Transcript

thus spake their Austra was made famous and Popular by 2001 and uh that is the music played at This Magic Moment when uh some primate suddenly gets an idea presumably one of our ancestors so uh how do we uh explain all that we've got all the ingredients on the table and today I want to talk about various ways of putting those ingredients together so we've talked about representations we've talked about methods and today we're going to talk about architectures and by the end of um the class you'll know how to put one of those things together actually no one knows how to put one of those things together but what you will know is about some alternatives for putting those things together so is to make something that's um arguably intelligent in the same way we are so that's our agenda for the day oh we'll also talk a little bit more about Stories the U I think it was in 2007 when uh the uh estonians moved a war memorial from the center of Talon off to a Russian uh War Cemetery prior to that time the estonians had been building up their National computer networks because they thought that uh computation was the way of the future and networks and all that shortly after the movement of that war memorial uh someone uh brought uh the uh Estonian national network down Cyber attack it was widely believed to be the Russians uh there's a large Russian ethnic population in Estonia to start with and the movement of that war and oil irritated directions and so everybody U everybody thinks that they did it but you know what uh no computer can understand the story I just told they can rubel through all of the worldwide web finding information that's relevant to that but no computer can understand the story I just told except one you'll see a demonstration of that later on today so if we um so by the way if you're interested in understanding nature of human intelligence this is of course the most intelligent most most important lecture of the semester and I should tell you a little bit about what we're going to do on Wednesday because for some reason uh the day before Thanksgiving uh tends to be a a lecture that's lightly populated except in this class because I'm going to talk about um business the artificial intelligence business and what can be learned from it about how to avoid going broke when you start your company so for many of you that will be the most important lecture of the semester yeah good so it all started uh back in the dawn age of artificial intelligence and really it all started at uh Carnegie melon sad to say because people at Carnegie melon notably new and Simon were the first to uh think about sort of a general purpose way of putting things together so as to build a a structure architecture for in which particular intelligent system could be built so their idea was called the general Problem Solver a long name for a simple idea and the simple idea is that you start your life out in a current state call it C and you want to get to some goal state call it s and the way you do that is you measure somehow the symbolic difference between where you are and where you want to be so that's the difference we'll call that difference D and when you uh observe that difference that's enough they say in this uh General approach to problem solving for you to select some operation that will move you from your current state to some new state an intermediate State call that I so I or that operator o is determined by the difference D and then of course the next thing to do is to measure the difference between that intermediate State and the state you want to be in and choose some operator that's relevant to reducing that state so we'll call that D2 and we'll call this O2 and D2 is what leads you to O2 and so it goes so that's the IDE and that's often called means ends analysis right because the end that you want to achieve is being in that final State s and the and the means is a that operator o so you have some notion of where you want to be and the difference between where you are and where you want to be and you pick an operator so as to reduce that difference this is all very abstract let's um exercise it in a in solving a problem that you all be faced with here in a day or two and that is for many of you most of you I hope the problem of going home so here you are you're at MIT and where you want to be is over here at home so you measure the difference between MIT and home and for many of you it's h further than you can go by car and not so far that you can't go at all so what you do is you say well the right operator is taking an airplane so here you are there's the operator take an airplane and this is the difference uh D and the difference D being sufficiently large you take the airplane trouble is if you happen to be sitting here in 10250 there's no way you can take an airplane because they don't fit in here so You' got another problem and that is to get to the airplane so the distance between here and Logan is such that the right way to do that is to take the MBTA and that's determined because you're you working on this difference reduction right here the difference from being at MIT and being at the airport so that difference dictates that you take the MBTA so you see you're recursing but you know uh there are no MBTA cars in here either so there's still a difference like so and that difference dictates that you walk so you've got D1 uh D2 and D3 and by the time you've exercised the operators relevant to those three differences you're at Logan and you take the airplane you get over to your hometown and you're faced with a smaller difference of getting from that airport to where you actually want to go so that's the general Problem Solver idea it's was such an exciting idea at the time it was it was such an exciting idea at the time because people would say to themselves ah this is a general purpose