Welcome to a new year and the January AI 2005 Expert. This month, I decided to take a cue from the Christmas special issue magazines and do something a little different. So I’ve put together an alphabetical Artificial Intelligence miscellany for your delight and delectation. AI being as rich and diverse as it is, there’s a lot of variety: new and old; applied and pure; people and programs; from AI-complete problems to a zomboid Santa Clause. Next month’s Newsletter will include a do-it-yourself on machine learning using Prolog and other free software, while in a future issue I hope to include a tutorial on the connection between symbolic and subsymbolic reasoning.
Happy New Year!
An AI Alphabet
According to that compendium of hacker slang and culture the Jargon File, an AI-complete problem is one that requires creating human-level intelligence. Computer vision, for example, is generally believed to be such a problem. However, this wasn’t always the case: legend has it that back in 1967, Marvin Minsky assigned computer vision to one of his students as a summer project. And despite the next entry in this alphabet, literary quality language translation - more than just converting reports and making approximate estimates at Google pages - is probably another. If setting or choosing student projects, it is wise to check for AI-completeness.
www.catb.org/~esr/jargon/ - Jargon File main page. To get to the dictionary, go to “Browse the Jargon File”, and then to the “Glossary” link about half way down the contents.
Small, yellow, mind-bogglingly improbable, and a biological universal translator when resident in your ear, the Babelfish in Douglas Adams’s Hitch Hiker’s Guide to the Galaxy gave its name to AltaVista’s Babel Fish online translation service. This is claimed to be the first to hit the Web, but it is no longer alone. Earlier this year, I found myself using the PROMT online English-Russian translator to help me parameterise the error messages in a Web-based Russian economic simulator. Type fragments of Russian text on a virtual Cyrillic keyboard, paste into a Web text box, bash the Translate button, check the English, then type, paste and translate back. Wouldn’t this have seemed utterly science-fictional back when the name AI was coined? Back when the First Officer in any self-respecting space opera would stuff xenocontaminant swabs into his utility belt and sling a universal translator pack over his shoulder just before venturing out onto the alien soil…
en.wikipedia.org/wiki/Babel_fish - Wikipedia entry for Douglas Adams’s fish, including his fideist disproof of the existence of God.
w4.systranlinks.com/ - Systran, the translation engine on which Babel Fish is based.
world.altavista.com/ - The Babel Fish online translator.
translation2.paralink.com/ - The PROMT online translator.
sirio.deusto.es/abaitua/konzeptu/ta/vic.htm - Is it worth learning translation technology? by Joseba Abaitua, University of Deusto, enquiring what human translators need know of computer translation. He examines Jaime de Ojeda’s Spanish translation of Lewis Carroll’s Twinkle, twinkle, little bat and the problems a mechanical Ojeda would have in solving the “formal hurdles” of Carroll’s original.
Compositional systems are those whose meanings can be calculated as a function of their parts and the way these are put together. Thus the meaning, or value, of the expression 12+34 can be calculated by applying the add function to the meaning of parts 12 and 34. Compositionality is good when designing data structures and programming languages, because it makes things built from them modular and easy to process by recursive decomposition, without having to worry about the context the parts occur in. Simon Peyton-Jones has an excellent example in his Composing Contracts: An Adventure in Financial Engineering, where he describes how to build representations for a large number of financial contracts, and a compositional semantics for calculating their value. The compositionality makes it much easier to add new kinds of contracts than in other valuation engines.
Compositionality in representational systems is deemed so important that AAAI organised an autumn conference this year on compositional connectionism in cognitive modeling. I’ll end this entry with a quote from the call for papers:
The open-ended productivity of the human capabilities aspired to by AI (e.g., perception, cognition, and language) is generally taken to be a consequence of compositionality; i.e., the ability to combine constituents recursively. The aim of this symposium is to expose connectionist researchers to the broadest possible range of conceptions of composition - including those conceptions that pose the greatest challenge for connectionism - while simultaneously alerting other AI and cognitive science researchers to the range of possibilities for connectionist implementation of composition.
www.cs.uu.nl/docs/vakken/ia/stof/compcontracts.pdf - Composing Contracts: An Adventure in Financial Engineering.
courses.essex.ac.uk/lg/LG619/Semantics1/ - Doug Arnold’s course notes on Prolog and natural language processing, Essex University. There are two links to an introduction to compositional semantics using lambda-calculus and Prolog.
www.aaai.org/Symposia/Fall/2004/fss-04.html - Page for the AAAI Fall 2004 Symposium Series, with a link to the symposium on compositional connectionism in cognitive modeling. I haven’t found any of the papers available online, although they exist in dumb media as an AAAI Press technical report - see www.aaai.org/Press/Reports/Symposia/Fall/fs-04-03.html. However, the two links below indicate the kind of work that goes on. Both papers describe attempts to design nets with compositional semantics. Rushton does this by using activation values which are matrices rather than scalars, and can represent several binary relations at once; Werning uses oscillatory neural networks.
www.cs.ttu.edu/~rushton/compsem.htm, Compositional Semantics in a Localist Neural Network J.N. Rushton, University of Georgia.
service.phil-fak.uni-duesseldorf.de/ezpublish/index.php/filemanager/download/4/salzburg.pdf - Semantic Models in Neural Networks, by Markus Werning, University of Düsseldorf.
Č is for two Čapeks. As everyone knows, it was Czech author Karel Čapek who wrote Rossum’s Universal Robots, the play in which mad inventor Old Rossum usurps the role of the Creator by artificially reproducing a man in painstaking detail, while the practical industrialist Young Rossum produces a stripped-down version of humanity to be sold as inexpensive workers. Many people also believe he invented the word “robot”, but apparently it was actually his brother Josef. Had Karel done so, we’d now be studying laborics.
The word itself is derived from a Slavonic root for “work”, but there’s some disagreement on its exact connotations. Some references say it didn’t just mean “work”, but work done as a serf - the most convincing I’ve seen is that when Czechoslovakia was still feudal, “robota” referred to those days of the week that peasants had to work without pay on the lands of noblemen. After feudalism passed, “robota” continued to be used for work that one wasn’t exactly doing voluntarily or for fun, while today’s younger Czechs and Slovaks tend to use it to mean work that’s boring or uninteresting.
Čapek did not write only about robots. The Gardener’s Year is a sweet little gardening diary, surprisingly similar in feeling to the way an English author might write. Apocryphal Tales views historical and mythical figures from an unusual angle, for example the baker whose business slumps because of the miracle of the loaves and fishes. War with the Newts pits humanity against giant newts (who behave better than most humans), with satire on journalistic writing, animal intelligence testing, and much else. And in The Absolute at Large, atom-splitting power stations release the God immanent within every particle of matter; unfortunately, since there is more than one power station, they each give rise to their own Gods, which then begin to fight.
en.wikipedia.org/wiki/Karel_Capek - Wikipedia entry for Čapek, with summaries and publication details for some of his books.
capek.misto.cz/english/robot.html - Dominik Zunt’s site on Čapek, with evidence for Josef as originator of the word “robot”.
www.maxmon.com/1921ad.htm - Maxfield and Montrose about the word.
jerz.setonhill.edu/resources/RUR/ - Dennis Jerz’s page on Čapek contains a detailed summary of Rossum’s Universal Robots.
www.catbirdpress.com/bookpages/apoc.htm - About Apocryphal Tales.
www.frc.ri.cmu.edu/robotics-faq/1.html - The Robotics FAQ on origin and early uses of the word “robot”, Asimov included. It suggests this nice little definition of robotics:
Force through intelligence. Where AI meet the real world.
