Mercurial > thoughts

annotate org/sloman.org @ 114:b8a6638e6e5e

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
begin additional refinement
author rlm
date Tue, 03 Jun 2014 13:36:48 -0400
parents 414a10d51d9f
children
rev   line source
rlm@57 1 #+TITLE:Transcript of Aaron Sloman - Artificial Intelligence - Psychology - Oxford Interview
rlm@57 2 #+AUTHOR:Dylan Holmes
rlm@57 3 #+EMAIL:
rlm@57 4 #+STYLE: <link rel="stylesheet" type="text/css" href="../css/sloman.css" />
rlm@57 5
rlm@57 6
rlm@109 7 #+begin_quote
rlm@109 8 *Update* (13 Oct): Aaron Sloman has produced an improved version of
rlm@109 9 this transcript, which includes follow-up thoughts and links to
rlm@109 10 related works. It is available on his website here:
rlm@109 11 [[http://www.cs.bham.ac.uk/research/projects/cogaff/movies/transcript-interview.html]].
rlm@109 12
rlm@109 13
rlm@109 14 This draft will remain available here for historical purposes.
rlm@109 15 #+end_quote
rlm@109 16
rlm@57 17 #+BEGIN_QUOTE
rlm@57 18
rlm@57 19
rlm@57 20 *Editor's note:* This is a working draft transcript which I made of
rlm@57 21 [[http://www.youtube.com/watch?feature=player_detailpage&v=iuH8dC7Snno][this nice interview]] of Aaron Sloman. Having just finished one
rlm@57 22 iteration of transcription, I still need to go in and clean up the
rlm@57 23 formatting and fix the parts that I misheard, so you can expect the
rlm@57 24 text to improve significantly in the near future.
rlm@57 25
rlm@57 26 To the extent that this is my work, you have my permission to make
rlm@57 27 copies of this transcript for your own purposes. Also, feel free to
rlm@57 28 e-mail me with comments or corrections.
rlm@57 29
rlm@109 30 (Addendum: This transcription is licensed by Aaron Sloman and Dylan Holmes, as
rlm@109 31 indicated here:
rlm@109 32 http://www.cs.bham.ac.uk/research/projects/cogaff/movies/transcript-interview.html#license)
rlm@109 33
rlm@109 34
rlm@57 35 You can send mail to =transcript@aurellem.org=.
rlm@57 36
rlm@57 37 Cheers,
rlm@57 38
rlm@57 39 ---Dylan
rlm@57 40 #+END_QUOTE
rlm@57 41
rlm@57 42
rlm@57 43
rlm@109 44
rlm@57 45 * Introduction
rlm@57 46
rlm@57 47 ** Aaron Sloman evolves into a philosopher of AI
rlm@57 48 [0:09] My name is Aaron Sloman. My first degree many years ago in
rlm@57 49 Capetown University was in Physics and Mathematics, and I intended to
rlm@57 50 go and be a mathematician. I came to Oxford and encountered
rlm@57 51 philosophers --- I had started reading philosophy and discussing
rlm@57 52 philosophy before then, and then I found that there were philosophers
rlm@57 53 who said things about mathematics that I thought were wrong, so
rlm@57 54 gradually got more and more involved in [philosophy] discussions and
rlm@57 55 switched to doing philosophy DPhil. Then I became a philosophy
rlm@57 56 lecturer and about six years later, I was introduced to artificial
rlm@57 57 intelligence when I was a lecturer at Sussex University in philosophy
rlm@57 58 and I very soon became convinced that the best way to make progress in
rlm@57 59 both areas of philosophy (including philosophy of mathematics which I
rlm@57 60 felt i hadn't dealt with adequately in my DPhil) about the philosophy
rlm@57 61 of mathematics, philosophy of mind, philsophy of language and all
rlm@57 62 those things---the best way was to try to design and test working
rlm@57 63 fragments of mind and maybe eventually put them all together but
rlm@57 64 initially just working fragments that would do various things.
rlm@57 65
rlm@57 66 [1:12] And I learned to program and ~ with various other people
rlm@57 67 including ~Margaret Boden whom you've interviewed, developed---helped
rlm@57 68 develop an undergraduate degree in AI and other things and also began
rlm@57 69 to do research in AI and so on which I thought of as doing philosophy,
rlm@57 70 primarily.
rlm@57 71
rlm@57 72 [1:29] And then I later moved to the University of Birmingham and I
rlm@57 73 was there --- I came in 1991 --- and I've been retired for a while but
rlm@57 74 I'm not interested in golf or gardening so I just go on doing full
rlm@57 75 time research and my department is happy to keep me on without paying
rlm@57 76 me and provide space and resources and I come, meeting bright people
rlm@57 77 at conferences and try to learn and make progress if I can.
rlm@57 78
rlm@57 79 ** AI is hard, in part because there are tempting non-problems.
rlm@57 80
rlm@57 81 One of the things I learnt and understood more and more over the many
rlm@57 82 years --- forty years or so since I first encountered AI --- is how
rlm@57 83 hard the problems are, and in part that's because it's very often
rlm@57 84 tempting to /think/ the problem is something different from what it
rlm@57 85 actually is, and then people design solutions to the non-problems, and
rlm@57 86 I think of most of my work now as just helping to clarify what the
rlm@57 87 problems are: what is it that we're trying to explain --- and maybe
rlm@57 88 this is leading into what you wanted to talk about:
rlm@57 89
rlm@57 90 I now think that one of the ways of getting a deep understanding of
rlm@57 91 that is to find out what were the problems that biological evolution
rlm@57 92 solved, because we are a product of /many/ solutions to /many/
rlm@57 93 problems, and if we just try to go in and work out what the whole
rlm@57 94 system is doing, we may get it all wrong, or badly wrong.
rlm@57 95
rlm@57 96
rlm@57 97 * What problems of intelligence did evolution solve?
rlm@57 98
rlm@57 99 ** Intelligence consists of solutions to many evolutionary problems; no single development (e.g. communication) was key to human-level intelligence.
rlm@57 100
rlm@57 101 [2:57] Well, first I would challenge that we are the dominant
rlm@57 102 species. I know it looks like that but actually if you count biomass,
rlm@57 103 if you count number of species, if you count number of individuals,
rlm@57 104 the dominant species are microbes --- maybe not one of them but anyway
rlm@57 105 they're the ones who dominate in that sense, and furthermore we are
rlm@57 106 mostly --- we are largely composed of microbes, without which we
rlm@57 107 wouldn't survive.
rlm@57 108
rlm@57 109
rlm@57 110 # ** Many nonlinguistic competences require sophisticated internal representations
rlm@57 111 [3:27] But there are things that make humans (you could say) best at
rlm@57 112 those things, or worst at those things, but it's a combination. And I
rlm@57 113 think it was a collection of developments of which there isn't any
rlm@57 114 single one. [] there might be, some people say, human language which
rlm@57 115 changed everything. By our human language, they mean human
rlm@57 116 communication in words, but I think that was a later development from
rlm@57 117 what must have started as the use of /internal/ forms of
rlm@57 118 representation --- which are there in nest-building birds, in
rlm@57 119 pre-verbal children, in hunting mammals --- because you can't take in
rlm@57 120 information about a complex structured environment in which things can
rlm@57 121 change and you may have to be able to work out what's possible and
rlm@57 122 what isn't possible, without having some way of representing the
rlm@57 123 components of the environment, their relationships, the kinds of
rlm@57 124 things they can and can't do, the kinds of things you might or might
rlm@57 125 not be able to do --- and /that/ kind of capability needs internal
rlm@57 126 languages, and I and colleagues [at Birmingham] have been referring to
rlm@57 127 them as generalized languages because some people object to
rlm@57 128 referring...to using language to refer to something that isn't used
rlm@57 129 for communication. But from that viewpoint, not only humans but many
rlm@57 130 other animals developed abilities to do things to their environment to
rlm@57 131 make them more friendly to themselves, which depended on being able to
rlm@57 132 represent possible futures, possible actions, and work out what's the
rlm@57 133 best thing to do.