Problem Solver so we can set it onto the problem of making itself smarter and so there was a kind of imagined chain reaction that would take place and the the developers of this architecture warned the public that within 10 years that is to say by about 1970 computer would be generally as smart as people and a lot of people made fun of them for that prediction but it was actually scientists attempting to be responsible because they thought something quite a quite serious dislocation was coming along and the people should know that it was coming and so they felt it was their responsibility in that age of scientific responsibility to warn the public it didn't turn out that way uh because well the problem of collecting the different es and finding the operators that's outside the scope of the architecture so the this is the problem that has to be solved by human before this architecture can be used you have to have identified the differences that you might encounter and The Operators that you might use and build this table which relates the two together so maybe that one that one some off diagonal elements and so on but building that table turned out to be a hard job so not surprisingly the idea evolved and eventually the folks at Carnegie who developed the general Problem Solver most notably Newell and his uh students developed a newer fresher more elaborate architecture called SAR and here's how sore works first of all what does sore mean it doesn't mean anything uh it used to mean State Operator Operator and result but for some reason the proponents of the sore architecture have decided they don't like the acronym and have asserted that Sor is merely a label that wouldn't be thought of as an acronym in any event Sor consists of various parts it has a long-term memory and has a short-term memory and it has connections to the outside world maybe a vision system and an action system but most of the activity of the sore problem solving architecture takes place in a short-term memory so you can view the contents of the long-term memory as shuttling in and out of short-term memory so you can see right away that this mechanism this architecture is heavily influenced by certain cognitive psychology experiments having to do with how much you can hold in your short-term memory nonsense syllables and all that sort of thing it was popular back in those days so this was a architecture devised primarily by psychologists and it had as amongst its features a short-term memory and a long-term memory so that's part one of this architecture now what what's in the what's in the long-term memory well assertions and rules AKA Productions a production being the Cary vernacular for rule it's just rule-based stuff like you saw on almost the first day of class so the whole thing is a gigantic rule-based system with assertions and rules that shuttle back and forth from long-term memory into short-term memory where processing takes place the third thing that you come comes to mind when you think of the sore architectures they had an elaborate preference system you recall that when we talked about rule based systems there was always a question what do you do when more than one rule would work you have to have some way of breaking those ties the sore architecture has an elaborate subsystem for doing that but I said that these are the first three things you think about and maybe that's not right because the next thing you you think about is perhaps a better way of identifi better thing to identify with a sor arure and that's the idea of problem spaces and that's the idea that if you're going to solve a problem you have to develop a space and do a search through that space just like we did when we talked about how we can get from here to home there's a space of places that's our problem space we can do a search through that space to find a way to get from one place to another that's the ort of thing that Sor is focused on finally The Fifth Element you tend to think about when you think about sore is the idea of universal subgoaling and that's the idea that whenever you can can't think of what to do next that becomes your next problem that deserves its own problem space SP and its own set of differences and operators and rules and assertions so you start off on a high level then you have to solve problems at a lower level just like you did up there with the general Problem Solver so if you have these two architectures you can begin to say well what are they centered on and this architecture is this General Problem Solver centered on the idea that everything is about problem solving is the problem solving hypothesis no one gave it that name but that's what it was and this architecture did get its name and it was called Always it was said always by new to be based on what he called the symbol system hypothesis the hypothesis that what we are as humans is symbol manipulators and we can uncover how that all works by giving people crypto arithmetic problems and having them talk out loud by thinking about what happens when you try to remember nonsense syllables by all that sort of stuff that was inv vogus in terms of psychology experiments in in the day when this architecture was first articulated but when you look at architectures you can sort of see where they come from and what their antecedents are uh it has a short-term memory and a long-term memory because new and his associates were cognitive scientists it has assertions and rules and preferences because nle and his associates were also AI people and it has problem spaces and Universal subgoaling because those are ideas that have been worked out in a more primitive form all already already in the general Problem Solver architecture so that's a glimpse of what Sora looked like its early days it's been very highly developed by a lot of smart people so although it's symbol centered they've