Damien Sullivan's Indiana Ramble

18 February 2004

I guess it's time for an upfront "whoami". I'm going for a joint PhD in computer and cognitive science here at Indiana University, working under the aegis of Douglas Hofstadter and the FARG/CRCC. My current long term goals are extending Metacat, making it smarter, more self-aware, maybe creative. My short term goal is porting Metacat's graphics so it can be usefully released to the community. I hope I can get Letter Spirit to be releasable by the same work.

Everything below this started off as a response to the advice that new grad students say something about who they are and what their research interests are. Then I kept adding stuff as I felt like it.

Pictures from June 2003: close up and the setting, the highest point in Tennessee, the name of which I forget. They both look a lot better (lighter) in xv than in my Solaris Mozilla.

My occasional alternative news page
26 Aug 2002

Oh look, just what I need, another web page. I've got a main page at Caltech, although if you're looking for my research interests that may be of limited use, so I should start talking here. My Caltech page has the science fiction websites I maintain -- Vinge, Brust, McKinley -- plus random stuff. I was part of the first Mosaic wave of webpages, and engage in only sporadic maintenance and in fitful ideas of what a page should be like, so I've already shown signs of needing the data archaeology of Vinge's Deepness in the Sky. And let's not talk about my e-mail archives, complicated through having been scattered among multiple small-quota accounts.

Oh right, research. I'm a PhD student here, aiming for a joint PhD with cognitive science. Like a lot of us, I've come here with ideas of working with Douglas Hofstadter. This reminds me uncomfortably of Caltech, where about half of the new freshmen come in intending to be the next Richard Feynman. I graduated in planetary science, a cousin of geology, so clearly I was one of the many who didn't stick with that plan. Of course there are some differences, such as my 'major' at the level of Caltech's consideration being fixed (CS or cog sci). Plus, unlike Feynman, Hofstadter is not dead. Vital difference, that.

But specifically I've been inspired by the FARG's work on concepts and analogies, and loops between perception and conception. Their work feels like stuff which could be in my own brain, unlike a lot of AI stuff I've been exposed to. And I agree that analogy making is a key and frequent thing humans do, although I won't claim all human thinking is by analogies. Steven Pinker's books have also been influential, in fact inspiring me in fall of 2001 to actually go to grad school. I think it may have been in one of them that I found the idea of most of our abstract concepts being expressed as analogies, or concept slips, of more physically grounded terms. 'got', 'have', 'go', 'pass', 'eat', 'fall', 'read' (an abstract process, but one easy to concretely demonstrate). We fall asleep and in love, have dreams and nightmares -- 'dream' is a verb by itself, but 'nightmare' isn't, and we don't have say "I dreamed a nightmare last night" -- read minds, eat chesspieces (at least when young; when I played with other 10 or 12 year olds they were really resistant to saying 'capture', as opposed to "I eat your knight"), etc.

I don't know specifically what I want to work on yet. I have a bunch of ideas, and no immediate leads on how to work on them.

Other random info: grew up in Chicago, went to Caltech 1992-1997, hung around for a year, went to San Francisco to make money programming. I read a lot -- science fiction, fantasy, biology, astronomy, history. Most of my CDs are Celtic music, but I've got some classical, which by default I treat as background music, because that's what it was when I was a kid. I'm an only child with two nieces and nephews. I prefer to go out to ethnic restaurants, and Italian doesn't really count, I mean Third World (plus most red sauces I've had in Italian restaurants have been bland compared to homemade.) I really didn't expect to see an Eritrean restaurant here, or sushi places, although I don't know yet if they're good. That is, I prefer to go out to ethnic restaurants when I go out; most of the time I cook at home. I'd like to get involved in dancing, although I've never stuck through lessons before. I'm looking at swing at the moment. At home I'll dance, or prance, or something, to wilder Irish music, but this is hard to share, on account of the high probability of collision with anything within ten feet. Although I once had this great romantic fantasy once of being so in tune with some girl that we could both prance around, whooshing by itself in interleaving patterns, like the motorcycles in the Sphere of Death at Magic Mountain.

Should I talk about politics or religion here? Hmm, maybe not just now.

Oh, finally, my ultimate plan: "To bring about strange aeons."

