view org/sloman.org @ 106:616f461ec98d

add links.
author rlm
date Tue, 03 Jun 2014 13:09:25 -0400
parents a72ac82bb785
children 414a10d51d9f
line wrap: on
line source
1 #+TITLE:Transcript of Aaron Sloman - Artificial Intelligence - Psychology - Oxford Interview
2 #+AUTHOR:Dylan Holmes
3 #+EMAIL:
4 #+STYLE: <link rel="stylesheet" type="text/css" href="../css/sloman.css" />
7 #+BEGIN_QUOTE
23 *Editor's note:* This is a working draft transcript which I made of
24 [[http://www.youtube.com/watch?feature=player_detailpage&v=iuH8dC7Snno][this nice interview]] of Aaron Sloman. Having just finished one
25 iteration of transcription, I still need to go in and clean up the
26 formatting and fix the parts that I misheard, so you can expect the
27 text to improve significantly in the near future.
29 To the extent that this is my work, you have my permission to make
30 copies of this transcript for your own purposes. Also, feel free to
31 e-mail me with comments or corrections.
33 You can send mail to =transcript@aurellem.org=.
35 Cheers,
37 ---Dylan
38 #+END_QUOTE
42 * Introduction
44 ** Aaron Sloman evolves into a philosopher of AI
45 [0:09] My name is Aaron Sloman. My first degree many years ago in
46 Capetown University was in Physics and Mathematics, and I intended to
47 go and be a mathematician. I came to Oxford and encountered
48 philosophers --- I had started reading philosophy and discussing
49 philosophy before then, and then I found that there were philosophers
50 who said things about mathematics that I thought were wrong, so
51 gradually got more and more involved in [philosophy] discussions and
52 switched to doing philosophy DPhil. Then I became a philosophy
53 lecturer and about six years later, I was introduced to artificial
54 intelligence when I was a lecturer at Sussex University in philosophy
55 and I very soon became convinced that the best way to make progress in
56 both areas of philosophy (including philosophy of mathematics which I
57 felt i hadn't dealt with adequately in my DPhil) about the philosophy
58 of mathematics, philosophy of mind, philsophy of language and all
59 those things---the best way was to try to design and test working
60 fragments of mind and maybe eventually put them all together but
61 initially just working fragments that would do various things.
63 [1:12] And I learned to program and ~ with various other people
64 including ~Margaret Boden whom you've interviewed, developed---helped
65 develop an undergraduate degree in AI and other things and also began
66 to do research in AI and so on which I thought of as doing philosophy,
67 primarily.
69 [1:29] And then I later moved to the University of Birmingham and I
70 was there --- I came in 1991 --- and I've been retired for a while but
71 I'm not interested in golf or gardening so I just go on doing full
72 time research and my department is happy to keep me on without paying
73 me and provide space and resources and I come, meeting bright people
74 at conferences and try to learn and make progress if I can.
76 ** AI is hard, in part because there are tempting non-problems.
78 One of the things I learnt and understood more and more over the many
79 years --- forty years or so since I first encountered AI --- is how
80 hard the problems are, and in part that's because it's very often
81 tempting to /think/ the problem is something different from what it
82 actually is, and then people design solutions to the non-problems, and
83 I think of most of my work now as just helping to clarify what the
84 problems are: what is it that we're trying to explain --- and maybe
85 this is leading into what you wanted to talk about:
87 I now think that one of the ways of getting a deep understanding of
88 that is to find out what were the problems that biological evolution
89 solved, because we are a product of /many/ solutions to /many/
90 problems, and if we just try to go in and work out what the whole
91 system is doing, we may get it all wrong, or badly wrong.
94 * What problems of intelligence did evolution solve?
96 ** Intelligence consists of solutions to many evolutionary problems; no single development (e.g. communication) was key to human-level intelligence.
98 [2:57] Well, first I would challenge that we are the dominant
99 species. I know it looks like that but actually if you count biomass,
100 if you count number of species, if you count number of individuals,
101 the dominant species are microbes --- maybe not one of them but anyway
102 they're the ones who dominate in that sense, and furthermore we are
103 mostly --- we are largely composed of microbes, without which we
104 wouldn't survive.
107 # ** Many nonlinguistic competences require sophisticated internal representations
108 [3:27] But there are things that make humans (you could say) best at
109 those things, or worst at those things, but it's a combination. And I
110 think it was a collection of developments of which there isn't any
111 single one. [] there might be, some people say, human language which
112 changed everything. By our human language, they mean human
113 communication in words, but I think that was a later development from
114 what must have started as the use of /internal/ forms of
115 representation --- which are there in nest-building birds, in
116 pre-verbal children, in hunting mammals --- because you can't take in
117 information about a complex structured environment in which things can
118 change and you may have to be able to work out what's possible and
119 what isn't possible, without having some way of representing the
120 components of the environment, their relationships, the kinds of
121 things they can and can't do, the kinds of things you might or might
122 not be able to do --- and /that/ kind of capability needs internal
123 languages, and I and colleagues [at Birmingham] have been referring to
124 them as generalized languages because some people object to
125 referring...to using language to refer to something that isn't used
126 for communication. But from that viewpoint, not only humans but many
127 other animals developed abilities to do things to their environment to
128 make them more friendly to themselves, which depended on being able to
129 represent possible futures, possible actions, and work out what's the
130 best thing to do.
132 [5:13] And nest-building in corvids for instance---crows, magpies,
133 [hawks], and so on --- are way beyond what current robots can do, and
134 in fact I think most humans would be challenged if they had to go and
135 find a collection of twigs, one at a time, maybe bring them with just
136 one hand --- or with your mouth --- and assemble them into a
137 structure that, you know, is shaped like a nest, and is fairly rigid,
138 and you could trust your eggs in them when wind blows. But they're
139 doing it, and so ... they're not our evolutionary ancestors, but
140 they're an indication --- and that example is an indication --- of
141 what must have evolved in order to provide control over the
142 environment in /that/ species.