attached to it things having to do with emotion and perception but generally with the view that the first thing to do when faced with a perception is to get it out of there and get it into a symbolic form that sort of the bias that you that the architecture that the architecture comes with so those are two architectures that are heavily biased toward thinking that the important part of what we do is problem solving but the most uh important perhaps at least from an MIT perspective of these problem solving solving oriented ways of thinking about the world is Marvin Minsky's architecture which he articulates in his book the emotion machine and Marvin is not just concerned with problem solving but also how problem solving might come in layers so let me show you a an example of the sort of problem that motivates some of uh Marvin's thinking so you can read that it's a short little vignette you have no trouble understanding it right no it's not difficult for us humans awfully tough for a computer in part because the thinking you need to the re the the ability to your ability to understand that story requ Ires you to think on many levels at the same time first of all there's a sort of um at the bottom there's instinctive reaction you see where there's instinctive reaction yeah that's the part where she hears a sound and turns her head that's Instinct right that's practically built in but then what she sees is a car so that's something that you know we don't have wired in it would be unlikely that we've evolved in the last 100 years to have an instinctive appreciation of cars barreling down the road so the next level in Marvin's architecture is learn reaction so that's the part about thinking about the car now spread throughout there well let's see where's it particularly good example she decides to Sprint across the road so that's where she's solving a problem so that's the deliberative thinking level but it doesn't stop there because later on she reflects on her impulsive decision so she thinks not only about stuff that's happening out there in the world but she also thinks about what's stuff that's going on in here so that's a level where we can which we can call reflective thinking well you know it doesn't stop there because she also considers in another part of the story something about being uneasy about arriving late so she's not only just thinking about events that are going on in her mind right now but events that are going on right now relative to plans she's made and so Marvin calls that the self-reflecting layer but that isn't the whole thing either because toward the end of the story she starts to worry about what her friends would think of her so there's a kind of reflective thinking in the cont in a more social context so he calls that self-conscious thinking so whereas the KY folks think the S architecture focuses mostly on problem solving Minsky's emotion machine book considers not just thinking but thinking on many layers and the blocker to doing any of that can be said to be uh the development of Common Sense which computers alas have never had much of so this is could be said to be based on the common sense hypothesis and the common sense hypothesis holds that in order to do all of that stuff you have to have common sense like people and if you have to have common sense like people you have to think about well how much of it is how much of that is there and how can we go get it and so this spawn a lot of activity in a media lab amongst people influenced by Marvin having to do with Gathering Common Sense the open mind project the work of Henry leam and others having to do with the Gathering of Common Sense from the worldwide web as a way of populating systems that would lay the foundation for doing this kind of layered thinking so that uh is a brief survey of some mechanisms some older than others but uh all but GPS GPS too let's face it it's hard to think of solving any problem without means ends analysis being involved so GPS isn't wrong it's just not all you need to think it's not all the only tool you need to think about what to do so these are early and late and still current but it's not the only thing there is because there have been reactions against this problemsolving way of thinking about the development of intelligence and the most prominent of those countercurrents of those alternative ideas belongs to Rod Brooks and his subsumption architecture so long about the early long about U oh the years surrounding 1990 uh Brooks became upset subsumption because robots couldn't do much uh they would turn them on at night and then the next morning they' come into the laboratory and they would have moved 25 ft uh nicely avoiding a table perhaps but not doing very much and taking a long time to do it so he decided that it's because people were thinking in the wrong way in those days people thought that the way you build a robot is you build a vision system and then you build a reasoning system and then you build an action system and it can do almost nothing but it does something so you improve the vision system and you improve the reasoning system and you improve the action system and now you've broken it because all the stuff that you used to be able to do doesn't work anymore so what uh what's the alternative well the alternative as articulated by Brooks is to turn this idea on its side so instead of having a a encapsulated vision system and encapsulated reasoning system and encapsulated Action System what you have is layers that are focused not on the sensing and the reasoning and the action but layers that are sens that are specialized to dealing with the world so in Brooks's way of thinking about things at the lowest level you might have a system that's capable of oh before we get to that avoiding objects and maybe the next level up is the wandering layer and maybe the next level up