28 Aug 2002

So I've talked more with Hofstadter, and it was fun. Assuming everything works out I think I'm spiralling toward 'simply' extending Metacat in various detailed and deep ways. I mean, really, this is why I applied to IU in the first place, and half the reason I finally applied to grad school at all last fall -- I'd read Fluid Concepts and Creative Analogies and Melanie Mitchell's thesis on Copycat, and I wanted to implement Metacat as described in FCCA. Later Helga Keller sent me the Metacat thesis, so I learned it existed, but there's plenty of room for that to grow too. Letter Spirit is interesting, but I think Metacat's problem domain is more amenable to graduating on time.

My TA assignment (AI for Associate Instructor is a really confusing acronym for me, given my field) is for a class which meets when Programming Methods in Cognitive Science does. What are "computer structures" as opposed to "data structures", anyway? Guess I'll find out tomorrow. And how do they do these TA assignments? No one called me up to probe what I might know and be suited to teach. I wasn't a CS major, so I'd hoped to avoid weird advanced courses, and I haven't touched Windows more than minimally, so A110 might not have been the top choice but it'd probably have been easy to pick up. Programming I'd have been good at -- I've done it since before high school, TAed it once, and have done it commercially recently. so I'd be useful in software engineering too. "Computer structures"? I fear hardware will be involved. Rant rant, whine whine.

Comics orgasm today. Lucifer, Girl Genius, and the League fo Extraordinary Gentlemen. And Mike Carey of Lucifer has taken over Hellblazer, so I'm trying that now.

9 October 2002

So Abhijit said he liked this page, and wondered if there'd be more. I hadn't planned on updating it regularly... us folks on gale make fun of weblogs. But hey.

"Computer structures" turns out to mean "assembly language for the Motorola 68000". And it's been fun and relatively easy. Little lecturing, a lot of serial tutoring, which I like. Tutoring, that is. I'm behind on grading, though. Of course they all know their grade on that assignment anyway, it was the cheap A of the first assignment. The prof turns out to be from Caltech, explaining the open book midterm they'll be suffering from. He'd probably make it unlimited time and take-home if the university would let him and he trusted the students.

I've gotten involved with Celtic dancing. I wanted to go for swing, but I've kept feeling really busy on Mondays. Celtic's funny: I love the music, but the dances can be sadistic. At least the Scottish ones. English country wasn't too bad. Maybe that's why the English conquered the islands: the Celts were wasting time on stuff which is as precise as ballet and takes more effort. In the meantime the English were practicing the longbow and law.

Apparently there are four prospective students circling around Hofstadter, and we all want to work on Metacat. Poor Letter Spirit. Oh hey, the CRCC web page got updated. Looks cool in general, but weirdly stretched out in the huge Netscape windows I favor.

This town is much nicer with a bicycle. Wheels! I need rain gear, though, especially now that it's getting colder.

I forgot to tape Buffy last night. As Giovanna would say, Waaaaaaaah!

Who's in charge here?
21 Oct 2002

Did Iraq expel UN inspectors 4 years ago? Not according to this comparison of articles.

6 Dec 2002

I came to grad school to do research, right? October/November felt like the Era of Nachos, the educational OS we extend for my Operating Systems class. Another grad student calls it "the tortilla chips of doom". Much time got sucked into it, or worrying about it, or procrastinating about it, or recovering from working on it. Fell behind in Hofstadter's Group Theory class, although I held my own in Comp Theory. (Weekly homework keeps me honest.) And I did finally catch up in Group -- gave a presentation before Thanksgiving on characteristic and commutator subgroups, and how much many textbooks suck by our standards.

But I've managed to scrape out some more time. Got to read some SF stories of my roommate's, and Greg Egan's Schild's Ladder, and Jonathan Weiner's The Beak of the Finch finally, about people watching evolution happen in real time. Awesome stuff! And right now I'm trudging through the singinst.org pages, and possibly the RealAI stuff later, just to see what they've been up to, and see if there are useful ideas or references to steal. Then I should get back to re-reading the Metacat thesis, or reading its code.

Five people so far have directly or indirectly liked this page. I guess this is how weblogs get started. I'll have to mock those less. I'm still not putting intimate personal info up, though. I notice this is now Google's third page for me, #1 being ugcs (as is appropriate -- I'll get worried if they swap!) and #2 being a little diet page I threw together. Are people actually linking to that more than my Vinge and Brust pages?