144 ** Speculation about how communication might have evolved from internal lanagues.
145 [5:56] And I think hunting mammals, fruit-picking mammals, mammals
146 that can rearrange parts of the environment, provide shelters, needed
147 to have .... also needed to have ways of representing possible
148 futures, not just what's there in the environment. I think at a later
149 stage, that developed into a form of communication, or rather the
150 /internal/ forms of representation became usable as a basis for
151 providing [context] to be communicated. And that happened, I think,
152 initially through performing actions that expressed intentions, and
153 probably led to situtations where an action (for instance, moving some
154 large object) was performed more easily, or more successfully, or more
155 accurately if it was done collaboratively. So someone who had worked
156 out what to do might start doing it, and then a conspecific might be
157 able to work out what the intention is, because that person has the
158 /same/ forms of representation and can build theories about what's
159 going on, and might then be able to help.
161 [7:11] You can imagine that if that started happening more (a lot of
162 collaboration based on inferred intentions and plans) then sometimes
163 the inferences might be obscure and difficult, so the /actions/ might
164 be enhanced to provide signals as to what the intention is, and what
165 the best way is to help, and so on.
167 [7:35] So, this is all handwaving and wild speculation, but I think
168 it's consistent with a large collection of facts which one can look at
169 --- and find if one looks for them, but one won't know if [some]one
170 doesn't look for them --- about the way children, for instance, who
171 can't yet talk, communicate, and the things they'll do, like going to
172 the mother and turning the face to point in the direction where the
173 child wants it to look and so on; that's an extreme version of action
174 indicating intention.
176 [8:03] Anyway. That's a very long roundabout answer to one conjecture
177 that the use of communicative language is what gave humans their
178 unique power to create and destroy and whatever, and I'm saying that
179 if by that you mean /communicative/ language, then I'm saying there
180 was something before that which was /non/-communicative language, and I
181 suspect that noncommunicative language continues to play a deep role
182 in /all/ human perception ---in mathematical and scientific reasoning, in
183 problem solving --- and we don't understand very much about it.
185 [8:48]
186 I'm sure there's a lot more to be said about the development of
187 different kinds of senses, the development of brain structures and
188 mechanisms is above all that, but perhaps I've droned on long enough
189 on that question.
192 * How do language and internal states relate to AI?
194 [9:09] Well, I think most of the human and animal capabilities that
195 I've been referring to are not yet to be found in current robots or
196 [computing] systems, and I think there are two reasons for that: one
197 is that it's intrinsically very difficult; I think that in particular
198 it may turn out that the forms of information processing that one can
199 implement on digital computers as we currently know them may not be as
200 well suited to performing some of these tasks as other kinds of
201 computing about which we don't know so much --- for example, I think
202 there may be important special features about /chemical/ computers
203 which we might [talk about in a little bit? find out about].
205 ** In AI, false assumptions can lead investigators astray.
206 [9:57] So, one of the problems then is that the tasks are hard ... but
207 there's a deeper problem as to why AI hasn't made a great deal of
208 progress on these problems that I'm talking about, and that is that
209 most AI researchers assume things---and this is not just AI
210 researchers, but [also] philsophers, and psychologists, and people
211 studying animal behavior---make assumptions about what it is that
212 animals or humans do, for instance make assumptions about what vision
213 is for, or assumptions about what motivation is and how motivation
214 works, or assumptions about how learning works, and then they try ---
215 the AI people try --- to model [or] build systems that perform those
216 assumed functions. So if you get the /functions/ wrong, then even if
217 you implement some of the functions that you're trying to implement,
218 they won't necessarily perform the tasks that the initial objective
219 was to imitate, for instance the tasks that humans, and nest-building
220 birds, and monkeys and so on can perform.
222 ** Example: Vision is not just about finding surfaces, but about finding affordances.
223 [11:09] I'll give you a simple example --- well, maybe not so simple,
224 but --- It's often assumed that the function of vision in humans (and
225 in other animals with good eyesight and so on) is to take in optical
226 information that hits the retina, and form into the (maybe changing
227 --- or, really, in our case definitely changing) patterns of
228 illumination where there are sensory receptors that detect those
229 patterns, and then somehow from that information (plus maybe other
230 information gained from head movement or from comparisons between two
231 eyes) to work out what there was in the environment that produced
232 those patterns, and that is often taken to mean \ldquo{}where were the
233 surfaces off which the light bounced before it came to me\rdquo{}. So
234 you essentially think of the task of the visual system as being to
235 reverse the image formation process: so the 3D structure's there, the
236 lens causes the image to form in the retina, and then the brain goes
237 back to a model of that 3D structure there. That's a very plausible
238 theory about vision, and it may be that that's a /subset/ of what
239 human vision does, but I think James Gibson pointed out that that kind
240 of thing is not necessarily going to be very useful for an organism,
241 and it's very unlikely that that's the main function of perception in
242 general, namely to produce some physical description of what's out
243 there.
245 [12:37] What does an animal /need/? It needs to know what it can do,
246 what it can't do, what the consequences of its actions will be
247 .... so, he introduced the word /affordance/, so from his point of
248 view, the function of vision, perception, are to inform the organism
249 of what the /affordances/ are for action, where that would mean what
250 the animal, /given/ its morphology (what it can do with its mouth, its
251 limbs, and so on, and the ways it can move) what it can do, what its
252 needs are, what the obstacles are, and how the environment supports or
253 obstructs those possible actions.