after that is explore and maybe the next level up after that is seek now in the old days when people took 61 I had no trouble um getting an answer the question of what does this remind you of in 61 well doesn't remind you of anything in 61 since you haven't taken it but as a if viewed as a generalization of a programming idea what is the programming idea there are only a few powerful ideas in programming and this is a generalization of one of them what is it do you have a name yes Andrew layers of abstraction and abstraction bound and abstraction barriers that Nails it pretty well because each of these guys can have its own Vision action and reasoning system and if you think of these as abstraction boundaries then when you got this thing working you don't screw with it anymore you build this layer on top and it may reach down in here from time to time but it doesn't fundamentally change it Brooks was inspired in part by the way our brains are constructed all that old stuff that we share with pigs is down in there deep and we put the neocortex over it so it looks layered in a way that would make Jerry susman proud so this then is the way that Brooks looks at the world and it's characterized by a few features just like just like sore is one of those features is no representation so this is a this is a detail that's probably right at the level that Brooks was operating and very questionable when you get above the level that Brooks is operating but before I go on let me say what the what the hypothesis is the hypothesis is the creature hypothesis it's the hypothesis that once you can get a machine to act as smart as an insect then the rest will be easy well how do you get a creature to be smart as an insect maybe you don't need representation we we focused on representation in this course so you can see that there's a little stress when I when when uh next thing is well what do you do if you don't have a representation uh let's see a representation makes a model possible models make it possible possible to predict uh to uh understand to explain and to control so if you don't have one what what can you possibly do well Brooks's answer is you use the world instead of a model so everything you do is reactive you don't have anything in your head that is a map of this room but maybe I don't need one because I can get out get around that table by constantly observing it and I don't have to fill up a memory with that information I can just react to it so no representation uh use a world instead of a model and the mechanisms in their purest form are just finite State machines so with that uh Brooks was able to do things that people were never able to do before and what's the modern uh instantiation of this architecture uh now according to Brooks in use in 5 million homes in the United States yeah the RBA it's the Rumba uh the Rumba robot is um by Brooks's account approximately the 13th business plan of iroot and it's it's the one that made it big because the Rumba vacuum cleaner uh has been very successful would you like to see a movie of its predecessor okay so this is a uh film made some time ago that shows in some sense the Suma of um that architecture what I want you to imagine very briefly is a robot that wanders around in the uh halls and rooms of the old Tech Square collecting the Coke cans okay you all got an image of that in your mind because I want you to compare the image you now have of that robot that's wandering around collecting the Coke can with the actual movie Herbert the soda collecting mobile robot it was built at the MIT AI lab in 198 89 the work was done by John canel under the supervision of Rodney Brooks hervert is a robot controlled by subsumption architecture this is a collection of small behaviors that uh influence the overall activities of the robot there are no centralized controllers and no world model Herbert navigates uh by using a number of infrared proximity sensors around its body and and basically following walls and corridors you can also look for the cams with a laser light striper right now it's come out of the door of an office followed along the wall and then his laser light striper has seen a tan on top of the desk in front of it when this happens a robot stops and deploys its arm you can see the arm going out now the arm has a number of sensors itself uh there fingertip sensors a brake beam in the Jaws and two infrared proximity sensors on the front of the hand it grabs cans in a stereotype fashion first it lowers down to find a surface somewhere then it bounces along the surface until it sees a can in front he uses the hand based IRS to reenter the arm by rotating the robot's body until the can comes between the jaws of the griffer at which point the brake beam senses the can after acquiring the can Herbert will tuck the arm back into its normal traveling configuration and attempt to go home since it has no Central representation it doesn't have any map of where it came from instead it has an algorithm which uses a magnetic compass to determine every time it comes through a door will it be able to find the door basically has a policy of always going north every time it exits the door so now the H can's being tucked away as the robot turns you'll see a red stripe from the Laser Rangefinder and now it's using the body IR to navigate back find the door and go through the door with it prize and there if you were pay attention you saw a little glimpse of John canel who was the uh student who developed that system so that was a tour to force that was a magic moment that when you open a champagne it's not what you expected of course because when I say imagine a robot wandering around in Tech Square picking up the cocan that leaves open a huge emble of possible hallucinations and usually our hallucinations about these things are we imagine things to be more more fluid more natural