - initial physical positions llike lying, sitting, lying on side, standing against some object etc - and vastly different room configurations. ---------------------------- There is a problem here, but the only true problem is one of complexity. The same problem occurs when the program tries to make sense out of the IO data that is input to the program so that it can recognize what kind of situation that it is in or responding to. The problem also occurs when it has to select the best kind of reaction to the situation when there are many otions that it can select from which are related to the complexities of the situation. There are no serious problems with implementing general AI other then the complexity of these kinds of problems. A situation can be recognized based on many components and there are many 'ideas' (about different kinds of situations) that can be recognized based on some of the components that may occur in the situation. It is a many-to-many kind of interpretation problem. Some problems are much simpler, but the complications is what makes general AI so complex. Jim Bromer On Sun, Dec 16, 2012 at 4:43 AM, Mike Tintner <[email protected]> wrote: Logan, You’re thinking of converting words into more words, here? How do the words translate into a physical course of action? How (i.e. in what forms of representation) would an infant AGI think that hadn’t yet learned language, when it say wants to “go across the room”... crawl, roll, whatever across the carpet to a toy? How do you get from “go” to its first and subsequent movements of limbs? Bear in mind, that the same wish – the same formulation – could apply to the infant in vastly diverse physical situations - initial physical positions llike lying, sitting, lying on side, standing against some object etc - and vastly different room configurations. AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
