Douglas R. Hofstadter er bedst kendt for bogen Gödel, Escher, Bach, som blev et fænomen da den udkom i 1979 – en svært definerbar moppedreng om bevidsthed, intelligens og begrebet “jeg”, som kommer ud af de særeste tangenter, og som endte med at vinde Pulitzer-prisen. Den gjorde stort indtryk på mig, da jeg læste den omkring 1990, og meget af min senere læsning var inspireret af emner og personer jeg blev introduceret til via denne bog og Hofstadters øvrige bøger. Flere af hvilke jeg fandt endnu bedre – jeg tror faktisk ikke engang jeg ville placere Gödel, Escher, Bach på min Hofstadter-top-tre.
En af hans bøger – den mest tekniske i forfatterskabet – hedder Fluid Concepts and Creative Analogies. To pudsigheder om den, en generel og en personlig:
1) Den allerførste bog der blev solgt på Amazon, var et eksemplar af den titel. Det var i 1995.
2) Da jeg selv ville have fat i titlen, ikke længe efter den var udkommet, forsøgte jeg at bestille den i en boghandel i Århus. Personalet i boghandelen gav mig meldingen, at de ikke kunne bestille den hjem til mig, hvis jeg ikke kendte bogens ISBN-nummer. Det er nok den mærkeligste spørgsmål jeg har fået i en boghandel.
Nå, men det lykkedes mig, selv uden ISBN, at få fat i bogen på en eller anden måde, og jeg læste den med interesse. Den beskriver diverse forskningsprojekter Hofstadter lavede sammen med diverse studerende på Indiana University og University of Michigan – “The Fluid Analogies Research Group” er listet som medforfatter på bogen, og medlemmernes navne fremgår af individuelle kapitler – Daniel Defays, Melanie Mitchell, Robert French, David Chalmers og Gary McGraw.
Bogens hovedinhold er detaljerede beskrivelser af projekter med at udvikle computerprogrammer der modellerer en form for “intelligens” – mønstergenkendelse, kreativitet – i et eller andet mikrodomæne. Således har vi fx programmet Jumbo, der finder engelsk-lydende anagrammer. Der er Letter Spirit, som, givet nogle bogstaver i en bestemt font/stil, frembringer øvrige bogstaver i alfabetet i samme stil. Og programmet Copycat, som finder analogier i et domæne af tekststrenge: (“Suppose the letter-string abc was changed to abd; how would you change the letter-string ijk in “the same way”?).
Allermest interessant finder jeg faktisk et indledende kapitel, i hvilket Hofstadter beskriver undersøgelser han lavede som teenager, for at finde ud af hvordan trekanttal (dvs tal der kan skrives som summen af de første N heltal) og kvadrattal fordeler sig mellem hinanden. Et studie i mønstergenkendelse; som beskrivelse af en kreativ proces er dette kapitel en tour de force. Mange forskningsartikler ville være betydeligt lettere at forstå, hvis de var udstyret med et afsnit i denne stil, så man forstår baggrunden for processen i tænkningen, og ikke bare præsenteres for det færdigpudsede resultat. Men der er næppe mange, der så præcist som som Hofstadter har dokumenteret egne tankeprocesser gennem årtier.
Hofstadter er, der i 90’erne, overbevist om, at vejen frem er at beskæftige sig med diverse mikro-domæner, en man-må-krybe-før-man-kan-gå-tilgang. I et kapitel kaldet The Knotty Problem of Evaluating Research erkender han, at mange kolleger langtfra er enige med ham. Der er så mange forskelligartede tilgange til forskning i kunstig intelligens, at Hofstadter må ty til “A Crazy Bazaar” som fællesbetegnelse for hvad der foregår i feltet. Han udtrykker vantro over for en radikalt anderledes tilgang, som han hører præsenteret i en forelæsning af Doug Lenat, der præsenterer et projekt han kalder CYC:
CYC (pronounced as, and related to, the “cyc” in “encyclopedia”) involves the encoding of millions upon millions of everyday facts into a uniform representation language based on predicate calculus, in an attempt to imbue a computer with common sense of the kind that any ten-year old has. The basic message of Lenat’s talk was that this type of knowledge-intensity is the only way to go, if you really want to make models of intelligence. Anything less is baby stuff. He said that in his opinion, within twenty years or so, thanks to advances in hardware (both memory size and raw speed), the complexity of programs, and the amount of knowledge available, computers will reach and surpass humans in intelligence level. He seemed perfectly confident and quite happy about this. I was nonplussed that he could say or believe such things, but in private conversation later, he reaffirmed his statements and incidentally remarked that work in microdomains is hopelessly outmoded. It was hard to even know how to reply to this, except to say “I don’t think so.”
