Mercurial > thoughts
diff org/sloman-old.html @ 109:414a10d51d9f
stuff from dylan?
| author | Robert McIntyre <rlm@mit.edu> |
|---|---|
| date | Tue, 03 Jun 2014 13:24:58 -0400 |
| parents | |
| children |
line wrap: on
line diff
1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/org/sloman-old.html Tue Jun 03 13:24:58 2014 -0400 1.3 @@ -0,0 +1,1348 @@ 1.4 +<?xml version="1.0" encoding="utf-8"?> 1.5 +<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" 1.6 + "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"> 1.7 +<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"> 1.8 +<head> 1.9 +<title>Transcript of Aaron Sloman - Artificial Intelligence - Psychology - Oxford Interview</title> 1.10 +<meta http-equiv="Content-Type" content="text/html;charset=utf-8"/> 1.11 +<meta name="title" content="Transcript of Aaron Sloman - Artificial Intelligence - Psychology - Oxford Interview"/> 1.12 +<meta name="generator" content="Org-mode"/> 1.13 +<meta name="generated" content="2013-10-04 18:49:53 UTC"/> 1.14 +<meta name="author" content="Dylan Holmes"/> 1.15 +<meta name="description" content=""/> 1.16 +<meta name="keywords" content=""/> 1.17 +<style type="text/css"> 1.18 + <!--/*--><![CDATA[/*><!--*/ 1.19 + html { font-family: Times, serif; font-size: 12pt; } 1.20 + .title { text-align: center; } 1.21 + .todo { color: red; } 1.22 + .done { color: green; } 1.23 + .tag { background-color: #add8e6; font-weight:normal } 1.24 + .target { } 1.25 + .timestamp { color: #bebebe; } 1.26 + .timestamp-kwd { color: #5f9ea0; } 1.27 + .right {margin-left:auto; margin-right:0px; text-align:right;} 1.28 + .left {margin-left:0px; margin-right:auto; text-align:left;} 1.29 + .center {margin-left:auto; margin-right:auto; text-align:center;} 1.30 + p.verse { margin-left: 3% } 1.31 + pre { 1.32 + border: 1pt solid #AEBDCC; 1.33 + background-color: #F3F5F7; 1.34 + padding: 5pt; 1.35 + font-family: courier, monospace; 1.36 + font-size: 90%; 1.37 + overflow:auto; 1.38 + } 1.39 + table { border-collapse: collapse; } 1.40 + td, th { vertical-align: top; } 1.41 + th.right { text-align:center; } 1.42 + th.left { text-align:center; } 1.43 + th.center { text-align:center; } 1.44 + td.right { text-align:right; } 1.45 + td.left { text-align:left; } 1.46 + td.center { text-align:center; } 1.47 + dt { font-weight: bold; } 1.48 + div.figure { padding: 0.5em; } 1.49 + div.figure p { text-align: center; } 1.50 + div.inlinetask { 1.51 + padding:10px; 1.52 + border:2px solid gray; 1.53 + margin:10px; 1.54 + background: #ffffcc; 1.55 + } 1.56 + textarea { overflow-x: auto; } 1.57 + .linenr { font-size:smaller } 1.58 + .code-highlighted {background-color:#ffff00;} 1.59 + .org-info-js_info-navigation { border-style:none; } 1.60 + #org-info-js_console-label { font-size:10px; font-weight:bold; 1.61 + white-space:nowrap; } 1.62 + .org-info-js_search-highlight {background-color:#ffff00; color:#000000; 1.63 + font-weight:bold; } 1.64 + /*]]>*/--> 1.65 +</style> 1.66 +<link rel="stylesheet" type="text/css" href="../css/sloman.css" /> 1.67 +<script type="text/javascript"> 1.68 +<!--/*--><![CDATA[/*><!--*/ 1.69 + function CodeHighlightOn(elem, id) 1.70 + { 1.71 + var target = document.getElementById(id); 1.72 + if(null != target) { 1.73 + elem.cacheClassElem = elem.className; 1.74 + elem.cacheClassTarget = target.className; 1.75 + target.className = "code-highlighted"; 1.76 + elem.className = "code-highlighted"; 1.77 + } 1.78 + } 1.79 + function CodeHighlightOff(elem, id) 1.80 + { 1.81 + var target = document.getElementById(id); 1.82 + if(elem.cacheClassElem) 1.83 + elem.className = elem.cacheClassElem; 1.84 + if(elem.cacheClassTarget) 1.85 + target.className = elem.cacheClassTarget; 1.86 + } 1.87 +/*]]>*///--> 1.88 +</script> 1.89 + 1.90 +</head> 1.91 +<body> 1.92 + 1.93 + 1.94 +<div id="content"> 1.95 +<h1 class="title">Transcript of Aaron Sloman - Artificial Intelligence - Psychology - Oxford Interview</h1> 1.96 + 1.97 + 1.98 +<blockquote> 1.99 + 1.100 + 1.101 + 1.102 + 1.103 + 1.104 + 1.105 + 1.106 + 1.107 + 1.108 + 1.109 + 1.110 + 1.111 + 1.112 + 1.113 + 1.114 + 1.115 +<p> 1.116 +<b>Editor's note:</b> This is a working draft transcript which I made of 1.117 +<a href="http://www.youtube.com/watch?feature=player_detailpage&v=iuH8dC7Snno">this nice interview</a> of Aaron Sloman. Having just finished one 1.118 +iteration of transcription, I still need to go in and clean up the 1.119 +formatting and fix the parts that I misheard, so you can expect the 1.120 +text to improve significantly in the near future. 1.121 +</p> 1.122 +<p> 1.123 +To the extent that this is my work, you have my permission to make 1.124 +copies of this transcript for your own purposes. Also, feel free to 1.125 +e-mail me with comments or corrections. 1.126 +</p> 1.127 +<p> 1.128 +You can send mail to <code>transcript@aurellem.org</code>. 1.129 +</p> 1.130 +<p> 1.131 +Cheers, 1.132 +</p> 1.133 +<p> 1.134 +—Dylan 1.135 +</p> 1.136 +</blockquote> 1.137 + 1.138 + 1.139 + 1.140 + 1.141 + 1.142 +<div id="table-of-contents"> 1.143 +<h2>Table of Contents</h2> 1.144 +<div id="text-table-of-contents"> 1.145 +<ul> 1.146 +<li><a href="#sec-1">1 Introduction</a> 1.147 +<ul> 1.148 +<li><a href="#sec-1-1">1.1 Aaron Sloman evolves into a philosopher of AI</a></li> 1.149 +<li><a href="#sec-1-2">1.2 AI is hard, in part because there are tempting non-problems.</a></li> 1.150 +</ul> 1.151 +</li> 1.152 +<li><a href="#sec-2">2 What problems of intelligence did evolution solve?</a> 1.153 +<ul> 1.154 +<li><a href="#sec-2-1">2.1 Intelligence consists of solutions to many evolutionary problems; no single development (e.g. communication) was key to human-level intelligence.</a></li> 1.155 +<li><a href="#sec-2-2">2.2 Speculation about how communication might have evolved from internal lanagues.</a></li> 1.156 +</ul> 1.157 +</li> 1.158 +<li><a href="#sec-3">3 How do language and internal states relate to AI?</a> 1.159 +<ul> 1.160 +<li><a href="#sec-3-1">3.1 In AI, false assumptions can lead investigators astray.</a></li> 1.161 +<li><a href="#sec-3-2">3.2 Example: Vision is not just about finding surfaces, but about finding affordances.</a></li> 1.162 +<li><a href="#sec-3-3">3.3 Online and offline intelligence</a></li> 1.163 +<li><a href="#sec-3-4">3.4 Example: Even toddlers use sophisticated geometric knowledge</a></li> 1.164 +</ul> 1.165 +</li> 1.166 +<li><a href="#sec-4">4 Animal intelligence</a> 1.167 +<ul> 1.168 +<li><a href="#sec-4-1">4.1 The priority is <i>cataloguing</i> what competences have evolved, not ranking them.</a></li> 1.169 +<li><a href="#sec-4-2">4.2 AI can be used to test philosophical theories</a></li> 1.170 +</ul> 1.171 +</li> 1.172 +<li><a href="#sec-5">5 Is abstract general intelligence feasible?</a> 1.173 +<ul> 1.174 +<li><a href="#sec-5-1">5.1 It's misleading to compare the brain and its neurons to a computer made of transistors</a></li> 1.175 +<li><a href="#sec-5-2">5.2 For example, brains may rely heavily on chemical information processing</a></li> 1.176 +<li><a href="#sec-5-3">5.3 Brain algorithms may simply be optimized for certain kinds of information processing other than bit manipulations</a></li> 1.177 +<li><a href="#sec-5-4">5.4 Example: find the shortest path by dangling strings</a></li> 1.178 +<li><a href="#sec-5-5">5.5 In sum, we know surprisingly little about the kinds of problems that evolution solved, and the manner in which they were solved.</a></li> 1.179 +</ul> 1.180 +</li> 1.181 +<li><a href="#sec-6">6 A singularity of cognitive catch-up</a> 1.182 +<ul> 1.183 +<li><a href="#sec-6-1">6.1 What if it will take a lifetime to learn enough to make something new?</a></li> 1.184 +</ul> 1.185 +</li> 1.186 +<li><a href="#sec-7">7 Spatial reasoning: a difficult problem</a> 1.187 +<ul> 1.188 +<li><a href="#sec-7-1">7.1 Example: Spatial proof that the angles of any triangle add up to a half-circle</a></li> 1.189 +<li><a href="#sec-7-2">7.2 Geometric results are fundamentally different than experimental results in chemistry or physics.</a></li> 1.190 +</ul> 1.191 +</li> 1.192 +<li><a href="#sec-8">8 Is near-term artificial general intelligence likely?</a> 1.193 +<ul> 1.194 +<li><a href="#sec-8-1">8.1 Two interpretations: a single mechanism for all problems, or many mechanisms unified in one program.</a></li> 1.195 +</ul> 1.196 +</li> 1.197 +<li><a href="#sec-9">9 Abstract General Intelligence impacts</a></li> 1.198 +</ul> 1.199 +</div> 1.200 +</div> 1.201 + 1.202 +<div id="outline-container-1" class="outline-2"> 1.203 +<h2 id="sec-1"><span class="section-number-2">1</span> Introduction</h2> 1.204 +<div class="outline-text-2" id="text-1"> 1.205 + 1.206 + 1.207 + 1.208 +</div> 1.209 + 1.210 +<div id="outline-container-1-1" class="outline-3"> 1.211 +<h3 id="sec-1-1"><span class="section-number-3">1.1</span> Aaron Sloman evolves into a philosopher of AI</h3> 1.212 +<div class="outline-text-3" id="text-1-1"> 1.213 + 1.214 +<p>[0:09] My name is Aaron Sloman. My first degree many years ago in 1.215 +Capetown University was in Physics and Mathematics, and I intended to 1.216 +go and be a mathematician. I came to Oxford and encountered 1.217 +philosophers — I had started reading philosophy and discussing 1.218 +philosophy before then, and then I found that there were philosophers 1.219 +who said things about mathematics that I thought were wrong, so 1.220 +gradually got more and more involved in [philosophy] discussions and 1.221 +switched to doing philosophy DPhil. Then I became a philosophy 1.222 +lecturer and about six years later, I was introduced to artificial 1.223 +intelligence when I was a lecturer at Sussex University in philosophy 1.224 +and I very soon became convinced that the best way to make progress in 1.225 +both areas of philosophy (including philosophy of mathematics which I 1.226 +felt i hadn't dealt with adequately in my DPhil) about the philosophy 1.227 +of mathematics, philosophy of mind, philsophy of language and all 1.228 +those things—the best way was to try to design and test working 1.229 +fragments of mind and maybe eventually put them all together but 1.230 +initially just working fragments that would do various things. 