Death and Downloading
We’d all like to avoid death, but biology doesn’t want to cooperate. So why not download our mind to a computer? This won’t be a new idea to most AI-ers. According to Raw Kurzweil, we’ll be able to do so by 2040. But in a Slashdot thread, “cshotton” suggests that Ray’s timetable is based on hardware extrapolation and ignores the complexity of the software needed, while “MercuryWings” imagines a Microsoft Brain XP download where downloadees have to be restored from tape after a crash, and any thought about open source software such as Linux causes a nasty pain down your side to prevent viral GPL contamination. I’m sure downloading will eventually be possible - but will it be by running a massively detailed neural or sub-neural simulation, or by emulating the mind’s symbol-level processing? And let’s hope the downloadees don’t become zombies. But anyway, this - the avoidance of death - must surely be the most noble goal of AI.
www.kurzweilai.net/ - Kurzweil’s site.
radio.cbc.ca/programs/quirks/archives/02-03/oct19.html - CBC Radio’s Quirks & Quarks, with links to Ray Kurzweil’s interview in which he sets out a timetable for the development of downloading, itself downloadable as MP3 or OGG.
www.penguinputnam.com/static/packages/us/kurzweil/ - Review of Kurzweil’s The Age of Spiritual Machines.
slashdot.org/article.pl?sid=02/10/21/0249259 - Slashdot mind-downloading thread. Some rubbish, but interesting and amusing postings too.
minduploading.org/ - Home page for the Society of Neural Prosthetics and Whole Brain Emulation Science. A quick browse through the references and resources will demonstrate just how much we still need to find out about neurobiology before downloading (or uploading as some call it) becomes reality.
Elephants Don’t Play Chess
Elephants don’t play chess. They can find food, locate and dock with mates, and do all the other things needed for a happy, long and trombipulative life in a complex and uncertain environment, but what they can’t do is play chess. In other words, they don’t do symbol manipulation, but they survive without needing to. This was the theme of Rodney Brooks’s paper Elephants Don’t Play Chess. He argued that the traditional AI approach of manipulating symbols and explicitly representing goals was flawed, leading to brittle software that, when embodied in robots, would fnd it difficult and computationally expensive to get around in the real world. As an alternative, he proposed “subsumption” architectures, based on layers of steadily more complex behaviours.
For example, a robot vacuum cleaner might have a level-1 layer which just makes it wander randomly round the room, the only sensory input being from a touch sensor which causes it to turn slightly and back off when it hits something. This could be supplemented with a level-2 layer which uses simple light sensing to bias movement towards darker regions - likely to be under furniture, where dust is often ignored. On top of this, a level-3 layer might monitor the weight of the dust bag, and rebias movement towards a fixed “emptying” station once the weight crosses a certain threshold.
The point is that these layers are simple to implement and test, involve direct sensor-to-motor links rather than computationally slow cognition, and allow systems to be developed incrementally, while having some ability to get around in the real world right from the start. As with any approach, subsumption architectures in particular, and Brooks’s approach to robotics in general, have their critics. However, Brooks’s ideas are now being applied, not only to Mars exploration robots, but to the literally more mundane matter of home robotics: his company iRobot markets the Roomba autonomous vacuum cleaner. For more on this, see my alphabet entry for “V”.
people.csail.mit.edu/u/b/brooks/public_html/ - Rodney Brooks’s home page. Elephants Don’t Play Chess is linked via his publications page here, as are other papers on the same topic.
ai.eecs.umich.edu/cogarch3/ - Comparative Reference of Cognitive Architectures by Scott Dexter, Daniel McKown, Seth Rogers, Richard Simpson, and William Walsh, University of Michigan. A useful comparison, covering subsumption amongst other architectures.
citeseer.ist.psu.edu/song96sumpy.html - Citeseer abstract and links to SUMPY: A Fuzzy Software Agent, by Hongjun Song, Stan Franklin, and Aregahegn Negatu. A well-known paper on an unusual use of subsumption, namely a software agent which helps maintain a Unix file system by compressing and backing up files.
www.oricomtech.com/misc/fandm.htm - Some ideas abstracted from reading Rodney Brooks' book Flesh and Machines. Point-by-point summary from Oricom Technologies, including (under the heading “Symbiotic Home Lifeforms”), three paragraphs on vacuum cleaners and the Roomba.
www.doc.ic.ac.uk/~nd/surprise_95/journal/vol4/pma/report.html - Exploring Mars Using Intelligent Robots by Paris Andreou and Adonis Charalambides, Imperial College. A fairly non-technical discussion of how to construct a Mars Exploration Rover designed around a subsumption architecture. There’s a good selection of links and printed references.
www.irobot.com/home.cfm - iRobot home page.
Fear of RDF
Many programmers find the Semantic Web language RDF scary, because it’s touted as a system for giving the Web artificial intelligence, and that sounds difficult. So reports the December 2004 AAAI news, following a feature in the Sydney Morning Herald. In fact though, it continues, RDF is just a metadata language, or language for describing data. Using RDF is about adding such metadata to the Web so content can be better used by programs. So to do my bit to dispel the fear, I link below to one beginner’s guide to RDF and two tutorials on using it with Prolog. It’s amusing to note that attitudes to AI vary; the quote below is from the first of those RDF-Prolog tutorials:
The AI label tends to mark things which aren’t yet implemented in a generally useful manner, often because hardware or general practices haven’t yet caught up. That seems to describe the Semantic Web pretty well.
www.aaai.org/AITopics/html/archvE12.html - AAAI December 2004 news page. Search for “metadata” to find the Sydney Morning Herald feature.
www.xml.com/pub/a/2001/01/24/rdf.html - What is RDF? by Tim Bray, updated by Dan Brickley. An easy-to-read XML.com tutorial on RDF for beginners.
www.xml.com/pub/a/2001/04/25/prologrdf/ and www.xml.com/pub/a/2001/07/25/prologrdf.html - Bijan Parsia’s XML.com tutorials on RDF for Prolog programmers, using SWI Prolog. A few broken links, but still useful.
www.swi-prolog.org/packages/ - The SWI-Prolog libraries page, including an RDF parser and utilities for storing and querying RDF.
It is rumoured that, in a certain UK university psychology lab, the monkeys used as one researcher’s experimental subjects have developed specialised neurons that fire only when confronted by him. Whether or not this is true, in theories of perception, a Grandmother Cell is a neuron so specialised that it fires only when confronted by the face of one’s grandmother. The term can, it seems, be traced back to a lecture series delivered by Jerome Lettvin in 1969, where he introduced, in a discussion on neural representation, a hypothetical neuroscientist who discovers in the brains of his animal subjects
some 18,000 neurons… that responded uniquely only to the animal’s mother, however displayed, whether animate or stuffed, seen from before or behind, upside down or on a diagonal, or offered by caricature, photograph or abstraction.
The idea may have originated with one of the early models of visual perception, Oliver Selfridge’s Pandemonium. When applied to letter recognition, this bottom-up model would start with detectors for low-level features such as vertical lines, horizontal lines, and parts of circles. These would feed to higher-level detectors for particular letters, which would themselves feed to a top-level detector that selects the most highly activated letter detector and hence the most probable letter. This is the analogue of the grandmother cell.
www.bbsonline.org/Preprints/OldArchive/bbs.page.html - Connectionist Modelling In Psychology: A Localist Manifesto by Mike Page of the Medical Research Council Cognition and Brain Sciences Unit. An interesting paper in which the author compares distributed representations in connectionist models with localised representations, distinguishes between distributed processing and distributed representation, and shows that distributed models supplemented with local representations can model pyschological phenomena that fully distributed models cannot. This paper was the source of my quote on grandmother cells; the author also explains the “Yellow Volkswagen cell” .
www.cs.nott.ac.uk/~esg/memory.html - Course notes on models of human memory by Elizabeth Gordon at Nottingham. These contain brief evaluations of Pandemonium and other models - including Minsky’s Society of Mind, Kanerva’s Sparse Distributed Memory and Hofstadter’s Copycat - against human performance. Other links on her home page include her games research, based around Nilsson’s teleo-reactive agents and the Gamebots toolkit.
vision.psy.mq.edu.au/~darrenb/objects.BW.pdf - Course slides on object recognition by Darren Burke at Macquarie, with brief sections on template matching, Pandemonium, Marr, and Beiderman. His home page links to other interesting stuff, including research on visual perception in penguins and sea lions, and on why echidna brains are so unusually big.