De senere års eksplosive udvikling inden for AI fik mig til at tænke tilbage på denne passage. For det synes jo unægtelig som om Lenat har fået mere ret i sine forudsigelser end Hofstadter. Jeg kunne godt tænke mig at se, om Hofstadter inden for nyere tid havde udtalt sig om emnet og eventuelt skiftet mening. Det har han, viser det sig. Eftertrykkeligt endda, og det er ikke noget der har beredt ham glæde. I et interview fra 2023 med Amy Jo Kim siger han om udviklingen: “It’s like a tidal wave that is washing over us at unprecedented and unimagined speed, and to me it’s quite terrifying, because it suggests that everything that I used to believe was the case is being overturned.”
Hele interviewet er dybt interessant, og kan p.t. ses på Kims YouTube-kanal Game Thinking TV her:
https://www.youtube.com/watch?v=R6e08RnJyxo
Et lidt længere citat, fra ca 30 minutter inde i interviewet:
Hofstadter: I grew up with a certain feeling about what computers can or cannot do, and I thought that artificial intelligence when I heard about it was a very fascinating goal which is to make rigid systems act fluid, but to me that was a very long, remote goal. It seemed infinitely far away, it felt as if artificial intelligence was the art of trying to make very rigid systems behave as if they were fluid, and I felt that would take enormous amounts of time, you know I felt it would be hundreds of years before anything even remotely like a human mind would be asymptotically approaching the level of the human mind, but from beneath. I never imagined that computers would rival or let alone surpass human intelligence, and in principle I thought they could rival human intelligence, I didn’t see any reason that they couldn’t but it seemed to me like it was a goal that was so far away I wasn’t worried about it, but when certain systems started appearing, maybe twenty years ago, they gave me pause, and then it started happening at an accelerating pace where unreachable goals and things that computers shouldn’t be able to do started toppling – the defeat of Garry Kasparov by Deep Blue and then going on to Go systems, Go programs… well, systems that could defeat some of the best Go players in the world, and systems got better and better at translation between languages, and then in producing intelligible responses to difficult questions in natural language, and, even writing poetry, and my whole intellectual edifice, my system of beliefs… it’s a very traumatic experience when some of your most core beliefs about the world start collapsing.
And especially when you think that the human beings are soon going to be eclipsed. It felt as if not only are my belief systems collapsing, but it feels as if the entire human race is going to be eclipsed and left in the dust. Soon. People ask me, “What do you mean by soon?”, and I don’t know what I really mean, I don’t have any way of knowing, but some part of me says five years, some part of me says twenty years, part of me says I don’t know, I have no idea. But the progress, the accelerating progress, has been so unexpected, so completely caught me off guard, not only myself, but many, many people, that there’s a certain kind of terror of an oncoming tsunami that is going to catch all of humanity off guard. It’s not clear whether that will mean the end of humanity in the sense of the systems we’ve created destroying us, it’s not clear if that’s the case, but it’s certainly conceivable. If not, it just renders humanity a small, a very small phenomenon compared to something else that is far more intelligent, and will become incomprehensible to us, as incomprehensible to us as we are to cockroaches.
Kim: That’s an interesting thought.
Hofstadter: Well I don’t think it’s interesting, I think it’s terrifying. I hate it…
Kim: You can’t help thinking it.
Hofstadter: I think about it practically all the time every single day…
Kim: Wow.
Hofstadter: … and it overwhelms me and depresses me in a way that I haven’t been depressed for a very long time.
Hofstadter er en meget højt begavet person, og han har beskæftiget sig indgående med emnet i årtier, så jeg må sige, at hans pessimisme gør indtryk. Der findes dog også andre højt begavede og vidende personer, der ikke deler hans syn på sagen, og de er måske i flertal. At der nok allerede er opnået så meget som der kan opnås med LLM-baserede systemer er jeg tilbøjelig til at give klap-lige-hesten-typerne ret i. Man finder hurtigt systemernes begrænsninger, når man prøver dem af.
Men det beroliger mig ikke stort, for jeg forventer, at mange andre gennembrud er lige om hjørnet, og hvem kan sige hvad konsekvenserne bliver af yderligere tre-fire stykker i samme størrelsesorden som dem vi har set de seneste år? Vi lever afgjort i interessante tider.