1.231 +</p> 1.232 +<p> 1.233 +[1:12] And I learned to program and ~ with various other people 1.234 +including ~Margaret Boden whom you've interviewed, developed—helped 1.235 +develop an undergraduate degree in AI and other things and also began 1.236 +to do research in AI and so on which I thought of as doing philosophy, 1.237 +primarily. 1.238 +</p> 1.239 +<p> 1.240 +[1:29] And then I later moved to the University of Birmingham and I 1.241 +was there — I came in 1991 — and I've been retired for a while but 1.242 +I'm not interested in golf or gardening so I just go on doing full 1.243 +time research and my department is happy to keep me on without paying 1.244 +me and provide space and resources and I come, meeting bright people 1.245 +at conferences and try to learn and make progress if I can. 1.246 +</p> 1.247 +</div> 1.248 + 1.249 +</div> 1.250 + 1.251 +<div id="outline-container-1-2" class="outline-3"> 1.252 +<h3 id="sec-1-2"><span class="section-number-3">1.2</span> AI is hard, in part because there are tempting non-problems.</h3> 1.253 +<div class="outline-text-3" id="text-1-2"> 1.254 + 1.255 + 1.256 +<p> 1.257 +One of the things I learnt and understood more and more over the many 1.258 +years — forty years or so since I first encountered AI — is how 1.259 +hard the problems are, and in part that's because it's very often 1.260 +tempting to <i>think</i> the problem is something different from what it 1.261 +actually is, and then people design solutions to the non-problems, and 1.262 +I think of most of my work now as just helping to clarify what the 1.263 +problems are: what is it that we're trying to explain — and maybe 1.264 +this is leading into what you wanted to talk about: 1.265 +</p> 1.266 +<p> 1.267 +I now think that one of the ways of getting a deep understanding of 1.268 +that is to find out what were the problems that biological evolution 1.269 +solved, because we are a product of <i>many</i> solutions to <i>many</i> 1.270 +problems, and if we just try to go in and work out what the whole 1.271 +system is doing, we may get it all wrong, or badly wrong. 1.272 +</p> 1.273 + 1.274 +</div> 1.275 +</div> 1.276 + 1.277 +</div> 1.278 + 1.279 +<div id="outline-container-2" class="outline-2"> 1.280 +<h2 id="sec-2"><span class="section-number-2">2</span> What problems of intelligence did evolution solve?</h2> 1.281 +<div class="outline-text-2" id="text-2"> 1.282 + 1.283 + 1.284 + 1.285 +</div> 1.286 + 1.287 +<div id="outline-container-2-1" class="outline-3"> 1.288 +<h3 id="sec-2-1"><span class="section-number-3">2.1</span> Intelligence consists of solutions to many evolutionary problems; no single development (e.g. communication) was key to human-level intelligence.</h3> 1.289 +<div class="outline-text-3" id="text-2-1"> 1.290 + 1.291 + 1.292 +<p> 1.293 +[2:57] Well, first I would challenge that we are the dominant 1.294 +species. I know it looks like that but actually if you count biomass, 1.295 +if you count number of species, if you count number of individuals, 1.296 +the dominant species are microbes — maybe not one of them but anyway 1.297 +they're the ones who dominate in that sense, and furthermore we are 1.298 +mostly — we are largely composed of microbes, without which we 1.299 +wouldn't survive. 1.300 +</p> 1.301 + 1.302 +<p> 1.303 +[3:27] But there are things that make humans (you could say) best at 1.304 +those things, or worst at those things, but it's a combination. And I 1.305 +think it was a collection of developments of which there isn't any 1.306 +single one. [] there might be, some people say, human language which 1.307 +changed everything. By our human language, they mean human 1.308 +communication in words, but I think that was a later development from 1.309 +what must have started as the use of <i>internal</i> forms of 1.310 +representation — which are there in nest-building birds, in 1.311 +pre-verbal children, in hunting mammals — because you can't take in 1.312 +information about a complex structured environment in which things can 1.313 +change and you may have to be able to work out what's possible and 1.314 +what isn't possible, without having some way of representing the 1.315 +components of the environment, their relationships, the kinds of 1.316 +things they can and can't do, the kinds of things you might or might 1.317 +not be able to do — and <i>that</i> kind of capability needs internal 1.318 +languages, and I and colleagues [at Birmingham] have been referring to 1.319 +them as generalized languages because some people object to 1.320 +referring…to using language to refer to something that isn't used 1.321 +for communication. But from that viewpoint, not only humans but many 1.322 +other animals developed abilities to do things to their environment to 1.323 +make them more friendly to themselves, which depended on being able to 1.324 +represent possible futures, possible actions, and work out what's the 1.325 +best thing to do. 1.326 +</p> 1.327 +<p> 1.328 +[5:13] And nest-building in corvids for instance—crows, magpies, 1.329 + [hawks], and so on — are way beyond what current robots can do, and 1.330 + in fact I think most humans would be challenged if they had to go and 1.331 + find a collection of twigs, one at a time, maybe bring them with just 1.332 + one hand — or with your mouth — and assemble them into a 1.333 + structure that, you know, is shaped like a nest, and is fairly rigid, 1.334 + and you could trust your eggs in them when wind blows. But they're 1.335 + doing it, and so … they're not our evolutionary ancestors, but 1.336 + they're an indication — and that example is an indication — of 1.337 + what must have evolved in order to provide control over the 1.338 + environment in <i>that</i> species. 1.339 +</p> 1.340 +</div> 1.341 + 1.342 +</div> 1.343 + 1.344 +<div id="outline-container-2-2" class="outline-3"> 1.345 +<h3 id="sec-2-2"><span class="section-number-3">2.2</span> Speculation about how communication might have evolved from internal lanagues.</h3> 1.346 +<div class="outline-text-3" id="text-2-2"> 1.347 + 1.348 +<p>[5:56] And I think hunting mammals, fruit-picking mammals, mammals 1.349 +that can rearrange parts of the environment, provide shelters, needed 1.350 +to have …. also needed to have ways of representing possible 1.351 +futures, not just what's there in the environment. I think at a later 1.352 +stage, that developed into a form of communication, or rather the 1.353 +<i>internal</i> forms of representation became usable as a basis for 1.354 +providing [context] to be communicated. And that happened, I think, 1.355 +initially through performing actions that expressed intentions, and 1.356 +probably led to situtations where an action (for instance, moving some 1.357 +large object) was performed more easily, or more successfully, or more 1.358 +accurately if it was done collaboratively. So someone who had worked 1.359 +out what to do might start doing it, and then a conspecific might be 1.360 +able to work out what the intention is, because that person has the 1.361 +<i>same</i> forms of representation and can build theories about what's 1.362 +going on, and might then be able to help. 1.363 +</p> 1.364 +<p> 1.365 +[7:11] You can imagine that if that started happening more (a lot of 1.366 +collaboration based on inferred intentions and plans) then sometimes 1.367 +the inferences might be obscure and difficult, so the <i>actions</i> might 1.368 +be enhanced to provide signals as to what the intention is, and what 1.369 +the best way is to help, and so on. 1.370 +</p> 1.371 +<p> 1.372 +[7:35] So, this is all handwaving and wild speculation, but I think 1.373 +it's consistent with a large collection of facts which one can look at 1.374 +— and find if one looks for them, but one won't know if [some]one 1.375 +doesn't look for them — about the way children, for instance, who 1.376 +can't yet talk, communicate, and the things they'll do, like going to 1.377 +the mother and turning the face to point in the direction where the 1.378 +child wants it to look and so on; that's an extreme version of action 1.379 +indicating intention. 