and more impressive than they actually are but that was impressive because no nobody came close to to doing anything like that uh before more more to be said about that U during the business lecture on on on Wednesday so that's this subsumption AR arure by the way uh maybe at this point we can say something about how the other architectures relate to what Minsky was talking about what's uh this deliberative thinking layer correspond to well that's what that's what sore is about and maybe GPS so what's what's subsumption about it's about stuff down here yeah it's about instinctive reaction and learn reaction but shoot what about Minsky's other if we're going to be building systems that are as smart as those things then we have to worry a little bit about that sort of thing too so that brings us to the to the Genesis uh architecture and now let me give you the standard caution uh that should be early in the presentation of any academic I will sometimes say I and what I mean is we and sometimes I'll say we and what I mean is they uh this was a system that was developed mostly by students of in mine who persuaded me after a great deal of time that they were thinking the right kinds of thoughts but here's how the Genesis architecture works as is no surprise uh given recent discussions it's all centered on language and the language part of the Genesis system has two roles one of which is to guide Marshall and interact with the perceptual systems and the other is to enable the description of events wow nice spelling that's yeah that's how it works so so so it's perception important I don't know I might ask you a question like is there anybody uh sitting in a front row uh wearing blue jeans and it's hard for you to resist under those circumstances your eyes from going over there and answering the question your eyes answer the question no symbol no symbol processing system is involved except in so far as my language system has communicated with your language system which drives your motor system and your vision system to go over there and answer the question for you but it's not just U the real stuff that the language system directs your attention to it's also the imagined stuff shoot it's been a long semester have I told you the story about my table saw probably not here's the deal I bought a table saw It's a Wonderful table saw I was installing it with a friend of mine who's a cabinet maker he said never wear gloves when you operate this saw why I said before you could answer the question I figured it out can you figure out why you never wear gloves when you operated table saw you know what a table saw is right it's a it's a table with a spinning blade in the middle and use it to cut wood and why she never wear gloves yes H well you know the answer it's not fair that's old Brett up there he's heard the story too many times yes Andrew you got it I've been told the answer before you've been told the answer how about somebody who hasn't been told the answer yes because the gloves might get caught because the glove might get caught and pull your hand into the blade and then what happens you know it's horrible your hand gets mangled and your fingers get cut off and this happens a lot to professionals it won't actually happen with that table saw that I bought because it uh its blade detects flesh and and stops the blade which then leads to stopping the blade and having the blade Retreat into the table in about two microseconds two milliseconds so in general though it's a bad idea and you always have toe you always have to suppose that the mechanism isn't working anyway in order to to to do use good safety practice but here's an example of something that nobody ever told you that he was able to figure out by imagining what would happen and reading the answers off of the scene that he imagined so nobody ever says many of the things that we know but we know them anyway here's another example imagine running down the street with a full bucket of water what happens the water splashes out and gets your leg wet right you won't find that in openms datab base nobody ever said that over the web it's not written down anywhere but you know it because you we human beings have a capacity to imagine perceptual things and read the answers to questions off of off of our imag in ation with that perceptual apparatus so that's a very important connection down there and then if you got the ability to describe events then you've got an ability to tell and understand stories and if you can do that then you can start to get a handle on culture both macro and micro and by macroculture I mean you're the country you grew up in the religion you grew up with and by micro I mean your family and personal experience and all shades in between so I don't know what inspires uh me and my associates to think in terms in these terms we talked a little about a little bit of it last time when I talked about Evolution and the apparent flowering of our species about 50,000 years ago at which time we got something and I believe that what we got and this is the characterization of this particular hypothesis what we got is the ability to tell stories and understand them so if we want to label this representation it's the label strong story hypothesis so what's the weak story hypothesis the weak story hypothesis is this is important the strong story hypothesis is this is allar is but is there any other evidence that this is uh really really really important so I've um queried chrishna here before the class starts and he tells me I haven't told you about the following uh experiment this in my way of thinking is is the most most important series of experiments ever done in cognitive psychology developmental psychology actually so here's the uh here's how we get started there's a rectangular room if you're a person if you're a rat it's a rectangular box all the walls