255 [13:15] And that's a very different collection of information
256 structures that you need from, say, \ldquo{}where are all the
257 surfaces?\rdquo{}: if you've got all the surfaces, /deriving/ the
258 affordances would still be a major task. So, if you think of the
259 perceptual system as primarily (for biological organisms) being
260 devices that provide information about affordances and so on, then the
261 tasks look very different. And most of the people working, doing
262 research on computer vision in robots, I think haven't taken all that
263 on board, so they're trying to get machines to do things which, even
264 if they were successful, would not make the robots very intelligent
265 (and in fact, even the ones they're trying to do are not really easy
266 to do, and they don't succeed very well--- although, there's progress;
267 I shouldn't disparage it too much.)
269 ** Online and offline intelligence
271 [14:10] It gets more complex as animals get more sophisticated. So, I
272 like to make a distinction between online intelligence and offline
273 intelligence. So, for example, if I want to pick something up --- like
274 this leaf <he plucks a leaf from the table> --- I was able to select
275 it from all the others in there, and while moving my hand towards it,
276 I was able to guide its trajectory, making sure it was going roughly
277 in the right direction --- as opposed to going out there, which
278 wouldn't have been able to pick it up --- and these two fingers ended
279 up with a portion of the leaf between them, so that I was able to tell
280 when I'm ready to do that <he clamps the leaf between two fingers>
281 and at that point, I clamped my fingers and then I could pick up the
282 leaf.
284 [14:54] Whereas, --- and that's an example of online intelligence:
285 during the performance of an action (both from the stage where it's
286 initiated, and during the intermediate stages, and where it's
287 completed) I'm taking in information relevant to controlling all those
288 stages, and that relevant information keeps changing. That means I
289 need stores of transient information which gets discarded almost
290 immediately and replaced or something. That's online intelligence. And
291 there are many forms; that's just one example, and Gibson discussed
292 quite a lot of examples which I won't try to replicate now.
294 [15:30] But in offline intelligence, you're not necessarily actually
295 /performing/ the actions when you're using your intelligence; you're
296 thinking about /possible/ actions. So, for instance, I could think
297 about how fast or by what route I would get back to the lecture room
298 if I wanted to [get to the next talk] or something. And I know where
299 the door is, roughly speaking, and I know roughly which route I would
300 take, when I go out, I should go to the left or to the right, because
301 I've stored information about where the spaces are, where the
302 buildings are, where the door was that we came out --- but in using
303 that information to think about that route, I'm not actually
304 performing the action. I'm not even /simulating/ it in detail: the
305 precise details of direction and speed and when to clamp my fingers,
306 or when to contract my leg muscles when walking, are all irrelevant to
307 thinking about a good route, or thinking about the potential things
308 that might happen on the way. Or what would be a good place to meet
309 someone who I think [for an acquaintance in particular] --- [barber]
310 or something --- I don't necessarily have to work out exactly /where/
311 the person's going to stand, or from what angle I would recognize
312 them, and so on.
314 [16:46] So, offline intelligence --- which I think became not just a
315 human competence; I think there are other animals that have aspects of
316 it: Squirrels are very impressive as you watch them. Gray squirrels at
317 any rate, as you watch them defeating squirrel-proof birdfeeders, seem
318 to have a lot of that [offline intelligence], as well as the online
319 intelligence when they eventually perform the action they've worked
320 out [] that will get them to the nuts.
322 [17:16] And I think that what happened during our evolution is that
323 mechanisms for acquiring and processing and storing and manipulating
324 information that is more and more remote from the performance of
325 actions developed. An example is taking in information about where
326 locations are that you might need to go to infrequently: There's a
327 store of a particular type of material that's good for building on
328 roofs of houses or something out around there in some
329 direction. There's a good place to get water somewhere in another
330 direction. There are people that you'd like to go and visit in
331 another place, and so on.
333 [17:59] So taking in information about an extended environment and
334 building it into a structure that you can make use of for different
335 purposes is another example of offline intelligence. And when we do
336 that, we sometimes use only our brains, but in modern times, we also
337 learned how to make maps on paper and walls and so on. And it's not
338 clear whether the stuff inside our heads has the same structures as
339 the maps we make on paper: the maps on paper have a different
340 function; they may be used to communicate with others, or meant for
341 /looking/ at, whereas the stuff in your head you don't /look/ at; you
342 use it in some other way.
344 [18:46] So, what I'm getting at is that there's a great deal of human
345 intelligence (and animal intelligence) which is involved in what's
346 possible in the future, what exists in distant places, what might have
347 happened in the past (sometimes you need to know why something is as
348 it is, because that might be relevant to what you should or shouldn't
349 do in the future, and so on), and I think there was something about
350 human evolution that extended that offline intelligence way beyond
351 that of animals. And I don't think it was /just/ human language, (but
352 human language had something to do with it) but I think there was
353 something else that came earlier than language which involves the
354 ability to use your offline intelligence to discover something that
355 has a rich mathematical structure.