1.380 +</p> 1.381 +<p> 1.382 +[8:03] Anyway. That's a very long roundabout answer to one conjecture 1.383 +that the use of communicative language is what gave humans their 1.384 +unique power to create and destroy and whatever, and I'm saying that 1.385 +if by that you mean <i>communicative</i> language, then I'm saying there 1.386 +was something before that which was <i>non</i>-communicative language, and I 1.387 +suspect that noncommunicative language continues to play a deep role 1.388 +in <i>all</i> human perception —in mathematical and scientific reasoning, in 1.389 +problem solving — and we don't understand very much about it. 1.390 +</p> 1.391 +<p> 1.392 +[8:48] 1.393 +I'm sure there's a lot more to be said about the development of 1.394 +different kinds of senses, the development of brain structures and 1.395 +mechanisms is above all that, but perhaps I've droned on long enough 1.396 +on that question. 1.397 +</p> 1.398 + 1.399 +</div> 1.400 +</div> 1.401 + 1.402 +</div> 1.403 + 1.404 +<div id="outline-container-3" class="outline-2"> 1.405 +<h2 id="sec-3"><span class="section-number-2">3</span> How do language and internal states relate to AI?</h2> 1.406 +<div class="outline-text-2" id="text-3"> 1.407 + 1.408 + 1.409 +<p> 1.410 +[9:09] Well, I think most of the human and animal capabilities that 1.411 +I've been referring to are not yet to be found in current robots or 1.412 +[computing] systems, and I think there are two reasons for that: one 1.413 +is that it's intrinsically very difficult; I think that in particular 1.414 +it may turn out that the forms of information processing that one can 1.415 +implement on digital computers as we currently know them may not be as 1.416 +well suited to performing some of these tasks as other kinds of 1.417 +computing about which we don't know so much — for example, I think 1.418 +there may be important special features about <i>chemical</i> computers 1.419 +which we might [talk about in a little bit? find out about]. 1.420 +</p> 1.421 + 1.422 +</div> 1.423 + 1.424 +<div id="outline-container-3-1" class="outline-3"> 1.425 +<h3 id="sec-3-1"><span class="section-number-3">3.1</span> In AI, false assumptions can lead investigators astray.</h3> 1.426 +<div class="outline-text-3" id="text-3-1"> 1.427 + 1.428 +<p>[9:57] So, one of the problems then is that the tasks are hard … but 1.429 +there's a deeper problem as to why AI hasn't made a great deal of 1.430 +progress on these problems that I'm talking about, and that is that 1.431 +most AI researchers assume things—and this is not just AI 1.432 +researchers, but [also] philsophers, and psychologists, and people 1.433 +studying animal behavior—make assumptions about what it is that 1.434 +animals or humans do, for instance make assumptions about what vision 1.435 +is for, or assumptions about what motivation is and how motivation 1.436 +works, or assumptions about how learning works, and then they try --- 1.437 +the AI people try — to model [or] build systems that perform those 1.438 +assumed functions. So if you get the <i>functions</i> wrong, then even if 1.439 +you implement some of the functions that you're trying to implement, 1.440 +they won't necessarily perform the tasks that the initial objective 1.441 +was to imitate, for instance the tasks that humans, and nest-building 1.442 +birds, and monkeys and so on can perform. 1.443 +</p> 1.444 +</div> 1.445 + 1.446 +</div> 1.447 + 1.448 +<div id="outline-container-3-2" class="outline-3"> 1.449 +<h3 id="sec-3-2"><span class="section-number-3">3.2</span> Example: Vision is not just about finding surfaces, but about finding affordances.</h3> 1.450 +<div class="outline-text-3" id="text-3-2"> 1.451 + 1.452 +<p>[11:09] I'll give you a simple example — well, maybe not so simple, 1.453 +but — It's often assumed that the function of vision in humans (and 1.454 +in other animals with good eyesight and so on) is to take in optical 1.455 +information that hits the retina, and form into the (maybe changing 1.456 +— or, really, in our case definitely changing) patterns of 1.457 +illumination where there are sensory receptors that detect those 1.458 +patterns, and then somehow from that information (plus maybe other 1.459 +information gained from head movement or from comparisons between two 1.460 +eyes) to work out what there was in the environment that produced 1.461 +those patterns, and that is often taken to mean “where were the 1.462 +surfaces off which the light bounced before it came to me”. So 1.463 +you essentially think of the task of the visual system as being to 1.464 +reverse the image formation process: so the 3D structure's there, the 1.465 +lens causes the image to form in the retina, and then the brain goes 1.466 +back to a model of that 3D structure there. That's a very plausible 1.467 +theory about vision, and it may be that that's a <i>subset</i> of what 1.468 +human vision does, but I think James Gibson pointed out that that kind 1.469 +of thing is not necessarily going to be very useful for an organism, 1.470 +and it's very unlikely that that's the main function of perception in 1.471 +general, namely to produce some physical description of what's out 1.472 +there. 1.473 +</p> 1.474 +<p> 1.475 +[12:37] What does an animal <i>need</i>? It needs to know what it can do, 1.476 +what it can't do, what the consequences of its actions will be 1.477 +…. so, he introduced the word <i>affordance</i>, so from his point of 1.478 +view, the function of vision, perception, are to inform the organism 1.479 +of what the <i>affordances</i> are for action, where that would mean what 1.480 +the animal, <i>given</i> its morphology (what it can do with its mouth, its 1.481 +limbs, and so on, and the ways it can move) what it can do, what its 1.482 +needs are, what the obstacles are, and how the environment supports or 1.483 +obstructs those possible actions. 1.484 +</p> 1.485 +<p> 1.486 +[13:15] And that's a very different collection of information 1.487 +structures that you need from, say, “where are all the 1.488 +surfaces?”: if you've got all the surfaces, <i>deriving</i> the 1.489 +affordances would still be a major task. So, if you think of the 1.490 +perceptual system as primarily (for biological organisms) being 1.491 +devices that provide information about affordances and so on, then the 1.492 +tasks look very different. And most of the people working, doing 1.493 +research on computer vision in robots, I think haven't taken all that 1.494 +on board, so they're trying to get machines to do things which, even 1.495 +if they were successful, would not make the robots very intelligent 1.496 +(and in fact, even the ones they're trying to do are not really easy 1.497 +to do, and they don't succeed very well— although, there's progress; 1.498 +I shouldn't disparage it too much.) 1.499 +</p> 1.500 +</div> 1.501 + 1.502 +</div> 1.503 + 1.504 +<div id="outline-container-3-3" class="outline-3"> 1.505 +<h3 id="sec-3-3"><span class="section-number-3">3.3</span> Online and offline intelligence</h3> 1.506 +<div class="outline-text-3" id="text-3-3"> 1.507 + 1.508 + 1.509 +<p> 1.510 +[14:10] It gets more complex as animals get more sophisticated. So, I 1.511 +like to make a distinction between online intelligence and offline 1.512 +intelligence. So, for example, if I want to pick something up — like 1.513 +this leaf <he plucks a leaf from the table> — I was able to select 1.514 +it from all the others in there, and while moving my hand towards it, 1.515 +I was able to guide its trajectory, making sure it was going roughly 1.516 +in the right direction — as opposed to going out there, which 1.517 +wouldn't have been able to pick it up — and these two fingers ended 1.518 +up with a portion of the leaf between them, so that I was able to tell 1.519 +when I'm ready to do that <he clamps the leaf between two fingers> 1.520 +and at that point, I clamped my fingers and then I could pick up the 1.521 +leaf. 1.522 +</p> 1.523 +<p> 1.524 +[14:54] Whereas, — and that's an example of online intelligence: 1.525 +during the performance of an action (both from the stage where it's 1.526 +initiated, and during the intermediate stages, and where it's 1.527 +completed) I'm taking in information relevant to controlling all those 1.528 +stages, and that relevant information keeps changing. That means I 1.529 +need stores of transient information which gets discarded almost 1.530 +immediately and replaced or something. That's online intelligence. And 1.531 +there are many forms; that's just one example, and Gibson discussed 1.532 +quite a lot of examples which I won't try to replicate now. 1.533 +</p> 1.534 +<p> 1.535 +[15:30] But in offline intelligence, you're not necessarily actually 1.536 +<i>performing</i> the actions when you're using your intelligence; you're 1.537 +thinking about <i>possible</i> actions. So, for instance, I could think 1.538 +about how fast or by what route I would get back to the lecture room 1.539 +if I wanted to [get to the next talk] or something. And I know where 1.540 +the door is, roughly speaking, and I know roughly which route I would 1.541 +take, when I go out, I should go to the left or to the right, because 1.542 +I've stored information about where the spaces are, where the 1.543 +buildings are, where the door was that we came out — but in using 1.544 +that information to think about that route, I'm not actually 1.545 +performing the action. I'm not even <i>simulating</i> it in detail: the 1.546 +precise details of direction and speed and when to clamp my fingers, 1.547 +or when to contract my leg muscles when walking, are all irrelevant to 