Douglas Hofstadter is the author of Gödel, Escher, Bach, a book which, with its whimsical form-follows-function dialogues and examinations of logic, truth and beauty, Gödel’s theorem, Zen, levels of explanation, and Lisp, became popular in the early 80s. He followed it with Metamagical Themas, a collection of essays first published in Scientific American, and which show a fascination with style, analogy, metaphor, and translation. For example: what do all the diverse typefaces we use to write an A have in common? How could we build a program to recognise that they’re all A’s? Turning this through 90 degrees, what is the essence that distinguishes Courier, say, from Helvetica? If we’d never seen a B before, how could we build a program to extract the essence of Helvetica-ness from a Helvetica A and apply it to make a Helvetica B? How could we apply this essence to letters in a different writing system, for example Cyrillic, Hebrew, or Chinese?
At the Indiana University Center for Research on Concepts and Cognition, Hofstadter and colleagues have implemented these ideas in various programs with “fluid concept” architecture. For example, the “Copycat” program was written to solve analogy problems such as this:
Suppose the letter-string abc were changed to abd; how would you change the letter-string ijk in “the same way”? Possible solutions are:
- The rightmost letter was replaced by d, hence: ijk ->ijd.
- The whole string was replaced by abd, hence ijk -> abd.
- All c’s were changed to d’s, hence ijk -> ijk.
- The rightmost letter was replaced by its alphabetic successor: ijk -> ijl. Most people would give this answer, and apparently so does Copycat, on 980 out of 1000 runs.
See the links below for more on his research, and my entry for “S” for thoughts on his approach to AI research.
Hofstadter’s interests in translation extend far beyond computing. A Person Paper on Purity in Language, first published in Metamagical Themas and purporting to be by William Satire - an allusion to New York Times columnist William Safire - is a satirical and extremely clever piece on racism and sexist language. This is how it begins:
It’s high time someone blew the whistle on all the silly prattle about revamping our language to suit the purposes of certain political fanatics. You know what I’m talking about-those who accuse speakers of English of what they call “racism.” This awkward neologism, constructed by analogy with the well-established term “sexism,” does not sit well in the ears, if I may mix my metaphors. But let us grant that in our society there may be injustices here and there in the treatment of either race from time to time, and let us even grant these people their terms “racism” and “racist.” How valid, however, are the claims of the self-proclaimed “black libbers,” or “negrists”-those who would radically change our language in order to “liberate” us poor dupes from its supposed racist bias? Most of the clamor, as you certainly know by now, revolves around the age-old usage of the noun “white” and words built from it, such as chairwhite, mailwhite, repairwhite, clergywhite, middlewhite, Frenchwhite, forewhite, whitepower, whiteslaughter, oneupwhiteship, straw white, whitehandle, and so on …
www.cogs.indiana.edu/people/homepages/hofstadter.html - Douglas Hofstadter’s home page. en.wikipedia.org/wiki/Douglas_Hofstadter - Wikipedia entry.
www.cogsci.indiana.edu/ - Center for Research on Concepts and Cognition home page. The link to “research topic pages” leads to more detailed explanations of the group’s research , though some of the pages on specific implementations are unfortunately still empty.
http://web.ulyssis.org/~joa/zsp/fluidconcepts/ - Fluid Concept Architecture: A Critical Evaluation by Joaquin Vanschoren, University of Leuven. An interesting attempt to get down inside the Copycat architecture and analyse how it works.
www.cs.pdx.edu/~mm/how-to-get-copycat.html - Links to the original source of Copycat, in a somewhat outdated Lisp, and to a Java reimplementation by Scott Bolland, University of Queensland.
pp.kpnet.fi/seirioa/cdenn/hofstadt.htm - Daniel Dennett’s highly complimentary review of Hofstadter and colleagues' Fluid Concepts and Creative Analogies, including a point-by-point summary of their work, and comparison with other AI architectures.
www.bloomington.in.us/~abangert/person.html - Links to various copies of A Person Paper on Purity in Language (with Post Scriptum).
I is for Intelligence Amplification, or the use of computing devices to supplement human brain power. Examples include pen and paper, slide rules, calculators, computer algebra, the Web, search engines, and online translators. Some people believe that, rather than work on general AI, which may be impossible in principle, or at least infeasible in practice, we should concentrate instead on IA tools for specific tasks.
www.businessintelligence.com/ex/asp/code.128/xe/article.htm - Where is the “Intelligence” behind “Business Intelligence”? Thoughts on AI and IA by Jay Liebowitz from the Department of Business and Management at Johns Hopkins University.
www.aleph.se/Trans/Individual/Intelligence/ - Anders Sandberg’s pages on transhumanism and IA. Topics linked range from mind downloading to intelligence amplification without computers, including memory-improvement systems and smart drugs.
In my time demonstrating the Oxford University AI Society at Freshers' Fair, the following two “jokes” have been thrown at me all too often: “What, Artificial Intelligence? That’s far too hard for me; I don’t even have any real intelligence”. And, “What, Artificial Intelligence? I need that; I don’t even have any real intelligence”. This entry is, however, not about jokes aimed at AI, but those generated and enjoyed by it.
Here’s a joke said to have been popular in Russia before the Soviet Union broke up:
Two people are talking about democracy. One is American and one is Russian. The American tells the Russian: “We have freedom in America - we can stand next to the White House and call our president stupid without fear of being punished.” The Russian replied, “We also have freedom in Russia. We can stand in front of the Kremlin and call your president stupid without being punished, too.”
Somehow, this joke seems to be an analogy that hasn’t quite worked out. Perhaps it’s like someone telling you the answer to
Suppose the letter-string ab were changed to ba; how would you change the letter-string abA in “the same way”?
is baA, when you would expect it to be baB. If that reminds you of my entry on Hofstadter’s work, that’s not surprising: he has taken a lot of interest in the subject, as the seminar linked below on computer puns and humour demonstrates. Now, as an exercise for the academic reader, what if anything is strange about the following analogy?
The juvenile sea squirt wanders through the sea searching for a suitable rock or hunk of coral to cling to and make its home for life. For this task, it has a rudimentary nervous system. When it finds its spot and takes root, it doesn’t need its brain anymore, so it eats it! (It’s rather like getting tenure.)
www.lyons42.com/seminars/HofstPunsAndHumor.html - Are Computers Approaching Human-Level Creativity? Seminar #4 of 5: Puns and Humor. Organised by Douglas Hofstadter, and including Marvin Minsky as panelist, this is an amusing account of a seminar on computer humour. I’m not sure whether that’s because of the panelists' insights, or despite them.
www.aaai.org/AITopics/html/toons.html - AAAI’s cartoons page. Below the cartoons are links to various pieces of work on humour research, including the feature just cited.
www.stenmorten.com/English/essays/analogies.php - This short Copycat-related page contains the Dennett sea-squirt/tenure analogy.