357 ** Example: Even toddlers use sophisticated geometric knowledge
358 #+<<example-gap>>
359 [19:44] I'll give you a simple example: if you look through a gap, you
360 can see something that's on the other side of the gap. Now, you
361 /might/ see what you want to see, or you might see only part of it. If
362 you want to see more of it, which way would you move? Well, you could
363 either move /sideways/, and see through the gap---and see it roughly
364 the same amount but a different part of it [if it's a ????], or you
365 could move /towards/ the gap and then your view will widen as you
366 approach the gap. Now, there's a bit of mathematics in there, insofar
367 as you are implicitly assuming that information travels in straight
368 lines, and as you go closer to a gap, the straight lines that you can
369 draw from where you are through the gap, widen as you approach that
370 gap. Now, there's a kind of theorem of Euclidean geometry in there
371 which I'm not going to try to state very precisely (and as far as I
372 know, wasn't stated explicitly in Euclidean geometry) but it's
373 something every toddler--- human toddler---learns. (Maybe other
374 animals also know it, I don't know.) But there are many more things,
375 actions to perform, to get you more information about things, actions
376 to perform to conceal information from other people, actions that will
377 enable you to operate, to act on a rigid object in one place in order
378 to produce an effect on another place. So, there's a lot of stuff that
379 involves lines and rotations and angles and speeds and so on that I
380 think humans (maybe, to a lesser extent, other animals) develop the
381 ability to think about in a generic way. That means that you could
382 take out the generalizations from the particular contexts and then
383 re-use them in a new contexts in ways that I think are not yet
384 represented at all in AI and in theories of human learning in any []
385 way --- although some people are trying to study learning of mathematics.
387 * Animal intelligence
389 ** The priority is /cataloguing/ what competences have evolved, not ranking them.
390 [22:03] I wasn't going to challenge the claim that humans can do more
391 sophisticated forms of [tracking], just to mention that there are some
392 things that other animals can do which are in some ways comparable,
393 and some ways superior to [things] that humans can do. In particular,
394 there are species of birds and also, I think, some rodents ---
395 squirrels, or something --- I don't know enough about the variety ---
396 that can hide nuts and remember where they've hidden them, and go back
397 to them. And there have been tests which show that some birds are able
398 to hide tens --- you know, [eighteen] or something nuts --- and to
399 remember which ones have been taken, which ones haven't, and so
400 on. And I suspect most humans can't do that. I wouldn't want to say
401 categorically that maybe we couldn't, because humans are very
402 [varied], and also [a few] people can develop particular competences
403 through training. But it's certainly not something I can do.
406 ** AI can be used to test philosophical theories
407 [23:01] But I also would like to say that I am not myself particularly
408 interested in trying to align animal intelligences according to any
409 kind of scale of superiority; I'm just trying to understand what it
410 was that biological evolution produced, and how it works, and I'm
411 interested in AI /mainly/ because I think that when one comes up with
412 theories about how these things work, one needs to have some way of
413 testing the theory. And AI provides ways of implementing and testing
414 theories that were not previously available: Immanuel Kant was trying
415 to come up with theories about how minds work, but he didn't have any
416 kind of a mechanism that he could build to test his theory about the
417 nature of mathematical knowledge, for instance, or how concepts were
418 developed from babyhood onward. Whereas now, if we do develop a
419 theory, we have a criterion of adequacy, namely it should be precise
420 enough and rich enough and detailed to enable a model to be
421 built. And then we can see if it works.
423 [24:07] If it works, it doesn't mean we've proved that the theory is
424 correct; it just shows it's a candidate. And if it doesn't work, then
425 it's not a candidate as it stands; it would need to be modified in
426 some way.
428 * Is abstract general intelligence feasible?
430 ** It's misleading to compare the brain and its neurons to a computer made of transistors
431 [24:27] I think there's a lot of optimism based on false clues:
432 the...for example, one of the false clues is to count the number of
433 neurons in the brain, and then talk about the number of transistors
434 you can fit into a computer or something, and then compare them. It
435 might turn out that the study of the way synapses work (which leads
436 some people to say that a typical synapse [] in the human brain has
437 computational power comparable to the Internet a few years ago,
438 because of the number of different molecules that are doing things,
439 the variety of types of things that are being done in those molecular
440 interactions, and the speed at which they happen, if you somehow count
441 up the number of operations per second or something, then you get
442 these comparable figures).
444 ** For example, brains may rely heavily on chemical information processing
445 Now even if the details aren't right, there may just be a lot of
446 information processing that...going on in brains at the /molecular/
447 level, not the neural level. Then, if that's the case, the processing
448 units will be orders of magnitude larger in number than the number of
449 neurons. And it's certainly the case that all the original biological
450 forms of information processing were chemical; there weren't brains
451 around, and still aren't in most microbes. And even when humans grow
452 their brains, the process of starting from a fertilized egg and
453 producing this rich and complex structure is, for much of the time,
454 under the control of chemical computations, chemical information
455 processing---of course combined with physical sorts of materials and
456 energy and so on as well.
458 [26:25] So it would seem very strange if all that capability was
459 something thrown away when you've got a brain and all the information
460 processing, the [challenges that were handled in making a brain],
461 ... This is handwaving on my part; I'm just saying that we /might/
462 learn that what brains do is not what we think they do, and that
463 problems of replicating them are not what we think they are, solely in
464 terms of numerical estimate of time scales, the number of components,
465 and so on.
467 ** Brain algorithms may simply be optimized for certain kinds of information processing other than bit manipulations
468 [26:56] But apart from that, the other basis of skepticism concerns
469 how well we understand what the problems are. I think there are many
470 people who try to formalize the problems of designing an intelligent
471 system in terms of streams of information thought of as bit streams or
472 collections of bit streams, and they think of as the problems of
473 intelligence as being the construction or detection of patterns in
474 those, and perhaps not just detection of patterns, but detection of
475 patterns that are useable for sending /out/ streams to control motors
476 and so on in order to []. And that way of conceptualizing the problem
477 may lead on the one hand to oversimplification, so that the things
478 that /would/ be achieved, if those goals were achieved, maybe much
479 simpler, in some ways inadequate. Or the replication of human
480 intelligence, or the matching of human intelligence---or for that
481 matter, squirrel intelligence---but in another way, it may also make
482 the problem harder: it may be that some of the kinds of things that
483 biological evolution has achieved can't be done that way. And one of
484 the ways that might turn out to be the case is not because it's not
485 impossible in principle to do some of the information processing on
486 artificial computers-based-on-transistors and other bit-manipulating
487 []---but it may just be that the computational complexity of solving
488 problems, processes, or finding solutions to complex problems, are
489 much greater and therefore you might need a much larger universe than
490 we have available in order to do things.