1.548 +thinking about a good route, or thinking about the potential things 1.549 +that might happen on the way. Or what would be a good place to meet 1.550 +someone who I think [for an acquaintance in particular] — [barber] 1.551 +or something — I don't necessarily have to work out exactly <i>where</i> 1.552 +the person's going to stand, or from what angle I would recognize 1.553 +them, and so on. 1.554 +</p> 1.555 +<p> 1.556 +[16:46] So, offline intelligence — which I think became not just a 1.557 +human competence; I think there are other animals that have aspects of 1.558 +it: Squirrels are very impressive as you watch them. Gray squirrels at 1.559 +any rate, as you watch them defeating squirrel-proof birdfeeders, seem 1.560 +to have a lot of that [offline intelligence], as well as the online 1.561 +intelligence when they eventually perform the action they've worked 1.562 +out [] that will get them to the nuts. 1.563 +</p> 1.564 +<p> 1.565 +[17:16] And I think that what happened during our evolution is that 1.566 +mechanisms for acquiring and processing and storing and manipulating 1.567 +information that is more and more remote from the performance of 1.568 +actions developed. An example is taking in information about where 1.569 +locations are that you might need to go to infrequently: There's a 1.570 +store of a particular type of material that's good for building on 1.571 +roofs of houses or something out around there in some 1.572 +direction. There's a good place to get water somewhere in another 1.573 +direction. There are people that you'd like to go and visit in 1.574 +another place, and so on. 1.575 +</p> 1.576 +<p> 1.577 +[17:59] So taking in information about an extended environment and 1.578 +building it into a structure that you can make use of for different 1.579 +purposes is another example of offline intelligence. And when we do 1.580 +that, we sometimes use only our brains, but in modern times, we also 1.581 +learned how to make maps on paper and walls and so on. And it's not 1.582 +clear whether the stuff inside our heads has the same structures as 1.583 +the maps we make on paper: the maps on paper have a different 1.584 +function; they may be used to communicate with others, or meant for 1.585 +<i>looking</i> at, whereas the stuff in your head you don't <i>look</i> at; you 1.586 +use it in some other way. 1.587 +</p> 1.588 +<p> 1.589 +[18:46] So, what I'm getting at is that there's a great deal of human 1.590 +intelligence (and animal intelligence) which is involved in what's 1.591 +possible in the future, what exists in distant places, what might have 1.592 +happened in the past (sometimes you need to know why something is as 1.593 +it is, because that might be relevant to what you should or shouldn't 1.594 +do in the future, and so on), and I think there was something about 1.595 +human evolution that extended that offline intelligence way beyond 1.596 +that of animals. And I don't think it was <i>just</i> human language, (but 1.597 +human language had something to do with it) but I think there was 1.598 +something else that came earlier than language which involves the 1.599 +ability to use your offline intelligence to discover something that 1.600 +has a rich mathematical structure. 1.601 +</p> 1.602 +</div> 1.603 + 1.604 +</div> 1.605 + 1.606 +<div id="outline-container-3-4" class="outline-3"> 1.607 +<h3 id="sec-3-4"><a name="example-gap" id="example-gap"></a><span class="section-number-3">3.4</span> Example: Even toddlers use sophisticated geometric knowledge</h3> 1.608 +<div class="outline-text-3" id="text-3-4"> 1.609 + 1.610 +<p>[19:44] I'll give you a simple example: if you look through a gap, you 1.611 +can see something that's on the other side of the gap. Now, you 1.612 +<i>might</i> see what you want to see, or you might see only part of it. If 1.613 +you want to see more of it, which way would you move? Well, you could 1.614 +either move <i>sideways</i>, and see through the gap—and see it roughly 1.615 +the same amount but a different part of it [if it's a ????], or you 1.616 +could move <i>towards</i> the gap and then your view will widen as you 1.617 +approach the gap. Now, there's a bit of mathematics in there, insofar 1.618 +as you are implicitly assuming that information travels in straight 1.619 +lines, and as you go closer to a gap, the straight lines that you can 1.620 +draw from where you are through the gap, widen as you approach that 1.621 +gap. Now, there's a kind of theorem of Euclidean geometry in there 1.622 +which I'm not going to try to state very precisely (and as far as I 1.623 +know, wasn't stated explicitly in Euclidean geometry) but it's 1.624 +something every toddler— human toddler—learns. (Maybe other 1.625 +animals also know it, I don't know.) But there are many more things, 1.626 +actions to perform, to get you more information about things, actions 1.627 +to perform to conceal information from other people, actions that will 1.628 +enable you to operate, to act on a rigid object in one place in order 1.629 +to produce an effect on another place. So, there's a lot of stuff that 1.630 +involves lines and rotations and angles and speeds and so on that I 1.631 +think humans (maybe, to a lesser extent, other animals) develop the 1.632 +ability to think about in a generic way. That means that you could 1.633 +take out the generalizations from the particular contexts and then 1.634 +re-use them in a new contexts in ways that I think are not yet 1.635 +represented at all in AI and in theories of human learning in any [] 1.636 +way — although some people are trying to study learning of mathematics. 1.637 +</p> 1.638 +</div> 1.639 +</div> 1.640 + 1.641 +</div> 1.642 + 1.643 +<div id="outline-container-4" class="outline-2"> 1.644 +<h2 id="sec-4"><span class="section-number-2">4</span> Animal intelligence</h2> 1.645 +<div class="outline-text-2" id="text-4"> 1.646 + 1.647 + 1.648 + 1.649 +</div> 1.650 + 1.651 +<div id="outline-container-4-1" class="outline-3"> 1.652 +<h3 id="sec-4-1"><span class="section-number-3">4.1</span> The priority is <i>cataloguing</i> what competences have evolved, not ranking them.</h3> 1.653 +<div class="outline-text-3" id="text-4-1"> 1.654 + 1.655 +<p>[22:03] I wasn't going to challenge the claim that humans can do more 1.656 +sophisticated forms of [tracking], just to mention that there are some 1.657 +things that other animals can do which are in some ways comparable, 1.658 +and some ways superior to [things] that humans can do. In particular, 1.659 +there are species of birds and also, I think, some rodents --- 1.660 +squirrels, or something — I don't know enough about the variety --- 1.661 +that can hide nuts and remember where they've hidden them, and go back 1.662 +to them. And there have been tests which show that some birds are able 1.663 +to hide tens — you know, [eighteen] or something nuts — and to 1.664 +remember which ones have been taken, which ones haven't, and so 1.665 +on. And I suspect most humans can't do that. I wouldn't want to say 1.666 +categorically that maybe we couldn't, because humans are very 1.667 +[varied], and also [a few] people can develop particular competences 1.668 +through training. But it's certainly not something I can do. 1.669 +</p> 1.670 + 1.671 +</div> 1.672 + 1.673 +</div> 1.674 + 1.675 +<div id="outline-container-4-2" class="outline-3"> 1.676 +<h3 id="sec-4-2"><span class="section-number-3">4.2</span> AI can be used to test philosophical theories</h3> 1.677 +<div class="outline-text-3" id="text-4-2"> 1.678 + 1.679 +<p>[23:01] But I also would like to say that I am not myself particularly 1.680 +interested in trying to align animal intelligences according to any 1.681 +kind of scale of superiority; I'm just trying to understand what it 1.682 +was that biological evolution produced, and how it works, and I'm 1.683 +interested in AI <i>mainly</i> because I think that when one comes up with 1.684 +theories about how these things work, one needs to have some way of 1.685 +testing the theory. And AI provides ways of implementing and testing 1.686 +theories that were not previously available: Immanuel Kant was trying 1.687 +to come up with theories about how minds work, but he didn't have any 1.688 +kind of a mechanism that he could build to test his theory about the 1.689 +nature of mathematical knowledge, for instance, or how concepts were 1.690 +developed from babyhood onward. Whereas now, if we do develop a 1.691 +theory, we have a criterion of adequacy, namely it should be precise 1.692 +enough and rich enough and detailed to enable a model to be 1.693 +built. And then we can see if it works. 1.694 +</p> 1.695 +<p> 1.696 +[24:07] If it works, it doesn't mean we've proved that the theory is 1.697 +correct; it just shows it's a candidate. And if it doesn't work, then 1.698 +it's not a candidate as it stands; it would need to be modified in 1.699 +some way. 1.700 +</p> 1.701 +</div> 1.702 +</div> 1.703 + 1.704 +</div> 1.705 + 1.706 +<div id="outline-container-5" class="outline-2"> 1.707 +<h2 id="sec-5"><span class="section-number-2">5</span> Is abstract general intelligence feasible?