7 Dec 2002

I think I'm officially refusing to accept any label related to AI or cognitive science. Computationalist, dynamicist, empiricist, whatever. I probably have computationalist leanings right now, but (a) I don't know enough about what any of these labels mean to anyone else to know what they'd be saying, and (b) I think labelling is counterproductive here; it leads to a tendency to see how other ideas mesh with Your Label or threaten the investment in Your Label. Of course refusing a label/tribe might mean I end up with no allies, but the grapevine whispers that's a risk of working with Hofstadter anyway because no one knows how to relate FARG stuff with Their Label, so I'll stand on my lack of principle. Call it youthful first-year idealism.

The immediate catalyst for claiming agnosticism is reading Egan and Weiner and the funky websites, actually. I used to be a fan of general intelligence and learning and skeptical of Chomskyan inbuilt grammar ideas. Then I read Steven Pinker's books and modularity totally seemed like the way to go. But Egan and the funkysites and a few talks I've been to here are pulling me back to wanting in general intelligence. Funkysites, by which I really mean some text of Eliezer Yudkowsky's, claimed that the famous facial-recognition module has actually been linked to recognition of fine visual detail, which would be interesting if true. Still a module, but not so narrow.

Weiner's contribution to all this was a quote I'll paraphrase: "The less data a science has, the more polarized are its scientists." He was talking about evolution debates ("Does stuff happen this way or that way? Rar!" *data comes in* "Oh, it's both. Who'd have thought of that?") but it seems to apply to cog sci too. Eliezer's got the idea too, in AI: "people have said symbol processing is the Single Key to Intelligence. Or parallelism. Or fuzziness. Or connectionism..."

So there I am. I don't know what I think, I need more data to have thoughts worth labelling.

Snow on the ground. I find being a Californian for ten years didn't totally spoil me. I am scared to use my bicycle at the moment, though, so I practice long walking instead.

21 Dec 2002
Feminine AI

So, I was thinking. There often seems to be a vague 'maleness' around the idea and field of AIs. Certainly of the cognitive scientists I've met here the majority have been male. And an AI itself in fiction tends to come across as... well, sexless, technically, but our culture defaults to 'maleish' for sexless, plus AI has connotations of Logic and Reason and not being so emotional, which in the US we map to men vs. women. I've heard Russia goes the other way -- men are passionate, women pragmatic. But women tend to be more religious, and math majors and programmers tend to be male.

On the other hand, while most mathematicians may be male, most men aren't mathematicians. Stereotypical male interactions are macho posturing, avoiding emotional or possibly verbal discussions, refusing to ask for directions or admit mistakes, competitiveness. Women get talking a lot, gossip, emotional support, cooperation, admission of ignorance or mistakes. (I make no claim here as to whether the stereotypes are true; I'm just dealing with the existence and perception of the stereotypes.)

Now, it seems to me that of the two it's the women who are much closer to the stereotypical highly verbal, curious, information driven AI. The Turing test is one step away from defining an AI as a computer which can gossip with you, and "rational" and "emotionless" as applied to AI tends to mean not engaging in ego protection and posturing. (Which latter aspects might suggest men aren't unemotional, just possessed of different emotions and an unwillingness to talk about it.) And women read more, and have taken more to the Internet and chat rooms as a communication tool, as opposed to a thing to take apart. So someone tell me why the default pronoun for AIs, if we're limited to 'he' and 'she', shouldn't be 'she'.

25 Dec 2002

I wonder if that last entry was worth anything...
I'm reading Andy Clark's Being There and in a way it's kind of depressing. 10^14 synapses, at least 10^16 ops/second, and we're not that bright, having to write lots of information into our information so that we can remind ourselves of what's going on every few seconds. He talks about rearranging Scrabble pieces so as to provoke new ideas; I went one step before that, to looking at the pieces frequently to remember what they are. Any AI program playing physical scrabble would look at the pieces once and remember them and manipulate the symbols in its head.

Which leaves me unconvinced that a successful AI need imitate human architecture. Humans have cheap perception, and just enough memory and sequential logic to get by, largely with environmental help. Oh, we have tons of laid down memory, and association up the wazoo, but it takes some discipline to remember 7 Scrabble pieces. Whereas our computers have expensive perception, but cheap memory (at least at the raw level, we don't know how to do high level) and cheap sequential logic. I'm not convinced that's any worse in the long run. It's not great at walking yet, but the human way isn't all that great at playing chess. And full human capability is backed up by a million times more power than any computer used in AI. So my jury's still out.

8 Jan 2003

Oil money and terror When did Arianna veer hard left?

14 Jan 2003

Free to Travel

6 March 2003

Spacecraft propulsion A lot of neat stuff. The Orion drive details (50 cents a pound to orbit?), talk about a nucler photon rocket as semi-practical, tripropellant 'staging'...