rlm@57 134
rlm@57 135 [5:13] And nest-building in corvids for instance---crows, magpies,
rlm@57 136 [hawks], and so on --- are way beyond what current robots can do, and
rlm@57 137 in fact I think most humans would be challenged if they had to go and
rlm@57 138 find a collection of twigs, one at a time, maybe bring them with just
rlm@57 139 one hand --- or with your mouth --- and assemble them into a
rlm@57 140 structure that, you know, is shaped like a nest, and is fairly rigid,
rlm@57 141 and you could trust your eggs in them when wind blows. But they're
rlm@57 142 doing it, and so ... they're not our evolutionary ancestors, but
rlm@57 143 they're an indication --- and that example is an indication --- of
rlm@57 144 what must have evolved in order to provide control over the
rlm@57 145 environment in /that/ species.
rlm@57 146
rlm@57 147 ** Speculation about how communication might have evolved from internal lanagues.
rlm@57 148 [5:56] And I think hunting mammals, fruit-picking mammals, mammals
rlm@57 149 that can rearrange parts of the environment, provide shelters, needed
rlm@57 150 to have .... also needed to have ways of representing possible
rlm@57 151 futures, not just what's there in the environment. I think at a later
rlm@57 152 stage, that developed into a form of communication, or rather the
rlm@57 153 /internal/ forms of representation became usable as a basis for
rlm@57 154 providing [context] to be communicated. And that happened, I think,
rlm@57 155 initially through performing actions that expressed intentions, and
rlm@57 156 probably led to situtations where an action (for instance, moving some
rlm@57 157 large object) was performed more easily, or more successfully, or more
rlm@57 158 accurately if it was done collaboratively. So someone who had worked
rlm@57 159 out what to do might start doing it, and then a conspecific might be
rlm@57 160 able to work out what the intention is, because that person has the
rlm@57 161 /same/ forms of representation and can build theories about what's
rlm@57 162 going on, and might then be able to help.
rlm@57 163
rlm@57 164 [7:11] You can imagine that if that started happening more (a lot of
rlm@57 165 collaboration based on inferred intentions and plans) then sometimes
rlm@57 166 the inferences might be obscure and difficult, so the /actions/ might
rlm@57 167 be enhanced to provide signals as to what the intention is, and what
rlm@57 168 the best way is to help, and so on.
rlm@57 169
rlm@57 170 [7:35] So, this is all handwaving and wild speculation, but I think
rlm@57 171 it's consistent with a large collection of facts which one can look at
rlm@57 172 --- and find if one looks for them, but one won't know if [some]one
rlm@57 173 doesn't look for them --- about the way children, for instance, who
rlm@57 174 can't yet talk, communicate, and the things they'll do, like going to
rlm@57 175 the mother and turning the face to point in the direction where the
rlm@57 176 child wants it to look and so on; that's an extreme version of action
rlm@57 177 indicating intention.
rlm@57 178
rlm@57 179 [8:03] Anyway. That's a very long roundabout answer to one conjecture
rlm@57 180 that the use of communicative language is what gave humans their
rlm@57 181 unique power to create and destroy and whatever, and I'm saying that
rlm@57 182 if by that you mean /communicative/ language, then I'm saying there
rlm@57 183 was something before that which was /non/-communicative language, and I
rlm@57 184 suspect that noncommunicative language continues to play a deep role
rlm@57 185 in /all/ human perception ---in mathematical and scientific reasoning, in
rlm@57 186 problem solving --- and we don't understand very much about it.
rlm@57 187
rlm@57 188 [8:48]
rlm@57 189 I'm sure there's a lot more to be said about the development of
rlm@57 190 different kinds of senses, the development of brain structures and
rlm@57 191 mechanisms is above all that, but perhaps I've droned on long enough
rlm@57 192 on that question.
rlm@57 193
rlm@57 194
rlm@57 195 * How do language and internal states relate to AI?
rlm@57 196
rlm@57 197 [9:09] Well, I think most of the human and animal capabilities that
rlm@57 198 I've been referring to are not yet to be found in current robots or
rlm@57 199 [computing] systems, and I think there are two reasons for that: one
rlm@57 200 is that it's intrinsically very difficult; I think that in particular
rlm@57 201 it may turn out that the forms of information processing that one can
rlm@57 202 implement on digital computers as we currently know them may not be as
rlm@57 203 well suited to performing some of these tasks as other kinds of
rlm@57 204 computing about which we don't know so much --- for example, I think
rlm@57 205 there may be important special features about /chemical/ computers
rlm@57 206 which we might [talk about in a little bit? find out about].
rlm@57 207
rlm@57 208 ** In AI, false assumptions can lead investigators astray.
rlm@57 209 [9:57] So, one of the problems then is that the tasks are hard ... but
rlm@57 210 there's a deeper problem as to why AI hasn't made a great deal of
rlm@57 211 progress on these problems that I'm talking about, and that is that
rlm@57 212 most AI researchers assume things---and this is not just AI
rlm@57 213 researchers, but [also] philsophers, and psychologists, and people
rlm@57 214 studying animal behavior---make assumptions about what it is that
rlm@57 215 animals or humans do, for instance make assumptions about what vision
rlm@57 216 is for, or assumptions about what motivation is and how motivation
rlm@57 217 works, or assumptions about how learning works, and then they try ---
rlm@57 218 the AI people try --- to model [or] build systems that perform those
rlm@57 219 assumed functions. So if you get the /functions/ wrong, then even if
rlm@57 220 you implement some of the functions that you're trying to implement,
rlm@57 221 they won't necessarily perform the tasks that the initial objective
rlm@57 222 was to imitate, for instance the tasks that humans, and nest-building
rlm@57 223 birds, and monkeys and so on can perform.
rlm@57 224
rlm@57 225 ** Example: Vision is not just about finding surfaces, but about finding affordances.
rlm@57 226 [11:09] I'll give you a simple example --- well, maybe not so simple,
rlm@57 227 but --- It's often assumed that the function of vision in humans (and
rlm@57 228 in other animals with good eyesight and so on) is to take in optical
rlm@57 229 information that hits the retina, and form into the (maybe changing
rlm@57 230 --- or, really, in our case definitely changing) patterns of
rlm@57 231 illumination where there are sensory receptors that detect those
rlm@57 232 patterns, and then somehow from that information (plus maybe other
rlm@57 233 information gained from head movement or from comparisons between two
rlm@57 234 eyes) to work out what there was in the environment that produced
rlm@57 235 those patterns, and that is often taken to mean \ldquo{}where were the
rlm@57 236 surfaces off which the light bounced before it came to me\rdquo{}. So
rlm@57 237 you essentially think of the task of the visual system as being to
rlm@57 238 reverse the image formation process: so the 3D structure's there, the
rlm@57 239 lens causes the image to form in the retina, and then the brain goes
rlm@57 240 back to a model of that 3D structure there. That's a very plausible
rlm@57 241 theory about vision, and it may be that that's a /subset/ of what
rlm@57 242 human vision does, but I think James Gibson pointed out that that kind
rlm@57 243 of thing is not necessarily going to be very useful for an organism,
rlm@57 244 and it's very unlikely that that's the main function of perception in
rlm@57 245 general, namely to produce some physical description of what's out
rlm@57 246 there.