492 ** Example: find the shortest path by dangling strings
493 [28:55] Then if the underlying mechanisms were different, the
494 information processing mechanisms, they might be better tailored to
495 particular sorts of computation. There's a [] example, which is
496 finding the shortest route if you've got a collection of roads, and
497 they may be curved roads, and lots of tangled routes from A to B to C,
498 and so on. And if you start at A and you want to get to Z --- a place
499 somewhere on that map --- the process of finding the shortest route
500 will involve searching through all these different possibilities and
501 rejecting some that are longer than others and so on. But if you make
502 a model of that map out of string, where these strings are all laid
503 out on the maps and so have the lengths of the routes. Then if you
504 hold the two knots in the string -- it's a network of string --- which
505 correspond to the start point and end point, then /pull/, then the
506 bits of string that you're left with in a straight line will give you
507 the shortest route, and that process of pulling just gets you the
508 solution very rapidly in a parallel computation, where all the others
509 just hang by the wayside, so to speak.
511 ** In sum, we know surprisingly little about the kinds of problems that evolution solved, and the manner in which they were solved.
512 [30:15] Now, I'm not saying brains can build networks of string and
513 pull them or anything like that; that's just an illustration of how if
514 you have the right representation, correctly implemented---or suitably
515 implemented---for a problem, then you can avoid very combinatorially
516 complex searches, which will maybe grow exponentially with the number
517 of components in your map, whereas with this thing, the time it takes
518 won't depend on how many strings you've [got on the map]; you just
519 pull, and it will depend only on the shortest route that exists in
520 there. Even if that shortest route wasn't obvious on the original map.
523 [30:59] So that's a rather long-winded way of formulating the
524 conjecture which---of supporting, a roundabout way of supporting the
525 conjecture that there may be something about the way molecules perform
526 computations where they have the combination of continuous change as
527 things move through space and come together and move apart, and
528 whatever --- and also snap into states that then persist, so [as you
529 learn from] quantum mechanics, you can have stable molecular
530 structures which are quite hard to separate, and then in catalytic
531 processes you can separate them, or extreme temperatures, or strong
532 forces, but they may nevertheless be able to move very rapidly in some
533 conditions in order to perform computations.
535 [31:49] Now there may be things about that kind of structure that
536 enable searching for solutions to /certain/ classes of problems to be
537 done much more efficiently (by brain) than anything we could do with
538 computers. It's just an open question.
540 [32:04] So it /might/ turn out that we need new kinds of technology
541 that aren't on the horizon in order to replicate the functions that
542 animal brains perform ---or, it might not. I just don't know. I'm not
543 claiming that there's strong evidence for that; I'm just saying that
544 it might turn out that way, partly because I think we know less than
545 many people think we know about what biological evolution achieved.
547 [32:28] There are some other possibilities: we may just find out that
548 there are shortcuts no one ever thought of, and it will all happen
549 much more quickly---I have an open mind; I'd be surprised, but it
550 could turn up. There /is/ something that worries me much more than the
551 singularity that most people talk about, which is machines achieving
552 human-level intelligence and perhaps taking over [the] planet or
553 something. There's what I call the /singularity of cognitive
554 catch-up/ ...
556 * A singularity of cognitive catch-up
558 ** What if it will take a lifetime to learn enough to make something new?
559 ... SCC, singularity of cognitive catch-up, which I think we're close
560 to, or maybe have already reached---I'll explain what I mean by
561 that. One of the products of biological evolution---and this is one of
562 the answers to your earlier questions which I didn't get on to---is
563 that humans have not only the ability to make discoveries that none of
564 their ancestors have ever made, but to shorten the time required for
565 similar achievements to be reached by their offspring and their
566 descendants. So once we, for instance, worked out ways of complex
567 computations, or ways of building houses, or ways of finding our way
568 around, we don't need...our children don't need to work it out for
569 themselves by the same lengthy trial and error procedure; we can help
570 them get there much faster.
572 Okay, well, what I've been referring to as the singularity of
573 cognitive catch-up depends on the fact that---fairly obvious, and it's
574 often been commented on---that in case of humans, it's not necessary
575 for each generation to learn what previous generations learned /in the
576 same way/. And we can speed up learning once something has been
577 learned, [it is able to] be learned by new people. And that has meant
578 that the social processes that support that kind of education of the
579 young can enormously accelerate what would have taken...perhaps
580 thousands [or] millions of years for evolution to produce, can happen in
581 a much shorter time.
584 [34:54] But here's the catch: in order for a new advance to happen ---
585 so for something new to be discovered that wasn't there before, like
586 Newtonian mechanics, or the theory of relativity, or Beethoven's music
587 or [style] or whatever --- the individuals have to have traversed a
588 significant amount of what their ancestors have learned, even if they
589 do it much faster than their ancestors, to get to the point where they
590 can see the gaps, the possibilities for going further than their
591 ancestors, or their parents or whatever, have done.
593 [35:27] Now in the case of knowledge of science, mathematics,
594 philosophy, engineering and so on, there's been a lot of accumulated
595 knowledge. And humans are living a /bit/ longer than they used to, but
596 they're still living for [whatever it is], a hundred years, or for
597 most people, less than that. So you can imagine that there might come
598 a time when in a normal human lifespan, it's not possible for anyone
599 to learn enough to understand the scope and limits of what's already
600 been achieved in order to see the potential for going beyond it and to
601 build on what's already been done to make that...those future steps.