are painted white you with me so far so now in each Cor corner there's a basket or a cloth or something in which under which in which on under which you can put some food now you put the food there while the rat watches you and then you give the rat a little spin to disorient it all right so then the rat stops and goes for the food and you can keep track of where the rat goes and the rat goes with approximately equal probability predominantly to those two corners so I bet you didn't know that rats were that smart so they understand the rectangular nature of the room and they don't go to the diagonal Corners where the food cannot be so these are are these genius rats or maybe we're just rats with big brains because we do the same thing so if you repeat this experiment and replace the rat by a with a with a small child and then you you put a toy in there instead of food and the rat and and not the r the child is usually held in a parents arms usually the child's mother uh usually because they think that if they participate in these experiments up there at Harvard they kid will get into Harvard someday so uh the kid goes to the diagonal Corners just like a rat and then uh then you uh next thing you do is you try and an an adult maybe an MIT student in that way case you can use food again and uh you get the same result who could be surprised so rats children and human adults pretty much all the same with the spe to this experiment until you paint one wall blue rats are not color blind in case you're wondering then what happens well if you paint one wall blue the rat still goes with equal probability to the two Di diagonal Corners if you plaint one wall blue the child still goes to the two diagonal Corners with approximately equal probability it's only us genius human adults who go only to that corner so this invites a couple of questions uh one of which is when does a child become an adult any ideas Leica what do you think yeah I me some you can pick a number greater than one and less than 10 manasa what do you think five pretty good guess do you have siblings at that age it's a surprise but why is it five is it because what does it relate to is it any correlate to the onset of that ability you might try everything as spelly does because she's extremely careful so she's tried gender she's tried the onset of language I don't know the appreciation of Music handedness and there's only one thing that matters and that is that the child becomes an adult at that time when they start to use the words left and right when they describe the world now I said that very carefully because they understand left and right in an earlier age but they only have started to use the words left and right when they describe the world at the time that they begin to break this they begin to break this Symmetry and go to the correct corner now for the next element of this I need a something to read has anyone got a textbook handy ah here we go ah China Illustrated history now I need a volunteer okay Andrew you want to do this so here's what you're going to do you can stay there but you can need to stand up so what I'm going to do is I'm going to read to you a passage from this book and I want you to say it back to me at the same time I read it it's as if you are doing in you know uh translation simultaneous translation except it's English to English go just thing's got words I can't pronounce okay you ready to go all right when overwhelmed by the magnitude of the problems he tackled he began to suspect that others were plotting against him or secretly ridiculing himid so it's not thank you very much and that's great so you see he could do it some people can't do it at least takes a little practice but he did it and guess what I've done to him I reduced his intelligence to that of a rat because if you do this experiment with an adult human with who who's who's doing this simultaneous English to English translation they go with equal probability to the two corners so what's happened what's happened is you jammed their language processor and when their language processor is jammed they can't put the blue wall together with a rectangular shape so it seems to be that language is the mediator of exactly the combinators you need in order to build descriptions because they can't even put those things together when their language processor is jammed by the simultaneous translation uh phenomenon so that brings us to the two goalstar ideas of the day and one is if you want to make yourself smarter you want to do those things look listen draw and talk because those are the particular mechanisms that surround this area down in here which is the center of what we do which is the center of our thinking so why do you take notes in class not because you'll ever look at them before but because it forces the engagement of your linguistic of your linguistic your motor and your visual apparatus and that makes you smarter because it's excise izing that stuff second thing you can say uh in conclusion especially from this experiment is beware of fast talkers why do you want to be aware of fast talkers it's not because they will talk you into anything is because when they talk fast they're jamming your language processor and you can't think that's why you want to be aware of fast talkers because if they Jam your language process or you won't think and you'll buy that car or you'll buy that drink or you do any manner of things that people who have want you to do those things have learned to do by by by by talking fast to jam your processor so that completes what we're going to do today and give you a demonstration of some of this stuff on another occasion e

Seek backward 5s ←
Longer seek (10s) Shift ←
Play Space Seek forward 5s →
Longer seek (10s) Shift →
0:00 Auto-scroll A Copy link at current time S

19. Architectures: GPS, SOAR, Subsumption, Society of Mind