rlm@57 247
rlm@57 248 [12:37] What does an animal /need/? It needs to know what it can do,
rlm@57 249 what it can't do, what the consequences of its actions will be
rlm@57 250 .... so, he introduced the word /affordance/, so from his point of
rlm@57 251 view, the function of vision, perception, are to inform the organism
rlm@57 252 of what the /affordances/ are for action, where that would mean what
rlm@57 253 the animal, /given/ its morphology (what it can do with its mouth, its
rlm@57 254 limbs, and so on, and the ways it can move) what it can do, what its
rlm@57 255 needs are, what the obstacles are, and how the environment supports or
rlm@57 256 obstructs those possible actions.
rlm@57 257
rlm@57 258 [13:15] And that's a very different collection of information
rlm@57 259 structures that you need from, say, \ldquo{}where are all the
rlm@57 260 surfaces?\rdquo{}: if you've got all the surfaces, /deriving/ the
rlm@57 261 affordances would still be a major task. So, if you think of the
rlm@57 262 perceptual system as primarily (for biological organisms) being
rlm@57 263 devices that provide information about affordances and so on, then the
rlm@57 264 tasks look very different. And most of the people working, doing
rlm@57 265 research on computer vision in robots, I think haven't taken all that
rlm@57 266 on board, so they're trying to get machines to do things which, even
rlm@57 267 if they were successful, would not make the robots very intelligent
rlm@57 268 (and in fact, even the ones they're trying to do are not really easy
rlm@57 269 to do, and they don't succeed very well--- although, there's progress;
rlm@57 270 I shouldn't disparage it too much.)
rlm@57 271
rlm@57 272 ** Online and offline intelligence
rlm@57 273
rlm@57 274 [14:10] It gets more complex as animals get more sophisticated. So, I
rlm@57 275 like to make a distinction between online intelligence and offline
rlm@57 276 intelligence. So, for example, if I want to pick something up --- like
rlm@57 277 this leaf <he plucks a leaf from the table> --- I was able to select
rlm@57 278 it from all the others in there, and while moving my hand towards it,
rlm@57 279 I was able to guide its trajectory, making sure it was going roughly
rlm@57 280 in the right direction --- as opposed to going out there, which
rlm@57 281 wouldn't have been able to pick it up --- and these two fingers ended
rlm@57 282 up with a portion of the leaf between them, so that I was able to tell
rlm@57 283 when I'm ready to do that <he clamps the leaf between two fingers>
rlm@57 284 and at that point, I clamped my fingers and then I could pick up the
rlm@57 285 leaf.
rlm@57 286
rlm@57 287 [14:54] Whereas, --- and that's an example of online intelligence:
rlm@57 288 during the performance of an action (both from the stage where it's
rlm@57 289 initiated, and during the intermediate stages, and where it's
rlm@57 290 completed) I'm taking in information relevant to controlling all those
rlm@57 291 stages, and that relevant information keeps changing. That means I
rlm@57 292 need stores of transient information which gets discarded almost
rlm@57 293 immediately and replaced or something. That's online intelligence. And
rlm@57 294 there are many forms; that's just one example, and Gibson discussed
rlm@57 295 quite a lot of examples which I won't try to replicate now.
rlm@57 296
rlm@57 297 [15:30] But in offline intelligence, you're not necessarily actually
rlm@57 298 /performing/ the actions when you're using your intelligence; you're
rlm@57 299 thinking about /possible/ actions. So, for instance, I could think
rlm@57 300 about how fast or by what route I would get back to the lecture room
rlm@57 301 if I wanted to [get to the next talk] or something. And I know where
rlm@57 302 the door is, roughly speaking, and I know roughly which route I would
rlm@57 303 take, when I go out, I should go to the left or to the right, because
rlm@57 304 I've stored information about where the spaces are, where the
rlm@57 305 buildings are, where the door was that we came out --- but in using
rlm@57 306 that information to think about that route, I'm not actually
rlm@57 307 performing the action. I'm not even /simulating/ it in detail: the
rlm@57 308 precise details of direction and speed and when to clamp my fingers,
rlm@57 309 or when to contract my leg muscles when walking, are all irrelevant to
rlm@57 310 thinking about a good route, or thinking about the potential things
rlm@57 311 that might happen on the way. Or what would be a good place to meet
rlm@57 312 someone who I think [for an acquaintance in particular] --- [barber]
rlm@57 313 or something --- I don't necessarily have to work out exactly /where/
rlm@57 314 the person's going to stand, or from what angle I would recognize
rlm@57 315 them, and so on.
rlm@57 316
rlm@57 317 [16:46] So, offline intelligence --- which I think became not just a
rlm@57 318 human competence; I think there are other animals that have aspects of
rlm@57 319 it: Squirrels are very impressive as you watch them. Gray squirrels at
rlm@57 320 any rate, as you watch them defeating squirrel-proof birdfeeders, seem
rlm@57 321 to have a lot of that [offline intelligence], as well as the online
rlm@57 322 intelligence when they eventually perform the action they've worked
rlm@57 323 out [] that will get them to the nuts.
rlm@57 324
rlm@57 325 [17:16] And I think that what happened during our evolution is that
rlm@57 326 mechanisms for acquiring and processing and storing and manipulating
rlm@57 327 information that is more and more remote from the performance of
rlm@57 328 actions developed. An example is taking in information about where
rlm@57 329 locations are that you might need to go to infrequently: There's a
rlm@57 330 store of a particular type of material that's good for building on
rlm@57 331 roofs of houses or something out around there in some
rlm@57 332 direction. There's a good place to get water somewhere in another
rlm@57 333 direction. There are people that you'd like to go and visit in
rlm@57 334 another place, and so on.
rlm@57 335
rlm@57 336 [17:59] So taking in information about an extended environment and
rlm@57 337 building it into a structure that you can make use of for different
rlm@57 338 purposes is another example of offline intelligence. And when we do
rlm@57 339 that, we sometimes use only our brains, but in modern times, we also
rlm@57 340 learned how to make maps on paper and walls and so on. And it's not
rlm@57 341 clear whether the stuff inside our heads has the same structures as
rlm@57 342 the maps we make on paper: the maps on paper have a different
rlm@57 343 function; they may be used to communicate with others, or meant for
rlm@57 344 /looking/ at, whereas the stuff in your head you don't /look/ at; you
rlm@57 345 use it in some other way.
rlm@57 346
rlm@57 347 [18:46] So, what I'm getting at is that there's a great deal of human
rlm@57 348 intelligence (and animal intelligence) which is involved in what's
rlm@57 349 possible in the future, what exists in distant places, what might have
rlm@57 350 happened in the past (sometimes you need to know why something is as
rlm@57 351 it is, because that might be relevant to what you should or shouldn't
rlm@57 352 do in the future, and so on), and I think there was something about
rlm@57 353 human evolution that extended that offline intelligence way beyond
rlm@57 354 that of animals. And I don't think it was /just/ human language, (but
rlm@57 355 human language had something to do with it) but I think there was
rlm@57 356 something else that came earlier than language which involves the
rlm@57 357 ability to use your offline intelligence to discover something that
rlm@57 358 has a rich mathematical structure.