603 [36:10] So if we reach that stage, we will have reached the
604 singularity of cognitive catch-up because the process of education
605 that enables individuals to learn faster than their ancestors did is
606 the catching-up process, and it may just be that we at some point
607 reach a point where catching up can only happen within a lifetime of
608 an individual, and after that they're dead and they can't go
609 beyond. And I have some evidence that there's a lot of that around
610 because I see a lot of people coming up with what /they/ think of as
611 new ideas which they've struggled to come up with, but actually they
612 just haven't taken in some of what was...some of what was done [] by
613 other people, in other places before them. And I think that despite
614 the availability of search engines which make it /easier/ for people
615 to get the information---for instance, when I was a student, if I
616 wanted to find out what other people had done in the field, it was a
617 laborious process---going to the library, getting books, and
618 ---whereas now, I can often do things in seconds that would have taken
619 hours. So that means that if seconds [are needed] for that kind of
620 work, my lifespan has been extended by a factor of ten or
621 something. So maybe that /delays/ the singularity, but it may not
622 delay it enough. But that's an open question; I don't know. And it may
623 just be that in some areas, this is more of a problem than others. For
624 instance, it may be that in some kinds of engineering, we're handing
625 over more and more of the work to machines anyways and they can go on
626 doing it. So for instance, most of the production of computers now is
627 done by a computer-controlled machine---although some of the design
628 work is done by humans--- a lot of /detail/ of the design is done by
629 computers, and they produce the next generation, which then produces
630 the next generation, and so on.
632 [37:57] I don't know if humans can go on having major advances, so
633 it'll be kind of sad if we can't.
635 * Spatial reasoning: a difficult problem
637 [38:15] Okay, well, there are different problems [ ] mathematics, and
638 they have to do with properties. So for instance a lot of mathematics
639 that can be expressed in terms of logical structures or algebraic
640 structures and those are pretty well suited for manipulation and...on
641 computers, and if a problem can be specified using the
642 logical/algebraic notation, and the solution method requires creating
643 something in that sort of notation, then computers are pretty good,
644 and there are lots of mathematical tools around---there are theorem
645 provers and theorem checkers, and all kinds of things, which couldn't
646 have existed fifty, sixty years ago, and they will continue getting
647 better.
650 But there was something that I was [[example-gap][alluding to earlier]] when I gave the
651 example of how you can reason about what you will see by changing your
652 position in relation to a door, where what you are doing is using your
653 grasp of spatial structures and how as one spatial relationship
654 changes namely you come closer to the door or move sideways and
655 parallel to the wall or whatever, other spatial relationships change
656 in parallel, so the lines from your eyes through to other parts of
657 the...parts of the room on the other side of the doorway change,
658 spread out more as you go towards the doorway, and as you move
659 sideways, they don't spread out differently, but focus on different
660 parts of the internal ... that they access different parts of the
661 ... of the room.
663 Now, those are examples of ways of thinking about relationships and
664 changing relationships which are not the same as thinking about what
665 happens if I replace this symbol with that symbol, or if I substitute
666 this expression in that expression in a logical formula. And at the
667 moment, I do not believe that there is anything in AI amongst the
668 mathematical reasoning community, the theorem-proving community, that
669 can model the processes that go on when a young child starts learning
670 to do Euclidean geometry and is taught things about---for instance, I
671 can give you a proof that the angles of any triangle add up to a
672 straight line, 180 degrees.
674 ** Example: Spatial proof that the angles of any triangle add up to a half-circle
675 There are standard proofs which involves starting with one triangle,
676 then adding a line parallel to the base one of my former students,
677 Mary Pardoe, came up with which I will demonstrate with this <he holds
678 up a pen> --- can you see it? If I have a triangle here that's got
679 three sides, if I put this thing on it, on one side --- let's say the
680 bottom---I can rotate it until it lies along the second...another
681 side, and then maybe move it up to the other end ~. Then I can rotate
682 it again, until it lies on the third side, and move it back to the
683 other end. And then I'll rotate it again and it'll eventually end up
684 on the original side, but it will have changed the direction it's
685 pointing in --- and it won't have crossed over itself so it will have
686 gone through a half-circle, and that says that the three angles of a
687 triangle add up to the rotations of half a circle, which is a
688 beautiful kind of proof and almost anyone can understand it. Some
689 mathematicians don't like it, because they say it hides some of the
690 assumptions, but nevertheless, as far as I'm concerned, it's an
691 example of a human ability to do reasoning which, once you've
692 understood it, you can see will apply to any triangle --- it's got to
693 be a planar triangle --- not a triangle on a globe, because then the
694 angles can add up to more than ... you can have three /right/ angles
695 if you have an equator...a line on the equator, and a line going up to
696 to the north pole of the earth, and then you have a right angle and
697 then another line going down to the equator, and you have a right
698 angle, right angle, right angle, and they add up to more than a
699 straight line. But that's because the triangle isn't in the plane,
700 it's on a curved surface. In fact, that's one of the
701 differences...definitional differences you can take between planar and
702 curved surfaces: how much the angles of a triangle add up to. But our
703 ability to /visualize/ and notice the generality in that process, and
704 see that you're going to be able to do the same thing using triangles
705 that stretch in all sorts of ways, or if it's a million times as
706 large, or if it's made...you know, written on, on...if it's drawn in
707 different colors or whatever --- none of that's going to make any
708 difference to the essence of that process. And that ability to see
709 the commonality in a spatial structure which enables you to draw some
710 conclusions with complete certainty---subject to the possibility that
711 sometimes you make mistakes, but when you make mistakes, you can
712 discover them, as has happened in the history of geometrical theorem
713 proving. Imre Lakatos had a wonderful book called [[http://en.wikipedia.org/wiki/Proofs_and_Refutations][/Proofs and
714 Refutations/]] --- which I won't try to summarize --- but he has
715 examples: mistakes were made; that was because people didn't always
716 realize there were subtle subcases which had slightly different
717 properties, and they didn't take account of that. But once they're
718 noticed, you rectify that.