</h2> 1.708 +<div class="outline-text-2" id="text-5"> 1.709 + 1.710 + 1.711 + 1.712 +</div> 1.713 + 1.714 +<div id="outline-container-5-1" class="outline-3"> 1.715 +<h3 id="sec-5-1"><span class="section-number-3">5.1</span> It's misleading to compare the brain and its neurons to a computer made of transistors</h3> 1.716 +<div class="outline-text-3" id="text-5-1"> 1.717 + 1.718 +<p>[24:27] I think there's a lot of optimism based on false clues: 1.719 +the…for example, one of the false clues is to count the number of 1.720 +neurons in the brain, and then talk about the number of transistors 1.721 +you can fit into a computer or something, and then compare them. It 1.722 +might turn out that the study of the way synapses work (which leads 1.723 +some people to say that a typical synapse [] in the human brain has 1.724 +computational power comparable to the Internet a few years ago, 1.725 +because of the number of different molecules that are doing things, 1.726 +the variety of types of things that are being done in those molecular 1.727 +interactions, and the speed at which they happen, if you somehow count 1.728 +up the number of operations per second or something, then you get 1.729 +these comparable figures). 1.730 +</p> 1.731 +</div> 1.732 + 1.733 +</div> 1.734 + 1.735 +<div id="outline-container-5-2" class="outline-3"> 1.736 +<h3 id="sec-5-2"><span class="section-number-3">5.2</span> For example, brains may rely heavily on chemical information processing</h3> 1.737 +<div class="outline-text-3" id="text-5-2"> 1.738 + 1.739 +<p>Now even if the details aren't right, there may just be a lot of 1.740 +information processing that…going on in brains at the <i>molecular</i> 1.741 +level, not the neural level. Then, if that's the case, the processing 1.742 +units will be orders of magnitude larger in number than the number of 1.743 +neurons. And it's certainly the case that all the original biological 1.744 +forms of information processing were chemical; there weren't brains 1.745 +around, and still aren't in most microbes. And even when humans grow 1.746 +their brains, the process of starting from a fertilized egg and 1.747 +producing this rich and complex structure is, for much of the time, 1.748 +under the control of chemical computations, chemical information 1.749 +processing—of course combined with physical sorts of materials and 1.750 +energy and so on as well. 1.751 +</p> 1.752 +<p> 1.753 +[26:25] So it would seem very strange if all that capability was 1.754 +something thrown away when you've got a brain and all the information 1.755 +processing, the [challenges that were handled in making a brain], 1.756 +… This is handwaving on my part; I'm just saying that we <i>might</i> 1.757 +learn that what brains do is not what we think they do, and that 1.758 +problems of replicating them are not what we think they are, solely in 1.759 +terms of numerical estimate of time scales, the number of components, 1.760 +and so on. 1.761 +</p> 1.762 +</div> 1.763 + 1.764 +</div> 1.765 + 1.766 +<div id="outline-container-5-3" class="outline-3"> 1.767 +<h3 id="sec-5-3"><span class="section-number-3">5.3</span> Brain algorithms may simply be optimized for certain kinds of information processing other than bit manipulations</h3> 1.768 +<div class="outline-text-3" id="text-5-3"> 1.769 + 1.770 +<p>[26:56] But apart from that, the other basis of skepticism concerns 1.771 +how well we understand what the problems are. I think there are many 1.772 +people who try to formalize the problems of designing an intelligent 1.773 +system in terms of streams of information thought of as bit streams or 1.774 +collections of bit streams, and they think of as the problems of 1.775 +intelligence as being the construction or detection of patterns in 1.776 +those, and perhaps not just detection of patterns, but detection of 1.777 +patterns that are useable for sending <i>out</i> streams to control motors 1.778 +and so on in order to []. And that way of conceptualizing the problem 1.779 +may lead on the one hand to oversimplification, so that the things 1.780 +that <i>would</i> be achieved, if those goals were achieved, maybe much 1.781 +simpler, in some ways inadequate. Or the replication of human 1.782 +intelligence, or the matching of human intelligence—or for that 1.783 +matter, squirrel intelligence—but in another way, it may also make 1.784 +the problem harder: it may be that some of the kinds of things that 1.785 +biological evolution has achieved can't be done that way. And one of 1.786 +the ways that might turn out to be the case is not because it's not 1.787 +impossible in principle to do some of the information processing on 1.788 +artificial computers-based-on-transistors and other bit-manipulating 1.789 +[]—but it may just be that the computational complexity of solving 1.790 +problems, processes, or finding solutions to complex problems, are 1.791 +much greater and therefore you might need a much larger universe than 1.792 +we have available in order to do things. 1.793 +</p> 1.794 +</div> 1.795 + 1.796 +</div> 1.797 + 1.798 +<div id="outline-container-5-4" class="outline-3"> 1.799 +<h3 id="sec-5-4"><span class="section-number-3">5.4</span> Example: find the shortest path by dangling strings</h3> 1.800 +<div class="outline-text-3" id="text-5-4"> 1.801 + 1.802 +<p>[28:55] Then if the underlying mechanisms were different, the 1.803 +information processing mechanisms, they might be better tailored to 1.804 +particular sorts of computation. There's a [] example, which is 1.805 +finding the shortest route if you've got a collection of roads, and 1.806 +they may be curved roads, and lots of tangled routes from A to B to C, 1.807 +and so on. And if you start at A and you want to get to Z — a place 1.808 +somewhere on that map — the process of finding the shortest route 1.809 +will involve searching through all these different possibilities and 1.810 +rejecting some that are longer than others and so on. But if you make 1.811 +a model of that map out of string, where these strings are all laid 1.812 +out on the maps and so have the lengths of the routes. Then if you 1.813 +hold the two knots in the string – it's a network of string — which 1.814 +correspond to the start point and end point, then <i>pull</i>, then the 1.815 +bits of string that you're left with in a straight line will give you 1.816 +the shortest route, and that process of pulling just gets you the 1.817 +solution very rapidly in a parallel computation, where all the others 1.818 +just hang by the wayside, so to speak. 1.819 +</p> 1.820 +</div> 1.821 + 1.822 +</div> 1.823 + 1.824 +<div id="outline-container-5-5" class="outline-3"> 1.825 +<h3 id="sec-5-5"><span class="section-number-3">5.5</span> In sum, we know surprisingly little about the kinds of problems that evolution solved, and the manner in which they were solved.</h3> 1.826 +<div class="outline-text-3" id="text-5-5"> 1.827 + 1.828 +<p>[30:15] Now, I'm not saying brains can build networks of string and 1.829 +pull them or anything like that; that's just an illustration of how if 1.830 +you have the right representation, correctly implemented—or suitably 1.831 +implemented—for a problem, then you can avoid very combinatorially 1.832 +complex searches, which will maybe grow exponentially with the number 1.833 +of components in your map, whereas with this thing, the time it takes 1.834 +won't depend on how many strings you've [got on the map]; you just 1.835 +pull, and it will depend only on the shortest route that exists in 1.836 +there. Even if that shortest route wasn't obvious on the original map. 1.837 +</p> 1.838 + 1.839 +<p> 1.840 +[30:59] So that's a rather long-winded way of formulating the 1.841 +conjecture which—of supporting, a roundabout way of supporting the 1.842 +conjecture that there may be something about the way molecules perform 1.843 +computations where they have the combination of continuous change as 1.844 +things move through space and come together and move apart, and 1.845 +whatever — and also snap into states that then persist, so [as you 1.846 +learn from] quantum mechanics, you can have stable molecular 1.847 +structures which are quite hard to separate, and then in catalytic 1.848 +processes you can separate them, or extreme temperatures, or strong 1.849 +forces, but they may nevertheless be able to move very rapidly in some 1.850 +conditions in order to perform computations. 1.851 +</p> 1.852 +<p> 1.853 +[31:49] Now there may be things about that kind of structure that 1.854 +enable searching for solutions to <i>certain</i> classes of problems to be 1.855 +done much more efficiently (by brain) than anything we could do with 1.856 +computers. It's just an open question. 1.857 +</p> 1.858 +<p> 1.859 +[32:04] So it <i>might</i> turn out that we need new kinds of technology 1.860 +that aren't on the horizon in order to replicate the functions that 1.861 +animal brains perform —or, it might not. I just don't know. I'm not 1.862 +claiming that there's strong evidence for that; I'm just saying that 1.863 +it might turn out that way, partly because I think we know less than 1.864 +many people think we know about what biological evolution achieved. 