rlm@57 359
rlm@57 360 ** Example: Even toddlers use sophisticated geometric knowledge
rlm@57 361 #+<<example-gap>>
rlm@57 362 [19:44] I'll give you a simple example: if you look through a gap, you
rlm@57 363 can see something that's on the other side of the gap. Now, you
rlm@57 364 /might/ see what you want to see, or you might see only part of it. If
rlm@57 365 you want to see more of it, which way would you move? Well, you could
rlm@57 366 either move /sideways/, and see through the gap---and see it roughly
rlm@57 367 the same amount but a different part of it [if it's a ????], or you
rlm@57 368 could move /towards/ the gap and then your view will widen as you
rlm@57 369 approach the gap. Now, there's a bit of mathematics in there, insofar
rlm@57 370 as you are implicitly assuming that information travels in straight
rlm@57 371 lines, and as you go closer to a gap, the straight lines that you can
rlm@57 372 draw from where you are through the gap, widen as you approach that
rlm@57 373 gap. Now, there's a kind of theorem of Euclidean geometry in there
rlm@57 374 which I'm not going to try to state very precisely (and as far as I
rlm@57 375 know, wasn't stated explicitly in Euclidean geometry) but it's
rlm@57 376 something every toddler--- human toddler---learns. (Maybe other
rlm@57 377 animals also know it, I don't know.) But there are many more things,
rlm@57 378 actions to perform, to get you more information about things, actions
rlm@57 379 to perform to conceal information from other people, actions that will
rlm@57 380 enable you to operate, to act on a rigid object in one place in order
rlm@57 381 to produce an effect on another place. So, there's a lot of stuff that
rlm@57 382 involves lines and rotations and angles and speeds and so on that I
rlm@57 383 think humans (maybe, to a lesser extent, other animals) develop the
rlm@57 384 ability to think about in a generic way. That means that you could
rlm@57 385 take out the generalizations from the particular contexts and then
rlm@57 386 re-use them in a new contexts in ways that I think are not yet
rlm@57 387 represented at all in AI and in theories of human learning in any []
rlm@57 388 way --- although some people are trying to study learning of mathematics.
rlm@57 389
rlm@57 390 * Animal intelligence
rlm@57 391
rlm@57 392 ** The priority is /cataloguing/ what competences have evolved, not ranking them.
rlm@57 393 [22:03] I wasn't going to challenge the claim that humans can do more
rlm@57 394 sophisticated forms of [tracking], just to mention that there are some
rlm@57 395 things that other animals can do which are in some ways comparable,
rlm@57 396 and some ways superior to [things] that humans can do. In particular,
rlm@57 397 there are species of birds and also, I think, some rodents ---
rlm@57 398 squirrels, or something --- I don't know enough about the variety ---
rlm@57 399 that can hide nuts and remember where they've hidden them, and go back
rlm@57 400 to them. And there have been tests which show that some birds are able
rlm@57 401 to hide tens --- you know, [eighteen] or something nuts --- and to
rlm@57 402 remember which ones have been taken, which ones haven't, and so
rlm@57 403 on. And I suspect most humans can't do that. I wouldn't want to say
rlm@57 404 categorically that maybe we couldn't, because humans are very
rlm@57 405 [varied], and also [a few] people can develop particular competences
rlm@57 406 through training. But it's certainly not something I can do.
rlm@57 407
rlm@57 408
rlm@57 409 ** AI can be used to test philosophical theories
rlm@57 410 [23:01] But I also would like to say that I am not myself particularly
rlm@57 411 interested in trying to align animal intelligences according to any
rlm@57 412 kind of scale of superiority; I'm just trying to understand what it
rlm@57 413 was that biological evolution produced, and how it works, and I'm
rlm@57 414 interested in AI /mainly/ because I think that when one comes up with
rlm@57 415 theories about how these things work, one needs to have some way of
rlm@57 416 testing the theory. And AI provides ways of implementing and testing
rlm@57 417 theories that were not previously available: Immanuel Kant was trying
rlm@57 418 to come up with theories about how minds work, but he didn't have any
rlm@57 419 kind of a mechanism that he could build to test his theory about the
rlm@57 420 nature of mathematical knowledge, for instance, or how concepts were
rlm@57 421 developed from babyhood onward. Whereas now, if we do develop a
rlm@57 422 theory, we have a criterion of adequacy, namely it should be precise
rlm@57 423 enough and rich enough and detailed to enable a model to be
rlm@57 424 built. And then we can see if it works.
rlm@57 425
rlm@57 426 [24:07] If it works, it doesn't mean we've proved that the theory is
rlm@57 427 correct; it just shows it's a candidate. And if it doesn't work, then
rlm@57 428 it's not a candidate as it stands; it would need to be modified in
rlm@57 429 some way.
rlm@57 430
rlm@57 431 * Is abstract general intelligence feasible?
rlm@57 432
rlm@57 433 ** It's misleading to compare the brain and its neurons to a computer made of transistors
rlm@57 434 [24:27] I think there's a lot of optimism based on false clues:
rlm@57 435 the...for example, one of the false clues is to count the number of
rlm@57 436 neurons in the brain, and then talk about the number of transistors
rlm@57 437 you can fit into a computer or something, and then compare them. It
rlm@57 438 might turn out that the study of the way synapses work (which leads
rlm@57 439 some people to say that a typical synapse [] in the human brain has
rlm@57 440 computational power comparable to the Internet a few years ago,
rlm@57 441 because of the number of different molecules that are doing things,
rlm@57 442 the variety of types of things that are being done in those molecular
rlm@57 443 interactions, and the speed at which they happen, if you somehow count
rlm@57 444 up the number of operations per second or something, then you get
rlm@57 445 these comparable figures).
rlm@57 446
rlm@57 447 ** For example, brains may rely heavily on chemical information processing
rlm@57 448 Now even if the details aren't right, there may just be a lot of
rlm@57 449 information processing that...going on in brains at the /molecular/
rlm@57 450 level, not the neural level. Then, if that's the case, the processing
rlm@57 451 units will be orders of magnitude larger in number than the number of
rlm@57 452 neurons. And it's certainly the case that all the original biological
rlm@57 453 forms of information processing were chemical; there weren't brains
rlm@57 454 around, and still aren't in most microbes. And even when humans grow
rlm@57 455 their brains, the process of starting from a fertilized egg and
rlm@57 456 producing this rich and complex structure is, for much of the time,
rlm@57 457 under the control of chemical computations, chemical information
rlm@57 458 processing---of course combined with physical sorts of materials and
rlm@57 459 energy and so on as well.
rlm@57 460
rlm@57 461 [26:25] So it would seem very strange if all that capability was
rlm@57 462 something thrown away when you've got a brain and all the information
rlm@57 463 processing, the [challenges that were handled in making a brain],
rlm@57 464 ... This is handwaving on my part; I'm just saying that we /might/
rlm@57 465 learn that what brains do is not what we think they do, and that
rlm@57 466 problems of replicating them are not what we think they are, solely in
rlm@57 467 terms of numerical estimate of time scales, the number of components,
rlm@57 468 and so on.