Kibertron, Kawada, Kawasaki, and Kokoro Dreams
OK, I admit it. I just wanted an excuse to look at pictures of some really B-I-I-I-G robots, and some of the manufacturers' names began with K.
www.androidworld.com/prod01.htm - World’s greatest android projects page. Lots of robot pin-ups, with links to the parent universities or companies. I was highly amused to see that the name of one Australian robot, GURoo, developed by Gordon Wyeth and colleagues at Queensland, started with the initial letters of the words “Grossly Underfunded”. Will this start a trend?
L is for Lenat. Roughly half-way between the birth of Lisp and the end of the Japanese Fifth Generation project, Douglas Lenat wrote the mathematical discovery program AM, to test the hypothesis that simple heuristics and a uniform control structure could generate creativity. AM was initialised with definitions of various set-theoretic functions, plus heuristics for creativity. These included rules like (I translate from the Lisp) “If a function is interesting, its inverse is worth investigating”; “A concept is more interesting if it has been discovered by several independent routes”; and “The result of making both arguments of a two-argument function the same may be interesting”. From these, AM conjectured the existence of integers, addition, multiplication, factorisation, primes, the Unique Factorisation Theorem, and Goldbach’s conjecture.
These results were impressive, and caused much argument about whether Lenat had biased his creativity heuristics towards such discoveries. But unfortunately, AM then spun off into unproductive conjectures about numbers with prime numbers of prime factors and other such stuff. This led Lenat to argue that as AM discovered new mathematical entities, it would need to discover new discovery rules too. Mathematicians realise that set theory is a different kind of beast from number theory: research methods that suit one don’t suit the other. To put this into practice, Lenat designed AM’s successor Eurisko, which did try to discover new discovery rules as it ran, evaluating each new rule by its usefulness at making discoveries.
Eurisko is fun because it gave rise to a lot of folklore. There’s an amusing little bit at the end of George Johnson’s popular account, Eurisko, The Computer With A Mind Of Its Own. He tells how, in Eurisko, sometimes a “mutant” heuristic would appear that did little more than continually cause itself to be triggered, causing an infinite loop. During one run, Lenat noticed that the number representing the usefulness of one newly discovered heuristic kept rising, indicating that Eurisko had made a particularly valuable find. As it turned out the heuristic performed no useful function. It simply examined the pool of new concepts, located those with the highest usefulness values, and marked them as having been created by itself.. Read the article also to find out how Eurisko was banned from the Traveller war-game tournament, because it proved too successful at generating unbeatable strategies.
These days, Lenat is best known for Cyc, his attempt to express as much as possible of human knowledge in logical form, as a resource of common-sense knowledge for AI reasoning systems. According to Marvin Minsky in a speech delivered in 2003, AI has been brain dead for 15 years, Cyc being one of the few worthwhile exceptions. Others may disagree - but where else can you find what Scientific American tells us is just one of the chunks of knowledge in Cyc’s knowledge base. Need I translate?
(holdsIn (YearFn 1998) (embarrassed BillClinton (sexualPartner MonicaLewinsky BillClinton)))
http://www.aliciapatterson.org/APF0704/Johnson/Johnson.html - Eurisko, The Computer With A Mind Of Its Own by freelance writer George Johnson. A nice popular-science account of AM and Eurisko, and of why Lenat wrote them.
http://www.agorics.com/Library/agoricpapers/ce/ce5.html - Paper on appling market discipline to Eurisko’s rules.
http://web.media.mit.edu/~haase/thesis/node52.html - Thesis on the discovery program Cyrano, by Ken Haase, MIT. Includes a reasonably detailed comparison with AM, and a run through AM’s discoveries.
https://www.wired.com/news/technology/0,1282,58714,00.html - Wired feature on Minsky’s recent pronouncement: AI Founder Blasts Modern Research.
http://www.cyc.com/ - Cycorp home page.
http://boole.stanford.edu/cyc.html - A somewhat sceptical 1994 report on Cyc by Vaughan Pratt, Stanford.
http://www.robotwisdom.com/ai/cycresources.html - An assortment of Cyc-related links from Jorn Barger.
http://www.aaai.org/AITopics/html/common.html - Some useful links on common-sense reasoning from AAAI, including classic papers such as John McCarthy’s 1959 Programs with Common Sense, news, and various items concerning Cyc. Search for “An Entity Named Monica” to find Scientific American’s Cyc translation of presidential mis-doings.
The word “meme” was coined by Richard Dawkins in his book The Selfish Gene to denote a piece of information that, like a gene, can replicate and evolve. Examples of memes include proselytizing religions, catchphrases, nursery rhymes, advertising jingles, and the idea of memes itself. If genes compete for physical resources such as food and light, memes can be seen as competing for resources such as storage space (whether in books, academic papers, Web pages, or brains) and replication efficiency (whether by word-of-mouth gossip, transfer of virus-infected floppies, or propagation of Internet worms). It’s hard to make memetics, the study of memes, as rigorous as genetics, because genes, or at least DNA strands, are clearly-definable physical objects. But there have been some interesting attempts at applying memetics to, for example, memory. Anyway, as Edmund Chattoe’s paper on Virtual Urban Legends: Investigating the Ecology of the World Wide Web illustrates, some of it needs no excuse other than that it’s fun.
http://users.ipfw.edu/waldschg/linx1l.htm - Huge list of memetics links from G.B. Waldschmidt at Indiana University - Purdue University Fort Wayne.
http://users.lycaeum.org/~sputnik/Memetics/ - A long list of links to memetics-related publications, including papers by Richard Dawkins, Daniel Dennett and Peter Medawar.
http://www.sosig.ac.uk/iriss/papers/paper37.htm - Virtual Urban Legends: Investigating the Ecology of the World Wide Web by Edmund Chattoe, Sociology Department, Oxford University.
http://users.lycaeum.org/~sputnik/Memetics/day.life.html - A Day in the Life of a Meme by Liane Gabora, Center for the Study of the Evolution and Origin of Life, University of California. The paper, which “outlines a theory of how memes evolve, and illustrates how a memetic perspective provides not only not only a foundation for research into the dynamics of concepts and artifacts at the societal level, but a synthetic framework for understanding how mental representations are generated, organized, stored, retrieved, and expressed at the level of the individual”, is a meme-based theory of memory.
http://vmyths.com/ - The Vmyths site, where you can “Learn about computer virus myths, hoaxes, urban legends, hysteria, and the implications if you believe in them”. Not unrelated to memes, since both computer viruses, and hoaxes about computer viruses, can be seen as memes.
Neats versus Scruffies
Is there one general principle underlying intelligence, or is it a collection of domain-specific hacks and tricks bodged together? Roughly speaking, “neats” believe the former and “scruffies” the latter. Or, as the Jargon File puts it, neats tend to believe that logic is king, while scruffies favor looser, more ad-hoc methods driven by empirical knowledge.
According to Wendy Lehnert, in her AI memoir Cognition, Computers, and Car Bombs: How Yale Prepared Me for the 90’s, the term originated with Roger Schank and Robert Abelson in their research into detailed symbolic models of human memory and belief. As Lehnert says,
In particular, certain personality traits go hand and hand with certain styles of research. Schank and Abelson hit upon one such phenomenon along these lines and dubbed it the neats vs. the scruffies. These terms moved into the mainstream AI community during the early 80s, shortly after Abelson presented the phenomenon in a keynote address at the Annual Meeting of the Cognitive Science Society in 1981.
She goes on to quote from this address to the effect that because the world is messy,
models of mind become garrulous and intractable as they become more and more realistic. If one’s emphasis is on science more than on cognition, however, the canons of hard science dictate a strategy of the isolation of idealized subsystems which can be modeled with elegant productive formalisms. Clarity and precision are highly prized, even at the expense of common sense realism. To caricature this tendency with a phrases from John Tukey, the motto of the narrow hard scientist is, “Be exactly wrong, rather than approximately right”.