720 ** Geometric results are fundamentally different than experimental results in chemistry or physics.
721 [43:28] But it's not the same as doing experiments in chemistry and
722 physics, where you can't be sure it'll be the same on [] or at a high
723 temperature, or in a very strong magnetic field --- with geometric
724 reasoning, in some sense you've got the full information in front of
725 you; even if you don't always notice an important part of it. So, that
726 kind of reasoning (as far as I know) is not implemented anywhere in a
727 computer. And most people who do research on trying to model
728 mathematical reasoning, don't pay any attention to that, because of
729 ... they just don't think about it. They start from somewhere else,
730 maybe because of how they were educated. I was taught Euclidean
731 geometry at school. Were you?
733 (Adam ford: Yeah)
735 Many people are not now. Instead they're taught set theory, and
736 logic, and arithmetic, and [algebra], and so on. And so they don't use
737 that bit of their brains, without which we wouldn't have built any of
738 the cathedrals, and all sorts of things we now depend on.
740 * Is near-term artificial general intelligence likely?
742 ** Two interpretations: a single mechanism for all problems, or many mechanisms unified in one program.
744 [44:35] Well, this relates to what's meant by general. And when I
745 first encountered the AGI community, I thought that what they all
746 meant by general intelligence was /uniform/ intelligence ---
747 intelligence based on some common simple (maybe not so simple, but)
748 single powerful mechanism or principle of inference. And there are
749 some people in the community who are trying to produce things like
750 that, often in connection with algorithmic information theory and
751 computability of information, and so on. But there's another sense of
752 general which means that the system of general intelligence can do
753 lots of different things, like perceive things, understand language,
754 move around, make things, and so on --- perhaps even enjoy a joke;
755 that's something that's not nearly on the horizon, as far as I
756 know. Enjoying a joke isn't the same as being able to make laughing
757 noises.
759 Given, then, that there are these two notions of general
760 intelligence---there's one that looks for one uniform, possibly
761 simple, mechanism or collection of ideas and notations and algorithms,
762 that will deal with any problem that's solvable --- and the other
763 that's general in the sense that it can do lots of different things
764 that are combined into an integrated architecture (which raises lots
765 of questions about how you combine these things and make them work
766 together) and we humans, certainly, are of the second kind: we do all
767 sorts of different things, and other animals also seem to be of the
768 second kind, perhaps not as general as humans. Now, it may turn out
769 that in some near future time, who knows---decades, a few
770 decades---you'll be able to get machines that are capable of solving
771 in a time that will depend on the nature of the problem, but any
772 problem that is solvable, and they will be able to do it in some sort
773 of tractable time --- of course, there are some problems that are
774 solvable that would require a larger universe and a longer history
775 than the history of the universe, but apart from that constraint,
776 these machines will be able to do anything []. But to be able to do
777 some of the kinds of things that humans can do, like the kinds of
778 geometrical reasoning where you look at the shape and you abstract
779 away from the precise angles and sizes and shapes and so on, and
780 realize there's something general here, as must have happened when our
781 ancestors first made the discoveries that eventually put together in
782 Euclidean geometry.
784 It may be that that requires mechanisms of a kind that we don't know
785 anything about at the moment. Maybe brains are using molecules and
786 rearranging molecules in some way that supports that kind of
787 reasoning. I'm not saying they are --- I don't know, I just don't see
788 any simple...any obvious way to map that kind of reasoning capability
789 onto what we currently do on computers. There is---and I just
790 mentioned this briefly beforehand---there is a kind of thing that's
791 sometimes thought of as a major step in that direction, namely you can
792 build a machine (or a software system) that can represent some
793 geometrical structure, and then be told about some change that's going
794 to happen to it, and it can predict in great detail what'll
795 happen. And this happens for instance in game engines, where you say
796 we have all these blocks on the table and I'll drop one other block,
797 and then [the thing] uses Newton's laws and properties of rigidity of
798 the parts and the elasticity and also stuff about geometries and space
799 and so on, to give you a very accurate representation of what'll
800 happen when this brick lands on this pile of things, [it'll bounce and
801 go off, and so on]. And you just, with more memory and more CPU power,
802 you can increase the accuracy--- but that's totally different than
803 looking at /one/ example, and working out what will happen in a whole
804 /range/ of cases at a higher level of abstraction, whereas the game
805 engine does it in great detail for /just/ this case, with /just/ those
806 precise things, and it won't even know what the generalizations are
807 that it's using that would apply to others []. So, in that sense, [we]
808 may get AGI --- artificial general intelligence --- pretty soon, but
809 it'll be limited in what it can do. And the other kind of general
810 intelligence which combines all sorts of different things, including
811 human spatial geometrical reasoning, and maybe other things, like the
812 ability to find things funny, and to appreciate artistic features and
813 other things may need forms of pattern-mechanism, and I have an open
814 mind about that.