rlm@57 469
rlm@57 470 ** Brain algorithms may simply be optimized for certain kinds of information processing other than bit manipulations
rlm@57 471 [26:56] But apart from that, the other basis of skepticism concerns
rlm@57 472 how well we understand what the problems are. I think there are many
rlm@57 473 people who try to formalize the problems of designing an intelligent
rlm@57 474 system in terms of streams of information thought of as bit streams or
rlm@57 475 collections of bit streams, and they think of as the problems of
rlm@57 476 intelligence as being the construction or detection of patterns in
rlm@57 477 those, and perhaps not just detection of patterns, but detection of
rlm@57 478 patterns that are useable for sending /out/ streams to control motors
rlm@57 479 and so on in order to []. And that way of conceptualizing the problem
rlm@57 480 may lead on the one hand to oversimplification, so that the things
rlm@57 481 that /would/ be achieved, if those goals were achieved, maybe much
rlm@57 482 simpler, in some ways inadequate. Or the replication of human
rlm@57 483 intelligence, or the matching of human intelligence---or for that
rlm@57 484 matter, squirrel intelligence---but in another way, it may also make
rlm@57 485 the problem harder: it may be that some of the kinds of things that
rlm@57 486 biological evolution has achieved can't be done that way. And one of
rlm@57 487 the ways that might turn out to be the case is not because it's not
rlm@57 488 impossible in principle to do some of the information processing on
rlm@57 489 artificial computers-based-on-transistors and other bit-manipulating
rlm@57 490 []---but it may just be that the computational complexity of solving
rlm@57 491 problems, processes, or finding solutions to complex problems, are
rlm@57 492 much greater and therefore you might need a much larger universe than
rlm@57 493 we have available in order to do things.
rlm@57 494
rlm@57 495 ** Example: find the shortest path by dangling strings
rlm@57 496 [28:55] Then if the underlying mechanisms were different, the
rlm@57 497 information processing mechanisms, they might be better tailored to
rlm@57 498 particular sorts of computation. There's a [] example, which is
rlm@57 499 finding the shortest route if you've got a collection of roads, and
rlm@57 500 they may be curved roads, and lots of tangled routes from A to B to C,
rlm@57 501 and so on. And if you start at A and you want to get to Z --- a place
rlm@57 502 somewhere on that map --- the process of finding the shortest route
rlm@57 503 will involve searching through all these different possibilities and
rlm@57 504 rejecting some that are longer than others and so on. But if you make
rlm@57 505 a model of that map out of string, where these strings are all laid
rlm@57 506 out on the maps and so have the lengths of the routes. Then if you
rlm@57 507 hold the two knots in the string -- it's a network of string --- which
rlm@57 508 correspond to the start point and end point, then /pull/, then the
rlm@57 509 bits of string that you're left with in a straight line will give you
rlm@57 510 the shortest route, and that process of pulling just gets you the
rlm@57 511 solution very rapidly in a parallel computation, where all the others
rlm@57 512 just hang by the wayside, so to speak.
rlm@57 513
rlm@57 514 ** In sum, we know surprisingly little about the kinds of problems that evolution solved, and the manner in which they were solved.
rlm@57 515 [30:15] Now, I'm not saying brains can build networks of string and
rlm@57 516 pull them or anything like that; that's just an illustration of how if
rlm@57 517 you have the right representation, correctly implemented---or suitably
rlm@57 518 implemented---for a problem, then you can avoid very combinatorially
rlm@57 519 complex searches, which will maybe grow exponentially with the number
rlm@57 520 of components in your map, whereas with this thing, the time it takes
rlm@57 521 won't depend on how many strings you've [got on the map]; you just
rlm@57 522 pull, and it will depend only on the shortest route that exists in
rlm@57 523 there. Even if that shortest route wasn't obvious on the original map.
rlm@57 524
rlm@57 525
rlm@57 526 [30:59] So that's a rather long-winded way of formulating the
rlm@57 527 conjecture which---of supporting, a roundabout way of supporting the
rlm@57 528 conjecture that there may be something about the way molecules perform
rlm@57 529 computations where they have the combination of continuous change as
rlm@57 530 things move through space and come together and move apart, and
rlm@57 531 whatever --- and also snap into states that then persist, so [as you
rlm@57 532 learn from] quantum mechanics, you can have stable molecular
rlm@57 533 structures which are quite hard to separate, and then in catalytic
rlm@57 534 processes you can separate them, or extreme temperatures, or strong
rlm@57 535 forces, but they may nevertheless be able to move very rapidly in some
rlm@57 536 conditions in order to perform computations.
rlm@57 537
rlm@57 538 [31:49] Now there may be things about that kind of structure that
rlm@57 539 enable searching for solutions to /certain/ classes of problems to be
rlm@57 540 done much more efficiently (by brain) than anything we could do with
rlm@57 541 computers. It's just an open question.
rlm@57 542
rlm@57 543 [32:04] So it /might/ turn out that we need new kinds of technology
rlm@57 544 that aren't on the horizon in order to replicate the functions that
rlm@57 545 animal brains perform ---or, it might not. I just don't know. I'm not
rlm@57 546 claiming that there's strong evidence for that; I'm just saying that
rlm@57 547 it might turn out that way, partly because I think we know less than
rlm@57 548 many people think we know about what biological evolution achieved.
rlm@57 549
rlm@57 550 [32:28] There are some other possibilities: we may just find out that
rlm@57 551 there are shortcuts no one ever thought of, and it will all happen
rlm@57 552 much more quickly---I have an open mind; I'd be surprised, but it
rlm@57 553 could turn up. There /is/ something that worries me much more than the
rlm@57 554 singularity that most people talk about, which is machines achieving
rlm@57 555 human-level intelligence and perhaps taking over [the] planet or
rlm@57 556 something. There's what I call the /singularity of cognitive
rlm@57 557 catch-up/ ...
rlm@57 558
rlm@57 559 * A singularity of cognitive catch-up
rlm@57 560
rlm@57 561 ** What if it will take a lifetime to learn enough to make something new?
rlm@57 562 ... SCC, singularity of cognitive catch-up, which I think we're close
rlm@57 563 to, or maybe have already reached---I'll explain what I mean by
rlm@57 564 that. One of the products of biological evolution---and this is one of
rlm@57 565 the answers to your earlier questions which I didn't get on to---is
rlm@57 566 that humans have not only the ability to make discoveries that none of
rlm@57 567 their ancestors have ever made, but to shorten the time required for
rlm@57 568 similar achievements to be reached by their offspring and their
rlm@57 569 descendants. So once we, for instance, worked out ways of complex
rlm@57 570 computations, or ways of building houses, or ways of finding our way
rlm@57 571 around, we don't need...our children don't need to work it out for
rlm@57 572 themselves by the same lengthy trial and error procedure; we can help
rlm@57 573 them get there much faster.
rlm@57 574
rlm@57 575 Okay, well, what I've been referring to as the singularity of
rlm@57 576 cognitive catch-up depends on the fact that---fairly obvious, and it's
rlm@57 577 often been commented on---that in case of humans, it's not necessary
rlm@57 578 for each generation to learn what previous generations learned /in the
rlm@57 579 same way/. And we can speed up learning once something has been
rlm@57 580 learned, [it is able to] be learned by new people. And that has meant
rlm@57 581 that the social processes that support that kind of education of the
rlm@57 582 young can enormously accelerate what would have taken...perhaps
rlm@57 583 thousands [or] millions of years for evolution to produce, can happen in
rlm@57 584 a much shorter time.
rlm@57 585
rlm@57 586
rlm@57 587 [34:54] But here's the catch: in order for a new advance to happen ---
rlm@57 588 so for something new to be discovered that wasn't there before, like
rlm@57 589 Newtonian mechanics, or the theory of relativity, or Beethoven's music
rlm@57 590 or [style] or whatever --- the individuals have to have traversed a
rlm@57 591 significant amount of what their ancestors have learned, even if they
rlm@57 592 do it much faster than their ancestors, to get to the point where they
rlm@57 593 can see the gaps, the possibilities for going further than their
rlm@57 594 ancestors, or their parents or whatever, have done.