But the conflict goes deeper. Indeed,
the stylistic division is the same polarization that arises in all fields of science, as well as in art, in politics, in religion, in child rearing - and in all spheres of human endeavor. Psychologist Silvan Tomkins characterizes this overriding conflict as that between characterologically left-wing and right-wing world views. The left-wing personality finds the sources of value and truth to lie within individuals, whose reactions to the world define what is important. The right-wing personality asserts that all human behavior is to be understood and judged according to rules or norms which exist independent of human reaction.
http://www.catb.org/~esr/jargon/html/N/neats-vs--scruffies.html - Jargon File entry.
http://www-nlp.cs.umass.edu/ciir-pubs/cognition.pdf - Cognition, Computers, and Car Bombs: How Yale Prepared Me for the 90’s, by Wendy Lehnert.
http://www.cs.bham.ac.uk/research/cogaff/sloman.scruffy.ai.pdf - Must Intelligent Systems Be Scruffy? by Aaron Sloman. An entertaining general paper, written in 1989 as connectionism was becoming popular, in which he asserts that yes, they must, even to the extent of forcing us to deal with “scruffy semantics”. But, he concludes, horrified rejection of AI is not the correct response.
http://scruffy.csail.mit.edu/scruffspace/index.php/Scruffy%3F - The Scruffspace forum article for “Scruffy?”, concerning first use of the words. Their main page says
scruffspace is dedicated to those interested in applying ai techniques to shape every aspect of humans' lives. more specifically, it is a common ground for exchanging resources pertaining to understanding how and what techniques in modern artificial intelligence can be applied to solving problems in ubiquitous computing. it is maintained by Max, with the hopes that it will attract other scruffies who will contribute to make this a more interesting resource.
I have just been told that Taiwan has no ambassador from the US, but does deploy an air-to-air missile called the AIM-7M. Or, as my informant expressed it,
<owl:Class rdf:ID="AIM-7M-AirToAirMissile"> <rdfs:comment>A kind of #$AirToAirMissile currently (2000) used by #$Taiwan-RepublicOfChina. </rdfs:comment> <rdf:type rdf:resource="#ExistingObjectType" /> <rdf:type rdf:resource="#ProductTypeByBrand" /> <rdf:type rdf:resource="#FormalProductType" /> <rdfs:subClassOf rdf:resource="#AirToAirMissile" /> <rdfs:subClassOf rdf:resource="#GuidedMissile" /> </owl:Class> <owl:Class rdf:ID="Ambassador"> <rdfs:comment>A collection of persons; a subset of #$Diplomat. Each element of #$Ambassador is a person who is the officially appointed chief representative of a country's government in dealing with another government or an international organization. For example, the collection #$Ambassador includes the U.S. ambassadors to #$Japan, #$Russia, #$China-PeoplesRepublic, and other countries, and also the U.S. ambassador to the #$UnitedNationsOrganization. A country sends an ambassador to another country only if it officially recognizes that country's sovereign status; e.g., currently there is no U.S. ambassador to #$Taiwan-RepublicOfChina. </rdfs:comment> <rdf:type rdf:resource="#PersonTypeByOccupation" /> <rdf:type rdf:resource="#PersonTypeByPositionInOrg" /> <rdfs:subClassOf rdf:resource="#Minister-Diplomatic" /> </owl:Class>
These are only two of the assertions in OpenCyc’s latest release of its “OWL scaffolding”, an ontology or knowledge-classification hierarchy containing, it is claimed, links relating over 60,000 items. Now, I’m not currently writing any military or diplomatic reasoning programs. But the same source also tells me that the abominable snowman is a sentient, mythological, hairy animal; abrin is a powdery biological toxin; absinthe is an illegal liquor with a bitter taste; and Absolut is a brand of vodka. Presumably, were I to search for “vodka”, I’d find it’s a legal liquor with a burning taste. Olympic pentathlon, skeet shooting, and soling; paper clips, paper bags, Paxil and petals; confusion (generic), Confucian gifts; these are a few of my favourite things. Sorry, got a bit carried away there. Anyway, this is a huge repository of common-sense information, and it’s all free! This is OpenCyc, the open-source version of Lenat’s Cyc. One must know how to use these ontologies, of course - I committed an error by saying the knowledge base had told me the AIM-7M is used by Taiwan, because that was only in the comment, so can’t (yet…) be interpreted by a machine. But the point is that this and the rest of OpenCyc’s knowledge are all open source. With such resources now available, combined with the power of modern computers, this is an excellent time to experiment with common-sense reasoning in AI.
http://www.opencyc.org/ - OpenCyc home page. The ontology referred to above, and the licence conditions, are both linked from here.
http://www.w3.org/2004/OWL/ - Home page for OWL, linking to a wide range of resources.
https://www.ainewsletter.com/ - AI Expert Newsletter for June 2003, with Dennis Merrit’s explanation of ontologies and common-sense reasoning, and links to ontologies and related software.
http://www.opensource.org/ - Home page for the Open Source Initiative. Lots of information on topics such as GPL and other open-source licenses, possible business models for open-source software, and the advantages of open-source development.
http://management.itmanagersjournal.com/management/04/05/10/2052216.shtml?tid=85 - Seven open source business strategies for competitive advantage. An interesting article from IT Manager’s Journal on profiting from open-sourcing one’s software.
I recently read Time’s Eye, the new Stephen Baxter and Arthur C Clarke novel which has mysterious satellites appear and scramble time, forming a patchwork of eras into which are thrown an ape-woman, some twenty first-century soldiers and astronauts and their futuristic mobile phone, Rudyard Kipling, and Alexander the Great and Genghis Khan, with their hordes. Perhaps it’s because numerically speaking, most of the others were soldiers looting and pillaging everything in sight, but I found the book’s most sympathetic character to be the mobile phone. Equipped with “sentience circuits”, the phone is ever-helpful, offering much-needed advice on everything from weather conditions to history; and it talks and jokes with its users as if it were human. When at the end of one chapter, it has to be turned off to conserve a no-longer-replaceable battery and asks “Will I dream?”, I felt a definite pang.
Many SF authors have coined names for their personal intelligence assistants - compad, comsole, belt, spex - and it’s exciting to think that we’ve advanced so far that Baxter and Clarke could extrapolate an existing device rather than having to invent one. But do I feel like that about my own mobile? Or Microsoft’s paperclip? Or any photocopier I have ever met anywhere in the world? No. Tools like these need to do much more to understand their user’s current state and goals, to plan how to help, and to understand how to give advice or assistance with least irritation to the user, bearing in mind his current mental and emotional state. There is actually a lot of research on this, but it’s so difficult that very little has yet entered our everyday applications. The AAAI page on interfaces has some interesting news items, such as this one from the January 2005 Scientific American: “‘If we could just give our computers and phones some understanding of the limits of human attention and memory, it would make them seem a lot more thoughtful and courteous,’ says Eric Horvitz of Microsoft Research. Horvitz, [Roel] Vertegaal, [Ted] Selker and [Rosalind] Picard are among a small but growing number of researchers trying to teach computers, phones, cars and other gadgets to behave less like egocentric oafs and more like considerate colleagues.”