1.865 +</p> 1.866 +<p> 1.867 +[32:28] There are some other possibilities: we may just find out that 1.868 +there are shortcuts no one ever thought of, and it will all happen 1.869 +much more quickly—I have an open mind; I'd be surprised, but it 1.870 +could turn up. There <i>is</i> something that worries me much more than the 1.871 +singularity that most people talk about, which is machines achieving 1.872 +human-level intelligence and perhaps taking over [the] planet or 1.873 +something. There's what I call the <i>singularity of cognitive catch-up</i> … 1.874 +</p> 1.875 +</div> 1.876 +</div> 1.877 + 1.878 +</div> 1.879 + 1.880 +<div id="outline-container-6" class="outline-2"> 1.881 +<h2 id="sec-6"><span class="section-number-2">6</span> A singularity of cognitive catch-up</h2> 1.882 +<div class="outline-text-2" id="text-6"> 1.883 + 1.884 + 1.885 + 1.886 +</div> 1.887 + 1.888 +<div id="outline-container-6-1" class="outline-3"> 1.889 +<h3 id="sec-6-1"><span class="section-number-3">6.1</span> What if it will take a lifetime to learn enough to make something new?</h3> 1.890 +<div class="outline-text-3" id="text-6-1"> 1.891 + 1.892 +<p>… SCC, singularity of cognitive catch-up, which I think we're close 1.893 +to, or maybe have already reached—I'll explain what I mean by 1.894 +that. One of the products of biological evolution—and this is one of 1.895 +the answers to your earlier questions which I didn't get on to—is 1.896 +that humans have not only the ability to make discoveries that none of 1.897 +their ancestors have ever made, but to shorten the time required for 1.898 +similar achievements to be reached by their offspring and their 1.899 +descendants. So once we, for instance, worked out ways of complex 1.900 +computations, or ways of building houses, or ways of finding our way 1.901 +around, we don't need…our children don't need to work it out for 1.902 +themselves by the same lengthy trial and error procedure; we can help 1.903 +them get there much faster. 1.904 +</p> 1.905 +<p> 1.906 +Okay, well, what I've been referring to as the singularity of 1.907 +cognitive catch-up depends on the fact that—fairly obvious, and it's 1.908 +often been commented on—that in case of humans, it's not necessary 1.909 +for each generation to learn what previous generations learned <i>in the same way</i>. And we can speed up learning once something has been 1.910 +learned, [it is able to] be learned by new people. And that has meant 1.911 +that the social processes that support that kind of education of the 1.912 +young can enormously accelerate what would have taken…perhaps 1.913 +thousands [or] millions of years for evolution to produce, can happen in 1.914 +a much shorter time. 1.915 +</p> 1.916 + 1.917 +<p> 1.918 +[34:54] But here's the catch: in order for a new advance to happen --- 1.919 +so for something new to be discovered that wasn't there before, like 1.920 +Newtonian mechanics, or the theory of relativity, or Beethoven's music 1.921 +or [style] or whatever — the individuals have to have traversed a 1.922 +significant amount of what their ancestors have learned, even if they 1.923 +do it much faster than their ancestors, to get to the point where they 1.924 +can see the gaps, the possibilities for going further than their 1.925 +ancestors, or their parents or whatever, have done. 1.926 +</p> 1.927 +<p> 1.928 +[35:27] Now in the case of knowledge of science, mathematics, 1.929 +philosophy, engineering and so on, there's been a lot of accumulated 1.930 +knowledge. And humans are living a <i>bit</i> longer than they used to, but 1.931 +they're still living for [whatever it is], a hundred years, or for 1.932 +most people, less than that. So you can imagine that there might come 1.933 +a time when in a normal human lifespan, it's not possible for anyone 1.934 +to learn enough to understand the scope and limits of what's already 1.935 +been achieved in order to see the potential for going beyond it and to 1.936 +build on what's already been done to make that…those future steps. 1.937 +</p> 1.938 +<p> 1.939 +[36:10] So if we reach that stage, we will have reached the 1.940 +singularity of cognitive catch-up because the process of education 1.941 +that enables individuals to learn faster than their ancestors did is 1.942 +the catching-up process, and it may just be that we at some point 1.943 +reach a point where catching up can only happen within a lifetime of 1.944 +an individual, and after that they're dead and they can't go 1.945 +beyond. And I have some evidence that there's a lot of that around 1.946 +because I see a lot of people coming up with what <i>they</i> think of as 1.947 +new ideas which they've struggled to come up with, but actually they 1.948 +just haven't taken in some of what was…some of what was done [] by 1.949 +other people, in other places before them. And I think that despite 1.950 +the availability of search engines which make it <i>easier</i> for people 1.951 +to get the information—for instance, when I was a student, if I 1.952 +wanted to find out what other people had done in the field, it was a 1.953 +laborious process—going to the library, getting books, and 1.954 +—whereas now, I can often do things in seconds that would have taken 1.955 +hours. So that means that if seconds [are needed] for that kind of 1.956 +work, my lifespan has been extended by a factor of ten or 1.957 +something. So maybe that <i>delays</i> the singularity, but it may not 1.958 +delay it enough. But that's an open question; I don't know. And it may 1.959 +just be that in some areas, this is more of a problem than others. For 1.960 +instance, it may be that in some kinds of engineering, we're handing 1.961 +over more and more of the work to machines anyways and they can go on 1.962 +doing it. So for instance, most of the production of computers now is 1.963 +done by a computer-controlled machine—although some of the design 1.964 +work is done by humans— a lot of <i>detail</i> of the design is done by 1.965 +computers, and they produce the next generation, which then produces 1.966 +the next generation, and so on. 1.967 +</p> 1.968 +<p> 1.969 +[37:57] I don't know if humans can go on having major advances, so 1.970 +it'll be kind of sad if we can't. 1.971 +</p> 1.972 +</div> 1.973 +</div> 1.974 + 1.975 +</div> 1.976 + 1.977 +<div id="outline-container-7" class="outline-2"> 1.978 +<h2 id="sec-7"><span class="section-number-2">7</span> Spatial reasoning: a difficult problem</h2> 1.979 +<div class="outline-text-2" id="text-7"> 1.980 + 1.981 + 1.982 +<p> 1.983 +[38:15] Okay, well, there are different problems [ ] mathematics, and 1.984 +they have to do with properties. So for instance a lot of mathematics 1.985 +that can be expressed in terms of logical structures or algebraic 1.986 +structures and those are pretty well suited for manipulation and…on 1.987 +computers, and if a problem can be specified using the 1.988 +logical/algebraic notation, and the solution method requires creating 1.989 +something in that sort of notation, then computers are pretty good, 1.990 +and there are lots of mathematical tools around—there are theorem 1.991 +provers and theorem checkers, and all kinds of things, which couldn't 1.992 +have existed fifty, sixty years ago, and they will continue getting 1.993 +better. 1.994 +</p> 1.995 + 1.996 +<p> 1.997 +But there was something that I was <a href="#sec-3-4">alluding to earlier</a> when I gave the 1.998 +example of how you can reason about what you will see by changing your 1.999 +position in relation to a door, where what you are doing is using your 1.1000 +grasp of spatial structures and how as one spatial relationship 1.1001 +changes namely you come closer to the door or move sideways and 1.1002 +parallel to the wall or whatever, other spatial relationships change 1.1003 +in parallel, so the lines from your eyes through to other parts of 1.1004 +the…parts of the room on the other side of the doorway change, 1.1005 +spread out more as you go towards the doorway, and as you move 1.1006 +sideways, they don't spread out differently, but focus on different 1.1007 +parts of the internal … that they access different parts of the 1.1008 +… of the room. 1.1009 +</p> 1.1010 +<p> 1.1011 +Now, those are examples of ways of thinking about relationships and 1.1012 +changing relationships which are not the same as thinking about what 1.1013 +happens if I replace this symbol with that symbol, or if I substitute 1.1014 +this expression in that expression in a logical formula. And at the 1.1015 +moment, I do not believe that there is anything in AI amongst the 1.1016 +mathematical reasoning community, the theorem-proving community, that 1.1017 +can model the processes that go on when a young child starts learning 1.1018 +to do Euclidean geometry and is taught things about—for instance, I 1.1019 +can give you a proof that the angles of any triangle add up to a 1.1020 +straight line, 180 degrees. 