rlm@57 595
rlm@57 596 [35:27] Now in the case of knowledge of science, mathematics,
rlm@57 597 philosophy, engineering and so on, there's been a lot of accumulated
rlm@57 598 knowledge. And humans are living a /bit/ longer than they used to, but
rlm@57 599 they're still living for [whatever it is], a hundred years, or for
rlm@57 600 most people, less than that. So you can imagine that there might come
rlm@57 601 a time when in a normal human lifespan, it's not possible for anyone
rlm@57 602 to learn enough to understand the scope and limits of what's already
rlm@57 603 been achieved in order to see the potential for going beyond it and to
rlm@57 604 build on what's already been done to make that...those future steps.
rlm@57 605
rlm@57 606 [36:10] So if we reach that stage, we will have reached the
rlm@57 607 singularity of cognitive catch-up because the process of education
rlm@57 608 that enables individuals to learn faster than their ancestors did is
rlm@57 609 the catching-up process, and it may just be that we at some point
rlm@57 610 reach a point where catching up can only happen within a lifetime of
rlm@57 611 an individual, and after that they're dead and they can't go
rlm@57 612 beyond. And I have some evidence that there's a lot of that around
rlm@57 613 because I see a lot of people coming up with what /they/ think of as
rlm@57 614 new ideas which they've struggled to come up with, but actually they
rlm@57 615 just haven't taken in some of what was...some of what was done [] by
rlm@57 616 other people, in other places before them. And I think that despite
rlm@57 617 the availability of search engines which make it /easier/ for people
rlm@57 618 to get the information---for instance, when I was a student, if I
rlm@57 619 wanted to find out what other people had done in the field, it was a
rlm@57 620 laborious process---going to the library, getting books, and
rlm@57 621 ---whereas now, I can often do things in seconds that would have taken
rlm@57 622 hours. So that means that if seconds [are needed] for that kind of
rlm@57 623 work, my lifespan has been extended by a factor of ten or
rlm@57 624 something. So maybe that /delays/ the singularity, but it may not
rlm@57 625 delay it enough. But that's an open question; I don't know. And it may
rlm@57 626 just be that in some areas, this is more of a problem than others. For
rlm@57 627 instance, it may be that in some kinds of engineering, we're handing
rlm@57 628 over more and more of the work to machines anyways and they can go on
rlm@57 629 doing it. So for instance, most of the production of computers now is
rlm@57 630 done by a computer-controlled machine---although some of the design
rlm@57 631 work is done by humans--- a lot of /detail/ of the design is done by
rlm@57 632 computers, and they produce the next generation, which then produces
rlm@57 633 the next generation, and so on.
rlm@57 634
rlm@57 635 [37:57] I don't know if humans can go on having major advances, so
rlm@57 636 it'll be kind of sad if we can't.
rlm@57 637
rlm@57 638 * Spatial reasoning: a difficult problem
rlm@57 639
rlm@57 640 [38:15] Okay, well, there are different problems [ ] mathematics, and
rlm@57 641 they have to do with properties. So for instance a lot of mathematics
rlm@57 642 that can be expressed in terms of logical structures or algebraic
rlm@57 643 structures and those are pretty well suited for manipulation and...on
rlm@57 644 computers, and if a problem can be specified using the
rlm@57 645 logical/algebraic notation, and the solution method requires creating
rlm@57 646 something in that sort of notation, then computers are pretty good,
rlm@57 647 and there are lots of mathematical tools around---there are theorem
rlm@57 648 provers and theorem checkers, and all kinds of things, which couldn't
rlm@57 649 have existed fifty, sixty years ago, and they will continue getting
rlm@57 650 better.
rlm@57 651
rlm@57 652
rlm@57 653 But there was something that I was [[example-gap][alluding to earlier]] when I gave the
rlm@57 654 example of how you can reason about what you will see by changing your
rlm@57 655 position in relation to a door, where what you are doing is using your
rlm@57 656 grasp of spatial structures and how as one spatial relationship
rlm@57 657 changes namely you come closer to the door or move sideways and
rlm@57 658 parallel to the wall or whatever, other spatial relationships change
rlm@57 659 in parallel, so the lines from your eyes through to other parts of
rlm@57 660 the...parts of the room on the other side of the doorway change,
rlm@57 661 spread out more as you go towards the doorway, and as you move
rlm@57 662 sideways, they don't spread out differently, but focus on different
rlm@57 663 parts of the internal ... that they access different parts of the
rlm@57 664 ... of the room.
rlm@57 665
rlm@57 666 Now, those are examples of ways of thinking about relationships and
rlm@57 667 changing relationships which are not the same as thinking about what
rlm@57 668 happens if I replace this symbol with that symbol, or if I substitute
rlm@57 669 this expression in that expression in a logical formula. And at the
rlm@57 670 moment, I do not believe that there is anything in AI amongst the
rlm@57 671 mathematical reasoning community, the theorem-proving community, that
rlm@57 672 can model the processes that go on when a young child starts learning
rlm@57 673 to do Euclidean geometry and is taught things about---for instance, I
rlm@57 674 can give you a proof that the angles of any triangle add up to a
rlm@57 675 straight line, 180 degrees.
rlm@57 676
rlm@57 677 ** Example: Spatial proof that the angles of any triangle add up to a half-circle
rlm@57 678 There are standard proofs which involves starting with one triangle,
rlm@57 679 then adding a line parallel to the base one of my former students,
rlm@57 680 Mary Pardoe, came up with which I will demonstrate with this <he holds
rlm@57 681 up a pen> --- can you see it? If I have a triangle here that's got
rlm@57 682 three sides, if I put this thing on it, on one side --- let's say the
rlm@57 683 bottom---I can rotate it until it lies along the second...another
rlm@57 684 side, and then maybe move it up to the other end ~. Then I can rotate
rlm@57 685 it again, until it lies on the third side, and move it back to the
rlm@57 686 other end. And then I'll rotate it again and it'll eventually end up
rlm@57 687 on the original side, but it will have changed the direction it's
rlm@57 688 pointing in --- and it won't have crossed over itself so it will have
rlm@57 689 gone through a half-circle, and that says that the three angles of a
rlm@57 690 triangle add up to the rotations of half a circle, which is a
rlm@57 691 beautiful kind of proof and almost anyone can understand it. Some
rlm@57 692 mathematicians don't like it, because they say it hides some of the
rlm@57 693 assumptions, but nevertheless, as far as I'm concerned, it's an
rlm@57 694 example of a human ability to do reasoning which, once you've
rlm@57 695 understood it, you can see will apply to any triangle --- it's got to
rlm@57 696 be a planar triangle --- not a triangle on a globe, because then the
rlm@57 697 angles can add up to more than ... you can have three /right/ angles
rlm@57 698 if you have an equator...a line on the equator, and a line going up to
rlm@57 699 to the north pole of the earth, and then you have a right angle and
rlm@57 700 then another line going down to the equator, and you have a right
rlm@57 701 angle, right angle, right angle, and they add up to more than a
rlm@57 702 straight line. But that's because the triangle isn't in the plane,
rlm@57 703 it's on a curved surface. In fact, that's one of the
rlm@57 704 differences...definitional differences you can take between planar and
rlm@57 705 curved surfaces: how much the angles of a triangle add up to. But our
rlm@57 706 ability to /visualize/ and notice the generality in that process, and
rlm@57 707 see that you're going to be able to do the same thing using triangles
rlm@57 708 that stretch in all sorts of ways, or if it's a million times as
rlm@57 709 large, or if it's made...you know, written on, on...if it's drawn in
rlm@57 710 different colors or whatever --- none of that's going to make any
rlm@57 711 difference to the essence of that process. And that ability to see
rlm@57 712 the commonality in a spatial structure which enables you to draw some
rlm@57 713 conclusions with complete certainty---subject to the possibility that
rlm@57 714 sometimes you make mistakes, but when you make mistakes, you can
rlm@57 715 discover them, as has happened in the history of geometrical theorem
rlm@57 716 proving. Imre Lakatos had a wonderful book called [[http://en.wikipedia.org/wiki/Proofs_and_Refutations][/Proofs and
rlm@57 717 Refutations/]] --- which I won't try to summarize --- but he has
rlm@57 718 examples: mistakes were made; that was because people didn't always
rlm@57 719 realize there were subtle subcases which had slightly different
rlm@57 720 properties, and they didn't take account of that. But once they're
rlm@57 721 noticed, you rectify that.