And in another novel, 3001: The Final Odyssey, Arthur C Clarke suggests that considerate software may make humans more considerate, too:
Illogical though it seemed, most of the human race had found it impossible not to be polite to its artificial children, however simple-minded they might be. Whole volumes of psychology, as well as popular guides (How Not to Hurt Your Computer’s Feelings; Artificial Intelligence - Real Irritation were two of the best-known titles) had been written on the subject of Man-Machine etiquette. Long ago it had been decided that, however inconsequential rudeness to robots might appear to be, it should be discouraged. All too easily, it could spread to human relationships as well.
http://www.aaai.org/AITopics/html/interfaces.html - AAAI page on interfaces, with a good variety of news and background reading. Search for “Considerate Computing” to find the quote above.
http://www.sfrevu.com/ISSUES/2004/0401/Stephen%20Baxter/Review.htm - SFRevu interview with Stephen Baxter. To quote the interviwer:
… I think we underestimate our ability to anthropomorphize devices. I’ve always felt bad for Clarke’s Hal-9000 in 2001, and yes, the cell phone in Time’s Eye is one of my favorite characters, though I wish it had gotten more air time.
Need an efficient sort? Here’s the Jargon File recommendation for a linear-time version:
A spectacular variant of bogo-sort [algorithm which repeatedly throws a deck of cards in the air, picks them up at random, and then tests whether they are in order] has been proposed which has the interesting property that, if the Many Worlds interpretation of quantum mechanics is true, it can sort an arbitrarily large array in linear time. (In the Many-Worlds model, the result of any quantum action is to split the universe-before into a sheaf of universes-after, one for each possible way the state vector can collapse; in any one of the universes-after the result appears random.) The steps are: 1. Permute the array randomly using a quantum process, 2. If the array is not sorted, destroy the universe (checking that the list is sorted requires O(n) time). Implementation of step 2 is left as an exercise for the reader.
As an implementation hint, should step 2 prove too challenging, I recommend inserting a small explosive charge into your PC and hooking it to the sortedness test, thus merely destroying yourself instead. Subjectively, the result will be the same, since all versions of you that perceive the array as unsorted will vanish, the only surviving subjective awareness being one that perceives the sorted version of the array. Quantum bogo-sort is a spoof, but if you were brave enough to try it, would it actually work? It is after all just a form of distributed computation, albeit an extreme one which grabs resources not from adjacent chips on a bus or computers in a network, but from adjacent universes in a multiverse. And this, roughly speaking, is the principle behind quantum computation, a topic which is taken very seriously indeed. To quote from an FAQ at the Centre for Quantum Computation,
What’s all this about parallel universes? If you only want to predict what quantum computers will be able to do, you only need the equations of quantum mechanics. But if you want to explain how they will do it, you need to understand that the computer you can see and touch is only one tiny facet of a far larger object, which is just as real even though its existence is only detectable indirectly, through the computational work it does for us. The best way to describe the structure of a quantum computer is not at present clear, but in some respects it is like many computers similar to the one we see, performing different but correlated computations which affect each other through quantum interference.
In fact, the Centre has such excellent tutorials on its site that I shall plead lack of space and leave you to follow their links. I have also included two links as starting points for the question of whether or not our minds use, or require, quantum effects - a belief held, as many readers will know, by cosmologist Roger Penrose amongst others. And to finish, for those with sufficient maths, I highly recommend his new book The Road to Reality as a guide to the mathematics underlying our universe.
http://www.catb.org/~esr/jargon/html/B/bogo-sort.html - Jargon File entry for bogo-sort.
http://www.qubit.org/ - Centre for Quantum Computation.
http://www.qubit.org/people/david/ - David Deutsch’s home page. Deutsch wrote The Fabric of Reality, a fascinating justification of the many-universes interpretation, based on four main strands: quantum physics; the theory of evolution; the theory of computation; and the theory of knowledge, explanation and understanding. The book has many enjoyable insights, of which I’ll mention just one: that many-universes gives us a way to distinguish knowledge (or useful information) from junk. Think of a sequence of DNA in some organism, containing one subsequence of meaningful genetic coding and one of junk DNA. The meaningful coding will have had an important causal role in each universe, because it helped determine the organism. So in those universes where the organism is the same, so will be the non-junk DNA. The junk DNA has no such role, and so will vary at random. home.earthlink.net/~djmp/FabricOfReality.html - David Park’s review of The Fabric of Reality.
http://www.edge.org/q2004/q04_print.html - Fun page of aphorisms and laws which I discovered linked from Deutsch’s blog.
http://en.wikipedia.org/wiki/Roger_Penrose - Wikipedia entry for Roger Penrose, including his views on consciousness and quantum mechanics.
http://www.consciousness.arizona.edu/quantum-mind2/ - Synopsis of the topics covered by the Tucson Quantum Mind Conference, 2003.
http://www.321books.co.uk/reviews/the-road-to-reality-by-roger-penrose.htm - About Penrose’s The Road to Reality.
Become the leader of one million artificial intelligences! This is the promise of Elixir Studio’s long-awaited game Republic: The Revolution, finally released in late 2003. Set in the ex-Soviet breakup country of Novistrana, Republic is a game of politics. Starting with a single loyal supporter, a tiny secret HQ and a very small base of local support, you must build up a nationwide faction powerful enough to oust the President and take over Novistrana, while fighting off other factions who vye for control. There is much devious social interaction, as you persuade, hire and recruit all manner of specialist characters to your cause or use less ethical methods such as blackmail to achieve your aims.
Rendered by Elixir’s “Totality Engine”, the graphics is said to be superbly detailed, and looks it from the screen shots - for example, each post-Revolution building has its own unique pattern of bullet holes. And the AI is said to be equally detailed: as your viewpoint roams the city, characters go about their daily business, buying loaves of bread, entering cafés, commuting on the city transport system. Many details remain confidential (I would love to see an open source version), but from a Strategyplanet interview, finite-state machines appear to form an important part of the implementation. The AAAI page on video games, toys, robotic pets and entertainment carries news on a huge variety of games: it’s fascinating to watch this and other sources to see the latest tricks used by designers as they build AI into their games.
http://www.elixir-studios.co.uk/nonflash/republic/republic.htm - Elixir’s Republic page.
http://archive.gamespy.com/interviews/august02/republic/ - In a GameSpy interview with James “Prophet” Fudge, Elixir explain why playing a shadow government could be more fun than attending the Republican National Convention.
http://www.deadalfs.co.uk/reviews/1023/ - “DNM”’s review of Republic for Deadalfs. One of the more critical I have read, suggesting that the AI made little difference to the reviewer’s enjoyment. www.strategyplanet.com/republic/ai.shtml - Jonathan Mayer’s detailed examination of where AI features in Republic.
http://www.aaai.org/AITopics/html/video.html - This AAAI page on video games, toys, robotic pets and entertainment has lots of interesting games material. Search for “Game sequel takes leaps in AI technology” and read about the new Sims game. Like Republic, this would appear to have some nifty AI.
http://ai-depot.com/FiniteStateMachines/FSM-Framework.html - A Finite State Machine Framework by Jason Brownlee. Short tutorial from AI depot, with state diagrams, showing how finite state machines are used in Quake. Of course, any commercial game will use many confidential tricks not covered in such publications.
Science, Simplification, and Special Somersaults
Back under my entry for “H”, I wrote about Douglas Hofstadter’s work on understanding analogical problem-solving and the nature of style. One project I hinted at but didn’t name is Letter Spirit. This tries to understand two important aspects of letters: the “categorical sameness” possessed by letters belonging to a given category, for example “a”; and the “stylistic sameness” possessed by letters belonging to a given style, for example Helvetica. It does so by designing “gridfonts” - lowercase alphabets whose letters are made by selecting straight line segments from a predefined grid. Style comes in because the program is first seeded with one or more letters drawn upon the grid in a particular way, and must then try to draw the other letters in the same style. This is made more difficult because the grids are usually small - 6 by 2 in the Letter Spirit page cited above - so there’s very little freedom to vary the way any given letter is drawn.