1.1021 +</p> 1.1022 + 1.1023 +</div> 1.1024 + 1.1025 +<div id="outline-container-7-1" class="outline-3"> 1.1026 +<h3 id="sec-7-1"><span class="section-number-3">7.1</span> Example: Spatial proof that the angles of any triangle add up to a half-circle</h3> 1.1027 +<div class="outline-text-3" id="text-7-1"> 1.1028 + 1.1029 +<p>There are standard proofs which involves starting with one triangle, 1.1030 +then adding a line parallel to the base one of my former students, 1.1031 +Mary Pardoe, came up with which I will demonstrate with this <he holds 1.1032 +up a pen> — can you see it? If I have a triangle here that's got 1.1033 +three sides, if I put this thing on it, on one side — let's say the 1.1034 +bottom—I can rotate it until it lies along the second…another 1.1035 +side, and then maybe move it up to the other end ~. Then I can rotate 1.1036 +it again, until it lies on the third side, and move it back to the 1.1037 +other end. And then I'll rotate it again and it'll eventually end up 1.1038 +on the original side, but it will have changed the direction it's 1.1039 +pointing in — and it won't have crossed over itself so it will have 1.1040 +gone through a half-circle, and that says that the three angles of a 1.1041 +triangle add up to the rotations of half a circle, which is a 1.1042 +beautiful kind of proof and almost anyone can understand it. Some 1.1043 +mathematicians don't like it, because they say it hides some of the 1.1044 +assumptions, but nevertheless, as far as I'm concerned, it's an 1.1045 +example of a human ability to do reasoning which, once you've 1.1046 +understood it, you can see will apply to any triangle — it's got to 1.1047 +be a planar triangle — not a triangle on a globe, because then the 1.1048 +angles can add up to more than … you can have three <i>right</i> angles 1.1049 +if you have an equator…a line on the equator, and a line going up to 1.1050 +to the north pole of the earth, and then you have a right angle and 1.1051 +then another line going down to the equator, and you have a right 1.1052 +angle, right angle, right angle, and they add up to more than a 1.1053 +straight line. But that's because the triangle isn't in the plane, 1.1054 +it's on a curved surface. In fact, that's one of the 1.1055 +differences…definitional differences you can take between planar and 1.1056 +curved surfaces: how much the angles of a triangle add up to. But our 1.1057 +ability to <i>visualize</i> and notice the generality in that process, and 1.1058 +see that you're going to be able to do the same thing using triangles 1.1059 +that stretch in all sorts of ways, or if it's a million times as 1.1060 +large, or if it's made…you know, written on, on…if it's drawn in 1.1061 +different colors or whatever — none of that's going to make any 1.1062 +difference to the essence of that process. And that ability to see 1.1063 +the commonality in a spatial structure which enables you to draw some 1.1064 +conclusions with complete certainty—subject to the possibility that 1.1065 +sometimes you make mistakes, but when you make mistakes, you can 1.1066 +discover them, as has happened in the history of geometrical theorem 1.1067 +proving. Imre Lakatos had a wonderful book called <a href="http://en.wikipedia.org/wiki/Proofs_and_Refutations"><i>Proofs and Refutations</i></a> — which I won't try to summarize — but he has 1.1068 +examples: mistakes were made; that was because people didn't always 1.1069 +realize there were subtle subcases which had slightly different 1.1070 +properties, and they didn't take account of that. But once they're 1.1071 +noticed, you rectify that. 1.1072 +</p> 1.1073 +</div> 1.1074 + 1.1075 +</div> 1.1076 + 1.1077 +<div id="outline-container-7-2" class="outline-3"> 1.1078 +<h3 id="sec-7-2"><span class="section-number-3">7.2</span> Geometric results are fundamentally different than experimental results in chemistry or physics.</h3> 1.1079 +<div class="outline-text-3" id="text-7-2"> 1.1080 + 1.1081 +<p>[43:28] But it's not the same as doing experiments in chemistry and 1.1082 +physics, where you can't be sure it'll be the same on [] or at a high 1.1083 +temperature, or in a very strong magnetic field — with geometric 1.1084 +reasoning, in some sense you've got the full information in front of 1.1085 +you; even if you don't always notice an important part of it. So, that 1.1086 +kind of reasoning (as far as I know) is not implemented anywhere in a 1.1087 +computer. And most people who do research on trying to model 1.1088 +mathematical reasoning, don't pay any attention to that, because of 1.1089 +… they just don't think about it. They start from somewhere else, 1.1090 +maybe because of how they were educated. I was taught Euclidean 1.1091 +geometry at school. Were you? 1.1092 +</p> 1.1093 +<p> 1.1094 +(Adam ford: Yeah) 1.1095 +</p> 1.1096 +<p> 1.1097 +Many people are not now. Instead they're taught set theory, and 1.1098 +logic, and arithmetic, and [algebra], and so on. And so they don't use 1.1099 +that bit of their brains, without which we wouldn't have built any of 1.1100 +the cathedrals, and all sorts of things we now depend on. 1.1101 +</p> 1.1102 +</div> 1.1103 +</div> 1.1104 + 1.1105 +</div> 1.1106 + 1.1107 +<div id="outline-container-8" class="outline-2"> 1.1108 +<h2 id="sec-8"><span class="section-number-2">8</span> Is near-term artificial general intelligence likely?</h2> 1.1109 +<div class="outline-text-2" id="text-8"> 1.1110 + 1.1111 + 1.1112 + 1.1113 +</div> 1.1114 + 1.1115 +<div id="outline-container-8-1" class="outline-3"> 1.1116 +<h3 id="sec-8-1"><span class="section-number-3">8.1</span> Two interpretations: a single mechanism for all problems, or many mechanisms unified in one program.</h3> 1.1117 +<div class="outline-text-3" id="text-8-1"> 1.1118 + 1.1119 + 1.1120 +<p> 1.1121 +[44:35] Well, this relates to what's meant by general. And when I 1.1122 +first encountered the AGI community, I thought that what they all 1.1123 +meant by general intelligence was <i>uniform</i> intelligence --- 1.1124 +intelligence based on some common simple (maybe not so simple, but) 1.1125 +single powerful mechanism or principle of inference. And there are 1.1126 +some people in the community who are trying to produce things like 1.1127 +that, often in connection with algorithmic information theory and 1.1128 +computability of information, and so on. But there's another sense of 1.1129 +general which means that the system of general intelligence can do 1.1130 +lots of different things, like perceive things, understand language, 1.1131 +move around, make things, and so on — perhaps even enjoy a joke; 1.1132 +that's something that's not nearly on the horizon, as far as I 1.1133 +know. Enjoying a joke isn't the same as being able to make laughing 1.1134 +noises. 1.1135 +</p> 1.1136 +<p> 1.1137 +Given, then, that there are these two notions of general 1.1138 +intelligence—there's one that looks for one uniform, possibly 1.1139 +simple, mechanism or collection of ideas and notations and algorithms, 1.1140 +that will deal with any problem that's solvable — and the other 1.1141 +that's general in the sense that it can do lots of different things 1.1142 +that are combined into an integrated architecture (which raises lots 1.1143 +of questions about how you combine these things and make them work 1.1144 +together) and we humans, certainly, are of the second kind: we do all 1.1145 +sorts of different things, and other animals also seem to be of the 1.1146 +second kind, perhaps not as general as humans. Now, it may turn out 1.1147 +that in some near future time, who knows—decades, a few 1.1148 +decades—you'll be able to get machines that are capable of solving 1.1149 +in a time that will depend on the nature of the problem, but any 1.1150 +problem that is solvable, and they will be able to do it in some sort 1.1151 +of tractable time — of course, there are some problems that are 1.1152 +solvable that would require a larger universe and a longer history 1.1153 +than the history of the universe, but apart from that constraint, 1.1154 +these machines will be able to do anything []. But to be able to do 1.1155 +some of the kinds of things that humans can do, like the kinds of 1.1156 +geometrical reasoning where you look at the shape and you abstract 1.1157 +away from the precise angles and sizes and shapes and so on, and 1.1158 +realize there's something general here, as must have happened when our 1.1159 +ancestors first made the discoveries that eventually put together in 1.1160 +Euclidean geometry. 