14 June 2003

I was having a simulated conversation in my head -- a common recreation -- and came up with something I thought interesting (which is why I have these things in my head.)

(imaginary dynamicist at IU): Isn't the AI work you're doing in FARG just glorified engineering?
Me: You say that as if making useful things is a bad thing.
imaginary Hofstadter: Do you really see what we're doing as engineering?
Me: I don't think there's a useful distinction for us. Geometry got started because the Egyptians needed to measure fields properly. I think the developeds of useful steam engines needed to effective discover thermodynamics before their engines could be practical. Conversely physical scientists seem to be building their own new instruments all the time, which scared me when I first learned of -- "I need all those skills too?"

I want to build artificial intelligences: this is an engineering goal which will be greatly helped by understading the science of intelligence. I want to understand human intelligence: this is a science goal which can be guided and strongly tested by engineering simulations of that intelligence (although for direct inverstigation neurology is obviously the vanguard.) And for that matter, if engineers cobble together something people come to regard as intelligent without slavish imitation of the human model, then the science of intelligence will have two different things to generalize from, rather than the single data point we have now, given the absence of alien subjects.

2 July 2003

Civil liberties are part of security by Harry Browne.

7 July 2003

ElectionMethods -- site advocating Condorcet or Approval voting, with which I think I agree. I *definitely* agree with moving up to approval voting ASAP, because it'd be an easy pure improvement; I think Condorcet is good too. IRV looks like a sham, in terms of actually giving third parties a real chance.

10 July 2003

Worked on Bongard problems and anagrams for a few hours. I feel like my neurotransmitters have been drained. At least that's how I visualize my headaches under these circumstances. I wonder if the brain can actually induce pain in itself like that. I mean, I know the brain has no pain nerves, but for this it doesn't need any.

11 July 2003

So, we're the Fluid Analogies Research Group. And one metaphor used for our programs has even been a model of water where molecules cling together for a while in clusterlets before breaking away and joining other clusterlets. So, in the spirit of stretching a metaphor to see how far it can go before it breaks, what would the solid and gas equivalents of our models be?

I figure Good Old-Fashioned AI (GOFAI) and neural networks, at least the Hopfield nets and simple feedforward ones trotted forward in introductory classes, have to be solids and gases respectively. All three (GOFAI, FARG, NN) are based on a hard computational core, just as microscopically all matter is governed by quantum electrodynamics, conservation laws, and Coulomb's Law. But in solids the low level properties propagate up fairly easily: the unit cell of a crystal determines macroscopic shape, giving you elegant and brittle structures, and GOFAI would be the equivalent, with the massive serialism of the implementation marching upward into the AI program. Ideal gases, on the other hand, practically obliterate their lower properties in simple statisical properties such as temperature and pressure, and the NNs under discussion pretty much *are* fancy statistics.

So is this metaphor worth anything? Does it get us anything besides stroking the ego of anyone working on more fluid programs? I don't know.

14 July 2003

Hell Or "Defiant Teen Behavior Modification"
Guardian article
Rocky Mountain News article
Court transcript

15 July 2003

Did my part for Harry's research. Hofstadter's problems are hard; I ran away after a few.

27 July 2003

Reactionaries vs. Conservatives -- conservatives are mostly happy with the American status quo, reactionaries want to return us to an imagine golden age. Bill Clinton was possibly the most conservative president of the last few decades (securing prosperity, not challenging any power groups); the self-styled conservatives of Reagan and the current Bushes are reactionary, seeking to overturn the social contract of the last several decades, even by bankrupting the country. And Nixon was progressive in some ways -- EPA, wage and price controls (not that those are good), considered negative income tax...

14 January 2004

Article claiming that the sin of Sodom had nothing to do with homosexuality, but with breaching hospitality and lack of compassion. And a related article.

20 January 2004

The real State of the Union courtesy of The Independent.

5 February 2004

my little myth of the evolution of sex
retcon of broken Pak biology in Niven's Known Space.

18 February 2004

Debt Clock -- US debt passes $7 trillion.

A: On February 23rd 1996, the Outstanding Public Debt jumped almost $30
to $5,017,056,630,040.53. This was the first time in history the U.S. National
Debt surpassed
         the $5 trillion mark.

         Almost exactly six years later, on February 26th, 2002, the Debt
         passed the $6 trillion dollar mark.
And, not quite two years later, it passed the $7 trillion dollar mark.