Now think of some well-known modern art movements - Impressionist, Fauvism, Art Nouveau, Art Deco, Cubism. Each has its own style, but it’s difficult to say just what distinguishes one from another. The history of artistic styles is a huge subject. Do the cognitive mechanisms underlying it have anything in common with styling Letter Spirit’s 26 letters in their tiny grids? Can we possibly learn anything useful about style in general by playing around with such microworlds? Hofstadter argues that we can, and that just as in other sciences, where to study a phenomenon we isolate it as far as possible from other influences, so should we when doing AI. Unfortunately, this is not the way most AI research proceeds.
Whether or not you believe that analogy and style are central to cognition, the use of microworlds is an important question for AI. Let me finish with a quote from a source I’ve mentioned before, Daniel Dennett’s review of Hofstadter’s work. Are your somersaults special?
Hofstadter has numerous important reflections to offer on “the knotty problem of evaluating research,” and one of the book’s virtues is to draw clearly for us “the vastness of the gulf that can separate different research projects that on the surface seem to belong to the same field. Those people who are interested in results will begin with a standard technology, not even questioning it at all, and then build a big system that solves many complex problems and impresses a lot of people.” He has taken a different path, and has often had difficulties convincing the grown-ups that it is a good one: “When there’s a little kid trying somersaults out for the first time next to a flashy gymnast doing flawless flips on a balance beam, who’s going to pay any attention to the kid?” A fair complaint, but part of the problem, now redressed by this book, was that the little kid didn’t try to explain (in an efficient format accessible to impatient grown-ups) why his somersaults were so special.
www.cigital.com/~gem/lspirit.html - The Letter Spirit project. Introduction to the project, with pictures of various gridfonts, and links to research papers.
pp.kpnet.fi/seirioa/cdenn/hofstadt.htm - Daniel Dennett’s highly complimentary review of Hofstadter and colleagues' Fluid Concepts and Creative Analogies, including a point-by-point summary of their work, and comparison with other AI architectures (also linked from my entry on “Hofstadter”).
www.acm.org/crossroads/xrds10-2/hofstadter.html - A Day in the Life of… Douglas Hofstadter, from the ACM student magazine Crossroads.
cf.hum.uva.nl/mmm/papers/h-93-a.html - Paper on A microworld approach to the formalization of musical knowledge by Henkjan Honing, Music Cognition Research Group (MMM), University of Amsterdam and Radboud University Nijmegen. Paper on applying AI to musical cognition. Microworlds play a key role in this research, and the paper includes a defence of their use.
Teledildonics is the integration of telepresence with sex. Classified by the Jargon File as “ha ha only serious”, the word is, according to Wikipedia, not so humorous and speculative that it can’t be used in serious contexts, and is indeed the only commonly-used word to express the precise concept. So far, teledildonics seems limited to mobile phone attachments and Internet-controlled sex toys on the tactile side, and virtual chatbot girlfriends (see AAAI link below) on the psychoemulatory. But as our understanding of neural interfacing increases, I am certain that this Erotic Computation Group page will no longer be just a spoof. Ah well, that’s quite enough work for tonight. Time to turn down the lights, open a bottle of wine, and load my favourite recording onto the pornograph…
www.catb.org/~esr/jargon/html/T/teledildonics.html - Jargon File entry.
http://en.wikipedia.org/wiki/Teledildonics - Wikipedia entry.
www.wired.com/news/culture/0,1284,66052,00.html - Wired on sex attachments for Cell Phones That Do It.
www.wired.com/news/culture/0,1284,65064,00.html - Wired on the Ins and Outs of Teledildonics, featuring the Internet-controllable sex toys.
www.aaai.org/AITopics/html/archvE11.html - AAAI report - search for “Sex, lies and AI” - on a Hong Kong company with whose virtual hi-res chatbot you can carry on a virtual affair by phone.
www.monzy.com/ecg/ - An amusing spoof page for the MIT Erotic Computation Group.
“Unconvincing” is an adjective that applies to almost every film ever made about AI. And as AAAI’s science fiction page reports from the Artificial Intelligence in the cinema site, almost every robot film ever, from Robby the Robot in The Mechanical Statue (1907) and Proteus IV in The Rubber Man (1909) onwards, concern mechanical men going out of control. The provision of counterexamples is left as an exercise for the viewer.
www.aaai.org/AITopics/html/scifi.html - AAAI’s science fiction page. This includes a link to Artificial Intelligence in the cinema and other film sites.
I have long believed that - as I suggested in the November 2004 AI Expert - one of the main forces driving robotics will be the non-threatening and enjoyable niche for robot pets and other toys. Now I learn there’s another. According to a US Today feature reported by the AAAI page on Smart Rooms, Smart Houses & Household Appliances, it’s aging baby boomers, who, as they get older, will need, and be happy to buy, increasing amounts of hi-tech domestic help, the new robot vacuum cleaners included.
This brings me back to my alphabet entry for “E”. Vacuum cleaners don’t play chess. This is because they - or at least the Roomba, manufactured by Rodney Brooks’s company iRobot - are, like elephants, controlled by a subsumption architecture. As “profesor” says on the robot-vacuum cleaner Slashdot thread:
it’s neat to watch it and see the subsumption architecture in action: “Oh look, it changed from spiraling behavior to wall following. Now it’s just going straight then turning when it finds a wall.”
One side-effect of these new robots is vacuum-cleaner hackers. Under the heading “Geek DIY”, the AAAI page on Software, Open Source Projects & Hardware reports that Phil Mass, Chris Casey and Elliot Mack, part of the Roomba design team, had in fact hoped that it would intrigue robotics enthusiasts:
The Roomba is a tempting hacker target - big payload, multiple onboard sensors. But its cleaning duties get in the way.
Despite the fact that the machine will insist in trying to clean up, there is a Roomba hackers' site, at www.roombacommunity.com/. One of their projects is the Zoomba, a Roomba with its microprocessor replaced so it can be used as a platform for robotics experimentation. Suggested applications include a Zoomba maze solver, and a mobile security robot made from a Zoomba with wireless Webcam. There’s even Zoomba tag - get two Zoombas, control them with Javelin Stamp micros, and program them to find and chase one another.
www.aaai.org/AITopics/html/rooms.html - AAAI page on Smart Rooms, Smart Houses & Household Appliances. The news item about baby boomers and home robots can be found under the heading “Domestic bliss through mechanical marvels”.
www.onrobo.com/reviews/At_Home/Vacuum_Cleaners/ - A good selection of robot vacuum cleaners from the OnRobo home and entertainment robotics site.
slashdot.org/articles/02/10/18/167257.shtml?tid=159 - Slashdot thread. The posting about watching the subsumption architecture is by “profesor”, October 18.
www.aaai.org/AITopics/html/soft.html - AAAI page on Software, Open Source Projects & Hardware. The item about hacking the Roomba is under the heading “Geek DIY”.
www.roombacommunity.com/ - The Roomba Community hacker site. This page links to the Zoomba project.
W is for Winter. Not the one currently reducing my neighbourhood to near-Siberian cold, but the AI Winter of the late 80s. The phrase was coined by analogy with “nuclear winter” - the theory that mass use of nuclear weapons would blot out the sun with smoke and dust, causing plunging global temperatures, a frozen Earth, and the extinction of humanity. The AI Winter merely caused the extinction of AI companies, partly because of the hype over expert systems and the disillusionment caused when business discovered their limitations. These included brittleness, and the inability to explain their advice at a level of abstraction naïve users could understand. H P Newquist’s book The Brain Makers is an interesting read on the business aspects of the Winter - he describes how the only demo one company could give of their technology was a baby expert system for choosing the best wine to accompany a dinner - but a review on Amazon advises taking his account with a grain of salt. Some posters on a recent AI winter comp.lang.lisp discussion advise doing the same, while others say it’s pretty accurate.
www.amazon.com/exec/obidos/tg/detail/-/0672304120/002-5934449-5620836?v=glance - Amazon review of The Brain Makers.
www.dreamsongs.com/NewFiles/Hopl2.pdf - Guy Steele and Richard Gabriel’s paper on the evolution of Lisp, from Lisp 1.5 in 1960 to standards development at the turn of the 90s. This looks briefly at the Winter from the viewpoint of Lisp.