816 * Abstract General Intelligence impacts
818 [49:53] Well, as far as the first type's concerned, it could be useful
819 for all kinds of applications --- there are people who worry about
820 where there's a system that has that type of intelligence, might in
821 some sense take over control of the planet. Well, humans often do
822 stupid things, and they might do something stupid that would lead to
823 disaster, but I think it's more likely that there would be other
824 things [] lead to disaster--- population problems, using up all the
825 resources, destroying ecosystems, and whatever. But certainly it would
826 go on being useful to have these calculating devices. Now, as for the
827 second kind of them, I don't know---if we succeeded at putting
828 together all the parts that we find in humans, we might just make an
829 artificial human, and then we might have some of them as your friends,
830 and some of them we might not like, and some of them might become
831 teachers or whatever, composers --- but that raises a question: could
832 they, in some sense, be superior to us, in their learning
833 capabilities, their understanding of human nature, or maybe their
834 wickedness or whatever --- these are all issues in which I expect the
835 best science fiction writers would give better answers than anything I
836 could do, but I did once fantasize when I [back] in 1978, that perhaps
837 if we achieved that kind of thing, that they would be wise, and gentle
838 and kind, and realize that humans are an inferior species that, you
839 know, have some good features, so they'd keep us in some kind of
840 secluded...restrictive kind of environment, keep us away from
841 dangerous weapons, and so on. And find ways of cohabitating with
842 us. But that's just fantasy.
844 Adam Ford: Awesome. Yeah, there's an interesting story /With Folded
845 Hands/ where [the computers] want to take care of us and want to
846 reduce suffering and end up lobotomizing everybody [but] keeping them
847 alive so as to reduce the suffering.
849 Aaron Sloman: Not all that different from /Brave New World/, where it
850 was done with drugs and so on, but different humans are given
851 different roles in that system, yeah.
853 There's also /The Time Machine/, H.G. Wells, where the ... in the
854 distant future, humans have split in two: the Eloi, I think they were
855 called, they lived underground, they were the [] ones, and then---no,
856 the Morlocks lived underground; Eloi lived on the planet; they were
857 pleasant and pretty but not very bright, and so on, and they were fed
858 on by ...
860 Adam Ford: [] in the future.
862 Aaron Sloman: As I was saying, if you ask science fiction writers,
863 you'll probably come up with a wide variety of interesting answers.
865 Adam Ford: I certainly have; I've spoken to [] of Birmingham, and
866 Sean Williams, ... who else?
868 Aaron Sloman: Did you ever read a story by E.M. Forrester called /The
869 Machine Stops/ --- very short story, it's [[http://archive.ncsa.illinois.edu/prajlich/forster.html][on the Internet somewhere]]
870 --- it's about a time when people sitting ... and this was written in
871 about [1914 ] so it's about...over a hundred years ago ... people are
872 in their rooms, they sit in front of screens, and they type things,
873 and they communicate with one another that way, and they don't meet;
874 they have debates, and they give lectures to their audiences that way,
875 and then there's a woman whose son says \ldquo{}I'd like to see
876 you\rdquo{} and she says \ldquo{}What's the point? You've got me at
877 this point \rdquo{} but he wants to come and talk to her --- I won't
878 tell you how it ends, but.
880 Adam Ford: Reminds me of the Internet.
882 Aaron Sloman: Well, yes; he invented ... it was just extraordinary
883 that he was able to do that, before most of the components that we
884 need for it existed.
886 Adam Ford: [Another person who did that] was Vernor Vinge [] /True
887 Names/.
889 Aaron Sloman: When was that written?
891 Adam Ford: The seventies.
893 Aaron Sloman: Okay, well a lot of the technology was already around
894 then. The original bits of internet were working, in about 1973, I was
895 sitting ... 1974, I was sitting at Sussex University trying to
896 use...learn LOGO, the programming language, to decide whether it was
897 going to be useful for teaching AI, and I was sitting [] paper
898 teletype, there was paper coming out, transmitting ten characters a
899 second from Sussex to UCL computer lab by telegraph cable, from there
900 to somewhere in Norway via another cable, from there by satellite to
901 California to a computer Xerox [] research center where they had
902 implemented a computer with a LOGO system on it, with someone I had
903 met previously in Edinburgh, Danny Bobrow, and he allowed me to have
904 access to this sytem. So there I was typing. And furthermore, it was
905 duplex typing, so every character I typed didn't show up on my
906 terminal until it had gone all the way there and echoed back, so I
907 would type, and the characters would come back four seconds later.
909 [55:26] But that was the Internet, and I think Vernor Vinge was
910 writing after that kind of thing had already started, but I don't
911 know. Anyway.
913 [55:41] Another...I mentioned H.G. Wells, /The Time Machine/. I
914 recently discovered, because [[http://en.wikipedia.org/wiki/David_Lodge_(author)][David Lodge]] had written a sort of
915 semi-novel about him, that he had invented Wikipedia, in advance--- he
916 had this notion of an encyclopedia that was free to everybody, and
917 everybody could contribute and [collaborate on it]. So, go to the
918 science fiction writers to find out the future --- well, a range of
919 possible futures.
921 Adam Ford: Well the thing is with science fiction writers, they have
922 to maintain some sort of interest for their readers, after all the
923 science fiction which reaches us is the stuff that publishers want to
924 sell, and so there's a little bit of a ... a bias towards making a
925 plot device there, and so the dramatic sort of appeals to our
926 amygdala, our lizard brain; we'll sort of stay there obviously to some
927 extent. But I think that they do come up with sort of amazing ideas; I
928 think it's worth trying to make these predictions; I think that we
929 should more time on strategic forecasting, I mean take that seriously.
931 Aaron Sloman: Well, I'm happy to leave that to others; I just want to
932 try to understand these problems that bother me about how things
933 work. And it may be that some would say that's irresponsible if I
934 don't think about what the implications will be. Well, understanding
935 how humans work /might/ enable us to make [] humans --- I suspect it
936 wont happen in this century; I think it's going to be too difficult.