rlm@57 722
rlm@57 723 ** Geometric results are fundamentally different than experimental results in chemistry or physics.
rlm@57 724 [43:28] But it's not the same as doing experiments in chemistry and
rlm@57 725 physics, where you can't be sure it'll be the same on [] or at a high
rlm@57 726 temperature, or in a very strong magnetic field --- with geometric
rlm@57 727 reasoning, in some sense you've got the full information in front of
rlm@57 728 you; even if you don't always notice an important part of it. So, that
rlm@57 729 kind of reasoning (as far as I know) is not implemented anywhere in a
rlm@57 730 computer. And most people who do research on trying to model
rlm@57 731 mathematical reasoning, don't pay any attention to that, because of
rlm@57 732 ... they just don't think about it. They start from somewhere else,
rlm@57 733 maybe because of how they were educated. I was taught Euclidean
rlm@57 734 geometry at school. Were you?
rlm@57 735
rlm@57 736 (Adam ford: Yeah)
rlm@57 737
rlm@57 738 Many people are not now. Instead they're taught set theory, and
rlm@57 739 logic, and arithmetic, and [algebra], and so on. And so they don't use
rlm@57 740 that bit of their brains, without which we wouldn't have built any of
rlm@57 741 the cathedrals, and all sorts of things we now depend on.
rlm@57 742
rlm@57 743 * Is near-term artificial general intelligence likely?
rlm@57 744
rlm@57 745 ** Two interpretations: a single mechanism for all problems, or many mechanisms unified in one program.
rlm@57 746
rlm@57 747 [44:35] Well, this relates to what's meant by general. And when I
rlm@57 748 first encountered the AGI community, I thought that what they all
rlm@57 749 meant by general intelligence was /uniform/ intelligence ---
rlm@57 750 intelligence based on some common simple (maybe not so simple, but)
rlm@57 751 single powerful mechanism or principle of inference. And there are
rlm@57 752 some people in the community who are trying to produce things like
rlm@57 753 that, often in connection with algorithmic information theory and
rlm@57 754 computability of information, and so on. But there's another sense of
rlm@57 755 general which means that the system of general intelligence can do
rlm@57 756 lots of different things, like perceive things, understand language,
rlm@57 757 move around, make things, and so on --- perhaps even enjoy a joke;
rlm@57 758 that's something that's not nearly on the horizon, as far as I
rlm@57 759 know. Enjoying a joke isn't the same as being able to make laughing
rlm@57 760 noises.
rlm@57 761
rlm@57 762 Given, then, that there are these two notions of general
rlm@57 763 intelligence---there's one that looks for one uniform, possibly
rlm@57 764 simple, mechanism or collection of ideas and notations and algorithms,
rlm@57 765 that will deal with any problem that's solvable --- and the other
rlm@57 766 that's general in the sense that it can do lots of different things
rlm@57 767 that are combined into an integrated architecture (which raises lots
rlm@57 768 of questions about how you combine these things and make them work
rlm@57 769 together) and we humans, certainly, are of the second kind: we do all
rlm@57 770 sorts of different things, and other animals also seem to be of the
rlm@57 771 second kind, perhaps not as general as humans. Now, it may turn out
rlm@57 772 that in some near future time, who knows---decades, a few
rlm@57 773 decades---you'll be able to get machines that are capable of solving
rlm@57 774 in a time that will depend on the nature of the problem, but any
rlm@57 775 problem that is solvable, and they will be able to do it in some sort
rlm@57 776 of tractable time --- of course, there are some problems that are
rlm@57 777 solvable that would require a larger universe and a longer history
rlm@57 778 than the history of the universe, but apart from that constraint,
rlm@57 779 these machines will be able to do anything []. But to be able to do
rlm@57 780 some of the kinds of things that humans can do, like the kinds of
rlm@57 781 geometrical reasoning where you look at the shape and you abstract
rlm@57 782 away from the precise angles and sizes and shapes and so on, and
rlm@57 783 realize there's something general here, as must have happened when our
rlm@57 784 ancestors first made the discoveries that eventually put together in
rlm@57 785 Euclidean geometry.
rlm@57 786
rlm@57 787 It may be that that requires mechanisms of a kind that we don't know
rlm@57 788 anything about at the moment. Maybe brains are using molecules and
rlm@57 789 rearranging molecules in some way that supports that kind of
rlm@57 790 reasoning. I'm not saying they are --- I don't know, I just don't see
rlm@57 791 any simple...any obvious way to map that kind of reasoning capability
rlm@57 792 onto what we currently do on computers. There is---and I just
rlm@57 793 mentioned this briefly beforehand---there is a kind of thing that's
rlm@57 794 sometimes thought of as a major step in that direction, namely you can
rlm@57 795 build a machine (or a software system) that can represent some
rlm@57 796 geometrical structure, and then be told about some change that's going
rlm@57 797 to happen to it, and it can predict in great detail what'll
rlm@57 798 happen. And this happens for instance in game engines, where you say
rlm@57 799 we have all these blocks on the table and I'll drop one other block,
rlm@57 800 and then [the thing] uses Newton's laws and properties of rigidity of
rlm@57 801 the parts and the elasticity and also stuff about geometries and space
rlm@57 802 and so on, to give you a very accurate representation of what'll
rlm@57 803 happen when this brick lands on this pile of things, [it'll bounce and
rlm@57 804 go off, and so on]. And you just, with more memory and more CPU power,
rlm@57 805 you can increase the accuracy--- but that's totally different than
rlm@57 806 looking at /one/ example, and working out what will happen in a whole
rlm@57 807 /range/ of cases at a higher level of abstraction, whereas the game
rlm@57 808 engine does it in great detail for /just/ this case, with /just/ those
rlm@57 809 precise things, and it won't even know what the generalizations are
rlm@57 810 that it's using that would apply to others []. So, in that sense, [we]
rlm@57 811 may get AGI --- artificial general intelligence --- pretty soon, but
rlm@57 812 it'll be limited in what it can do. And the other kind of general
rlm@57 813 intelligence which combines all sorts of different things, including
rlm@57 814 human spatial geometrical reasoning, and maybe other things, like the
rlm@57 815 ability to find things funny, and to appreciate artistic features and
rlm@57 816 other things may need forms of pattern-mechanism, and I have an open
rlm@57 817 mind about that.