The thread mentioned on comp.lang.lisp started in April 2003 under the title “history of AI Winter?”.
X is for Xenopsychology, the study of extraterrestrial cognition. We don’t know any E.T.’s, apart perhaps from some fossil Martian bacteria which wouldn’t have done much cognition anyway, but we can still hypothesise about the way they think. Fredzzpyggl and Bobsqrppyx may have 17 tentacles per head, look like molten bagpipes swimming in a sodium sea, and text one another by smell-phone, but according to Marvin Minsky, we’ll still be able to communicate with them, because as problem-solvers, they’ll face the same limitations on space, time, and materials as us, and hence evolve similar mental processes. These include symbolic representations for plans and goals, and economic thinking to allocate sparse resources. Not everyone would agree, as I’ve indicated in the links below.
web.media.mit.edu/~minsky/ - Marvin Minsky’s home page. His paper, Communication with Alien Intelligence, is linked from there.
www.lehigh.edu/~mhb0/aiforcogs7.html - Mark Bickhard’s tutorial on the physical symbol system hypothesis. He cites Tim Smither’s view that the idea of organisms making symbolic representations of plans and goals is just folk psychology: unhelpful in designing robots, and unlikely in natural intelligences.
Under the heading You, Robot, AAAI News reports that Hans Moravec, like Ray Kurzweil, believes we’ll one day download our minds into computers. But that’s a long way off. In the meantime, he has started a company, Seegrid, to build vision systems that enable robots to move supplies around warehouses with no human direction. Most industrial robots don’t have vision, and hence find it extremely difficult to deal with uncertainty.
As Moravec says, it’s a long way from downloading minds, but you have to start somewhere. Nearly everything sold has to be warehoused at some point, and at some point it must also be rerouted and shipped. Seegrid’s robots can automate the work now done by human workers who move millions of tons of supplies and products using dollies, pallet jacks and forklifts. In the warehouse or out of it, if robots are going to succeed, the world cannot be adapted to them; they have to adapt to the world, just like the rest of us.
www.frc.ri.cmu.edu/~hpm/ - Hans Moravec’s home page.
www.seegrid.com/ - Seegrid home page.
www.aaai.org/AITopics/html/current.html - AAAI news page with the Seegrid report. The original feature, from Scientific American, is linked from there.
Once upon a time, there was a man who like everybody else, had a brain and a mind; and who not like everybody else, believed his consciousness to be separate from both. Perhaps caused by one or the other - an epiphenomenon - but certainly not an influence on either. One day his much-loved pet hamster was run over by a bus. Heart is pierced by grief, all our hero wanted was to die. At the same time, he realised how distressing his suicide would be to friends and family.
Suddenly, from a puff of smoke, stepped an angel who proffered a marvellous potion. One sip, the angel explained, and consciousness would be entirely annulled - our hero would never feel anything, never be aware, ever again. But since consciousness is merely an epiphenomenon of the brain, this would not affect body and behaviour, which would carry on exactly as if he were still conscious. Barely hesitating, the grieving protagonist snatched the potion from the angel’s hand and downed it in one, extinguishing his subjective awareness for ever. His body, however, continued its daily round. Twenty-four hours later, with another puff of smoke, the angel reappeared and asked him how he felt. “That potion you gave me”, complained the protagonist, “didn’t work at all. I still feel as much grief, as much loss, as I ever did.”
This, roughly, is the plot of An Unfortunate Dualist, a little fable by philosopher Raymond Smullyan, which appears reprinted in the excellent collection of philosophical essays The Mind’s I: Fantasies and Reflections on Self and Soul by Hofstadter and Dennett. The protagonist is a “zombie” - a character who lacks consciousness, but who acts otherwise exactly as he would if he had it. Zombies frequently participate in philosophical thought experiments. Often, they do so to demonstrate how ridiculous it is to assume that one can run an exact simulation of a human, for example a downloaded brain, without it being conscious.
www.indiana.edu/~alldrp/members/smullyan.html - Indiana University page for Raymond Smullyan.
www.artsci.wustl.edu/~philos/MindDict/zombie.html - The entry for “zombie” in Chris Eliasmith’s Web dictionary of Philosphy of Mind.
ase.tufts.edu/cogstud/papers/unzombie.htm - An excerpt from Daniel Dennett’s paper The Unimagined Preposterousness of Zombies, which he begins with the following provocative remarks:
Sometimes philosophers clutch an insupportable hypothesis to their bosoms and run headlong over the cliff edge. Then, like cartoon characters, they hang there in mid-air, until they notice what they have done and gravity takes over. Just such a boon is the philosophers' concept of a zombie, a strangely attractive notion that sums up, in one leaden lump, almost everything that I think is wrong with current thinking about consciousness. Philosophers ought to have dropped the zombie like a hot potato, but since they persist in their embrace, this gives me a golden opportunity to focus attention on the most seductive error in current thinking.
members.aol.com/lshauser/zomboid.html - Zombies Invade Philosophy! home page. Links to a lot of interesting philosophy, including papers and a links page by philosopher David Chalmers. As with Dennett, his stuff is definitely worth reading.
www.atomicdeathray.com/unprofessional/zombies/zombies.html - It’s Christmas, so I’ll finish with this little story, Zombies of the North Pole. Written by David Bryant, it features Santa Clause as you’ve never seen him before.
Islands of Truth in the Net of a Million Lies
I want to credit AAAI, the American Association for Artificial Intelligence, many of whose pages I’ve linked to as backup for my alphabet entries. They have a diverse collection of resources on various aspects of AI, with pointers to background reading as well as news. Less seriously, I’ve referred to pages from the Jargon File. A few of my links point at Wikipedia entries. These can be handy summaries, but one should read them with caution - some appear to be incomplete, or even wrong in parts. It’s a shame that many Wikipedia authors seem not to declare themselves; doing so would provide some chance of evaluating their biases and experience. And to anyone new to the Web, it’s worth noting that in his novel A Fire Upon the Deep, Vernor Vinge called his galaxy-spanning communications network, " the Net of a Million Lies". Or as Matt Visser’s relativity resource page has it, “the net is an example of semi-organized anarchy with zero quality control”. And of course, textbooks, course notes and research papers can contain errors and oversimplifications, on the Web or off it.
Tekkotsu and Cognitive Robotics
Following last issue’s feature on programming the Aibo, David Touretzky mailed to say that he is creating a “Cognitive Robotics” course for Carnegie Mellon that will use AIBO as the platform. Together with Ethan Tira-Thompson and colleagues, David is developing higher-level cognitive primitives for the Tekkotsu open-source robot-programming framework that will allow AIBO to be programmed at a more abstract level than currently possible. Some preliminary information is available in their paper Cognitive Primitives for Mobile Robots.
www-2.cs.cmu.edu/~dst/ - David Touretzky’s home page.
www-2.cs.cmu.edu/~tekkotsu/ - Tekkotsu home page.
www-2.cs.cmu.edu/~tekkotsu/media/FS5-04Tira-ThompsonE.pdf - Cognitive Primitives for Mobile Robots, by Ethan Tira-Thompson, Neil Halelamien, Jordan Wales, and David Touretzky, Carnegie Mellon University.