1.1161 +</p> 1.1162 +<p> 1.1163 +It may be that that requires mechanisms of a kind that we don't know 1.1164 +anything about at the moment. Maybe brains are using molecules and 1.1165 +rearranging molecules in some way that supports that kind of 1.1166 +reasoning. I'm not saying they are — I don't know, I just don't see 1.1167 +any simple…any obvious way to map that kind of reasoning capability 1.1168 +onto what we currently do on computers. There is—and I just 1.1169 +mentioned this briefly beforehand—there is a kind of thing that's 1.1170 +sometimes thought of as a major step in that direction, namely you can 1.1171 +build a machine (or a software system) that can represent some 1.1172 +geometrical structure, and then be told about some change that's going 1.1173 +to happen to it, and it can predict in great detail what'll 1.1174 +happen. And this happens for instance in game engines, where you say 1.1175 +we have all these blocks on the table and I'll drop one other block, 1.1176 +and then [the thing] uses Newton's laws and properties of rigidity of 1.1177 +the parts and the elasticity and also stuff about geometries and space 1.1178 +and so on, to give you a very accurate representation of what'll 1.1179 +happen when this brick lands on this pile of things, [it'll bounce and 1.1180 +go off, and so on]. And you just, with more memory and more CPU power, 1.1181 +you can increase the accuracy— but that's totally different than 1.1182 +looking at <i>one</i> example, and working out what will happen in a whole 1.1183 +<i>range</i> of cases at a higher level of abstraction, whereas the game 1.1184 +engine does it in great detail for <i>just</i> this case, with <i>just</i> those 1.1185 +precise things, and it won't even know what the generalizations are 1.1186 +that it's using that would apply to others []. So, in that sense, [we] 1.1187 +may get AGI — artificial general intelligence — pretty soon, but 1.1188 +it'll be limited in what it can do. And the other kind of general 1.1189 +intelligence which combines all sorts of different things, including 1.1190 +human spatial geometrical reasoning, and maybe other things, like the 1.1191 +ability to find things funny, and to appreciate artistic features and 1.1192 +other things may need forms of pattern-mechanism, and I have an open 1.1193 +mind about that. 1.1194 +</p> 1.1195 +</div> 1.1196 +</div> 1.1197 + 1.1198 +</div> 1.1199 + 1.1200 +<div id="outline-container-9" class="outline-2"> 1.1201 +<h2 id="sec-9"><span class="section-number-2">9</span> Abstract General Intelligence impacts</h2> 1.1202 +<div class="outline-text-2" id="text-9"> 1.1203 + 1.1204 + 1.1205 +<p> 1.1206 +[49:53] Well, as far as the first type's concerned, it could be useful 1.1207 +for all kinds of applications — there are people who worry about 1.1208 +where there's a system that has that type of intelligence, might in 1.1209 +some sense take over control of the planet. Well, humans often do 1.1210 +stupid things, and they might do something stupid that would lead to 1.1211 +disaster, but I think it's more likely that there would be other 1.1212 +things [] lead to disaster— population problems, using up all the 1.1213 +resources, destroying ecosystems, and whatever. But certainly it would 1.1214 +go on being useful to have these calculating devices. Now, as for the 1.1215 +second kind of them, I don't know—if we succeeded at putting 1.1216 +together all the parts that we find in humans, we might just make an 1.1217 +artificial human, and then we might have some of them as your friends, 1.1218 +and some of them we might not like, and some of them might become 1.1219 +teachers or whatever, composers — but that raises a question: could 1.1220 +they, in some sense, be superior to us, in their learning 1.1221 +capabilities, their understanding of human nature, or maybe their 1.1222 +wickedness or whatever — these are all issues in which I expect the 1.1223 +best science fiction writers would give better answers than anything I 1.1224 +could do, but I did once fantasize when I [back] in 1978, that perhaps 1.1225 +if we achieved that kind of thing, that they would be wise, and gentle 1.1226 +and kind, and realize that humans are an inferior species that, you 1.1227 +know, have some good features, so they'd keep us in some kind of 1.1228 +secluded…restrictive kind of environment, keep us away from 1.1229 +dangerous weapons, and so on. And find ways of cohabitating with 1.1230 +us. But that's just fantasy. 1.1231 +</p> 1.1232 +<p> 1.1233 +Adam Ford: Awesome. Yeah, there's an interesting story <i>With Folded Hands</i> where [the computers] want to take care of us and want to 1.1234 +reduce suffering and end up lobotomizing everybody [but] keeping them 1.1235 +alive so as to reduce the suffering. 1.1236 +</p> 1.1237 +<p> 1.1238 +Aaron Sloman: Not all that different from <i>Brave New World</i>, where it 1.1239 +was done with drugs and so on, but different humans are given 1.1240 +different roles in that system, yeah. 1.1241 +</p> 1.1242 +<p> 1.1243 +There's also <i>The Time Machine</i>, H.G. Wells, where the … in the 1.1244 +distant future, humans have split in two: the Eloi, I think they were 1.1245 +called, they lived underground, they were the [] ones, and then—no, 1.1246 +the Morlocks lived underground; Eloi lived on the planet; they were 1.1247 +pleasant and pretty but not very bright, and so on, and they were fed 1.1248 +on by … 1.1249 +</p> 1.1250 +<p> 1.1251 +Adam Ford: [] in the future. 1.1252 +</p> 1.1253 +<p> 1.1254 +Aaron Sloman: As I was saying, if you ask science fiction writers, 1.1255 +you'll probably come up with a wide variety of interesting answers. 1.1256 +</p> 1.1257 +<p> 1.1258 +Adam Ford: I certainly have; I've spoken to [] of Birmingham, and 1.1259 +Sean Williams, … who else? 1.1260 +</p> 1.1261 +<p> 1.1262 +Aaron Sloman: Did you ever read a story by E.M. Forrester called <i>The Machine Stops</i> — very short story, it's <a href="http://archive.ncsa.illinois.edu/prajlich/forster.html">on the Internet somewhere</a> 1.1263 +— it's about a time when people sitting … and this was written in 1.1264 +about [1914 ] so it's about…over a hundred years ago … people are 1.1265 +in their rooms, they sit in front of screens, and they type things, 1.1266 +and they communicate with one another that way, and they don't meet; 1.1267 +they have debates, and they give lectures to their audiences that way, 1.1268 +and then there's a woman whose son says “I'd like to see 1.1269 +you” and she says “What's the point? You've got me at 1.1270 +this point ” but he wants to come and talk to her — I won't 1.1271 +tell you how it ends, but. 1.1272 +</p> 1.1273 +<p> 1.1274 +Adam Ford: Reminds me of the Internet. 1.1275 +</p> 1.1276 +<p> 1.1277 +Aaron Sloman: Well, yes; he invented … it was just extraordinary 1.1278 +that he was able to do that, before most of the components that we 1.1279 +need for it existed. 1.1280 +</p> 1.1281 +<p> 1.1282 +Adam Ford: [Another person who did that] was Vernor Vinge [] <i>True Names</i>. 1.1283 +</p> 1.1284 +<p> 1.1285 +Aaron Sloman: When was that written? 1.1286 +</p> 1.1287 +<p> 1.1288 +Adam Ford: The seventies. 1.1289 +</p> 1.1290 +<p> 1.1291 +Aaron Sloman: Okay, well a lot of the technology was already around 1.1292 +then. The original bits of internet were working, in about 1973, I was 1.1293 +sitting … 1974, I was sitting at Sussex University trying to 1.1294 +use…learn LOGO, the programming language, to decide whether it was 1.1295 +going to be useful for teaching AI, and I was sitting [] paper 1.1296 +teletype, there was paper coming out, transmitting ten characters a 1.1297 +second from Sussex to UCL computer lab by telegraph cable, from there 1.1298 +to somewhere in Norway via another cable, from there by satellite to 1.1299 +California to a computer Xerox [] research center where they had 1.1300 +implemented a computer with a LOGO system on it, with someone I had 1.1301 +met previously in Edinburgh, Danny Bobrow, and he allowed me to have 1.1302 +access to this sytem. So there I was typing. And furthermore, it was 1.1303 +duplex typing, so every character I typed didn't show up on my 1.1304 +terminal until it had gone all the way there and echoed back, so I 1.1305 +would type, and the characters would come back four seconds later. 1.1306 +</p> 1.1307 +<p> 1.1308 +[55:26] But that was the Internet, and I think Vernor Vinge was 1.1309 +writing after that kind of thing had already started, but I don't 1.1310 +know. Anyway. 1.1311 +</p> 1.1312 +<p> 1.1313 +[55:41] Another…I mentioned H.G. Wells, <i>The Time Machine</i>. I 1.1314 +recently discovered, because <a href="http://en.wikipedia.org/wiki/David_Lodge_(author)">David Lodge</a> had written a sort of 1.1315 +semi-novel about him, that he had invented Wikipedia, in advance— he 1.1316 +had this notion of an encyclopedia that was free to everybody, and 1.1317 +everybody could contribute and [collaborate on it]. So, go to the 1.1318 +science fiction writers to find out the future — well, a range of 1.1319 +possible futures. 1.1320 +</p> 1.1321 +<p> 1.1322 +Adam Ford: Well the thing is with science fiction writers, they have 1.1323 +to maintain some sort of interest for their readers, after all the 1.1324 +science fiction which reaches us is the stuff that publishers want to 1.1325 +sell, and so there's a little bit of a … a bias towards making a 1.1326 +plot device there, and so the dramatic sort of appeals to our 1.1327 +amygdala, our lizard brain; we'll sort of stay there obviously to some 1.1328 +extent. But I think that they do come up with sort of amazing ideas; I 1.1329 +think it's worth trying to make these predictions; I think that we 1.1330 +should more time on strategic forecasting, I mean take that seriously. 1.1331 +</p> 1.1332 +<p> 1.1333 +Aaron Sloman: Well, I'm happy to leave that to others; I just want to 1.1334 +try to understand these problems that bother me about how things 1.1335 +work. And it may be that some would say that's irresponsible if I 1.1336 +don't think about what the implications will be. Well, understanding 1.1337 +how humans work <i>might</i> enable us to make [] humans — I suspect it 1.1338 +wont happen in this century; I think it's going to be too difficult. 1.1339 +</p></div> 1.1340 +</div> 1.1341 +</div> 1.1342 + 1.1343 +<div id="postamble"> 1.1344 +<p class="date">Date: 2013-10-04 18:49:53 UTC</p> 1.1345 +<p class="author">Author: Dylan Holmes</p> 1.1346 +<p class="creator">Org version 7.7 with Emacs version 23</p> 1.1347 +<a href="http://validator.w3.org/check?uri=referer">Validate XHTML 1.0</a> 1.1348 + 1.1349 +</div> 1.1350 +</body> 1.1351 +</html>