rlm@57 818
rlm@57 819 * Abstract General Intelligence impacts
rlm@57 820
rlm@57 821 [49:53] Well, as far as the first type's concerned, it could be useful
rlm@57 822 for all kinds of applications --- there are people who worry about
rlm@57 823 where there's a system that has that type of intelligence, might in
rlm@57 824 some sense take over control of the planet. Well, humans often do
rlm@57 825 stupid things, and they might do something stupid that would lead to
rlm@57 826 disaster, but I think it's more likely that there would be other
rlm@57 827 things [] lead to disaster--- population problems, using up all the
rlm@57 828 resources, destroying ecosystems, and whatever. But certainly it would
rlm@57 829 go on being useful to have these calculating devices. Now, as for the
rlm@57 830 second kind of them, I don't know---if we succeeded at putting
rlm@57 831 together all the parts that we find in humans, we might just make an
rlm@57 832 artificial human, and then we might have some of them as your friends,
rlm@57 833 and some of them we might not like, and some of them might become
rlm@57 834 teachers or whatever, composers --- but that raises a question: could
rlm@57 835 they, in some sense, be superior to us, in their learning
rlm@57 836 capabilities, their understanding of human nature, or maybe their
rlm@57 837 wickedness or whatever --- these are all issues in which I expect the
rlm@57 838 best science fiction writers would give better answers than anything I
rlm@57 839 could do, but I did once fantasize when I [back] in 1978, that perhaps
rlm@57 840 if we achieved that kind of thing, that they would be wise, and gentle
rlm@57 841 and kind, and realize that humans are an inferior species that, you
rlm@57 842 know, have some good features, so they'd keep us in some kind of
rlm@57 843 secluded...restrictive kind of environment, keep us away from
rlm@57 844 dangerous weapons, and so on. And find ways of cohabitating with
rlm@57 845 us. But that's just fantasy.
rlm@57 846
rlm@57 847 Adam Ford: Awesome. Yeah, there's an interesting story /With Folded
rlm@57 848 Hands/ where [the computers] want to take care of us and want to
rlm@57 849 reduce suffering and end up lobotomizing everybody [but] keeping them
rlm@57 850 alive so as to reduce the suffering.
rlm@57 851
rlm@57 852 Aaron Sloman: Not all that different from /Brave New World/, where it
rlm@57 853 was done with drugs and so on, but different humans are given
rlm@57 854 different roles in that system, yeah.
rlm@57 855
rlm@57 856 There's also /The Time Machine/, H.G. Wells, where the ... in the
rlm@57 857 distant future, humans have split in two: the Eloi, I think they were
rlm@57 858 called, they lived underground, they were the [] ones, and then---no,
rlm@57 859 the Morlocks lived underground; Eloi lived on the planet; they were
rlm@57 860 pleasant and pretty but not very bright, and so on, and they were fed
rlm@57 861 on by ...
rlm@57 862
rlm@57 863 Adam Ford: [] in the future.
rlm@57 864
rlm@57 865 Aaron Sloman: As I was saying, if you ask science fiction writers,
rlm@57 866 you'll probably come up with a wide variety of interesting answers.
rlm@57 867
rlm@57 868 Adam Ford: I certainly have; I've spoken to [] of Birmingham, and
rlm@57 869 Sean Williams, ... who else?
rlm@57 870
rlm@57 871 Aaron Sloman: Did you ever read a story by E.M. Forrester called /The
rlm@57 872 Machine Stops/ --- very short story, it's [[http://archive.ncsa.illinois.edu/prajlich/forster.html][on the Internet somewhere]]
rlm@57 873 --- it's about a time when people sitting ... and this was written in
rlm@57 874 about [1914 ] so it's about...over a hundred years ago ... people are
rlm@57 875 in their rooms, they sit in front of screens, and they type things,
rlm@57 876 and they communicate with one another that way, and they don't meet;
rlm@57 877 they have debates, and they give lectures to their audiences that way,
rlm@57 878 and then there's a woman whose son says \ldquo{}I'd like to see
rlm@57 879 you\rdquo{} and she says \ldquo{}What's the point? You've got me at
rlm@57 880 this point \rdquo{} but he wants to come and talk to her --- I won't
rlm@57 881 tell you how it ends, but.
rlm@57 882
rlm@57 883 Adam Ford: Reminds me of the Internet.
rlm@57 884
rlm@57 885 Aaron Sloman: Well, yes; he invented ... it was just extraordinary
rlm@57 886 that he was able to do that, before most of the components that we
rlm@57 887 need for it existed.
rlm@57 888
rlm@57 889 Adam Ford: [Another person who did that] was Vernor Vinge [] /True
rlm@57 890 Names/.
rlm@57 891
rlm@57 892 Aaron Sloman: When was that written?
rlm@57 893
rlm@57 894 Adam Ford: The seventies.
rlm@57 895
rlm@57 896 Aaron Sloman: Okay, well a lot of the technology was already around
rlm@57 897 then. The original bits of internet were working, in about 1973, I was
rlm@57 898 sitting ... 1974, I was sitting at Sussex University trying to
rlm@57 899 use...learn LOGO, the programming language, to decide whether it was
rlm@57 900 going to be useful for teaching AI, and I was sitting [] paper
rlm@57 901 teletype, there was paper coming out, transmitting ten characters a
rlm@57 902 second from Sussex to UCL computer lab by telegraph cable, from there
rlm@57 903 to somewhere in Norway via another cable, from there by satellite to
rlm@57 904 California to a computer Xerox [] research center where they had
rlm@57 905 implemented a computer with a LOGO system on it, with someone I had
rlm@57 906 met previously in Edinburgh, Danny Bobrow, and he allowed me to have
rlm@57 907 access to this sytem. So there I was typing. And furthermore, it was
rlm@57 908 duplex typing, so every character I typed didn't show up on my
rlm@57 909 terminal until it had gone all the way there and echoed back, so I
rlm@57 910 would type, and the characters would come back four seconds later.
rlm@57 911
rlm@57 912 [55:26] But that was the Internet, and I think Vernor Vinge was
rlm@57 913 writing after that kind of thing had already started, but I don't
rlm@57 914 know. Anyway.
rlm@57 915
rlm@57 916 [55:41] Another...I mentioned H.G. Wells, /The Time Machine/. I
rlm@57 917 recently discovered, because [[http://en.wikipedia.org/wiki/David_Lodge_(author)][David Lodge]] had written a sort of
rlm@57 918 semi-novel about him, that he had invented Wikipedia, in advance--- he
rlm@57 919 had this notion of an encyclopedia that was free to everybody, and
rlm@57 920 everybody could contribute and [collaborate on it]. So, go to the
rlm@57 921 science fiction writers to find out the future --- well, a range of
rlm@57 922 possible futures.
rlm@57 923
rlm@57 924 Adam Ford: Well the thing is with science fiction writers, they have
rlm@57 925 to maintain some sort of interest for their readers, after all the
rlm@57 926 science fiction which reaches us is the stuff that publishers want to
rlm@57 927 sell, and so there's a little bit of a ... a bias towards making a
rlm@57 928 plot device there, and so the dramatic sort of appeals to our
rlm@57 929 amygdala, our lizard brain; we'll sort of stay there obviously to some
rlm@57 930 extent. But I think that they do come up with sort of amazing ideas; I
rlm@57 931 think it's worth trying to make these predictions; I think that we
rlm@57 932 should more time on strategic forecasting, I mean take that seriously.
rlm@57 933
rlm@57 934 Aaron Sloman: Well, I'm happy to leave that to others; I just want to
rlm@57 935 try to understand these problems that bother me about how things
rlm@57 936 work. And it may be that some would say that's irresponsible if I
rlm@57 937 don't think about what the implications will be. Well, understanding
rlm@57 938 how humans work /might/ enable us to make [] humans --- I suspect it
rlm@57 939 wont happen in this century; I think it's going to be too difficult.