LOVERS' CONTROLLER

  • MINDS AND MACHINES AND MYSTERIES

    judge writes about the esoteric and beloved debut science and philosophy book by Douglas Hofstadter

    by judge russell

    April 20 2020

  • GÖDEL ESCHER BACH: an ETERNAL GOLDEN BRAID (1979) by DOUGLAS R. HOFSTADTER

    Hofstadter's GEB is a tangle of ideas woven so deep that the reader starts to see all of the many hidden links between every little piece of their world. Earlier in my college career I had this professor who mentioned this book as one would mention the Bible -- often enough that when I saw it in a stack of books to dispose of in the basement of La Mariposa during a Books Through Bars volunteering day I snatched it immediately, curious about what kind of knowledge was inside this tome.

    At first glance it seems convoluted and a little pointless -- but like all my favorite things, it reveals its genius slowly and humbly. It's based around a firm belief that close study of certain fields can reveal truths in seemingly unrelated fields. Mostly, Hofstadter concerns himself with number theory, computer science, visual art, classical and experimental music, philosophy, neuroscience, metamathematics, and molecular biology. That sort of esotericness makes me hard-pressed to recommend this book to anyone, (...in fact there is one charming neuroscience student who I did recommend it to, and she replied with this quote from The Psychopath Test by Jon Ronson: "'If you're a geek,' sighed Deborah, 'and you're just discovering the Internet, and especially if you're a boy, Gödel, Escher, Bach, would be like your Bible.'". I can only imagine she's not reading it.) but nevertheless I believe it's really a good (if large) collection of thoughts.

    Hofstadter is concerned with uncovering the secrets of human intelligence by means of artificial intelligence, and carefully hypothesizes all sorts of claims by comparing the mind to formal systems, the LISP language to human cells, Escher drawings to "tangled hierarchies". However, the book doesn't read like a scientific paper, and due to its structure (a dialogue (parodying Carroll parodying Plato) alternating with a real chapter) often makes this sort of thinking lots of fun. Hofstadter has both endless curiosity and a strong adherence to methodology, and it's this which makes him a great thinker. He believes that the fun connections and references, analogies, isomorphisms, he strings together are actually the root of the greatest mysteries of intelligence. That an idea can so quickly bring up another, tangentially related one, means that our brains can process the two ideas and extract something that is the same out of them, and to this day (as far as I know, but I am a bad computer scientist) there is nothing in AI that approaches that sort of thing.

    In a very interesting article I read as a post-script to this book in The Atlantic Hofstadter explains his disdain (or at least disinterest) for the way AI has gone in the past 30 or so years and it's not hard to see why he feels this way. GEB feels like it's on the brink of a very new way of computing -- even now, 40 years later. Yet AI research seems to have taken a different course, concerning itself with the politically/militarily advantageous, the profitable, and at its most benign, with machines that just spit out right answers. None of the grand breakthroughs in AI, no matter how impressive they can be (translators, image recognizers, the weird neural-net-cat-creator that my girlfriend spent yesterday afternoon drawing dicks on) are concerned with being "intelligent beings". Neural nets are inspired by neurons in the brain of course, but they take liberties with the model to get better results and they don't have any higher level models that have any "thoughts".

    Maybe there's nothing wrong with abandoning the Hofstadter way of AI-- I mean, they're machines after all, why not let them "think" like machines? Plus, all of this neural net reinforcement learning stuff really works. His argument to that (and perhaps mine too, now) would be that there should be an isomorphic mapping between different forms of intelligence at a higher level. It shouldn't have to be a perfect mapping, and it's sort of absurd to believe that one is possible, but at least you should be able to relate a human brain to a sufficiently intelligent computer program at a level above the hardware (the neurons, the machine or assembly language). After all, if there isn't any similarity between a so-called artificial intelligence and the only model of intelligence we have, is it fair to call it intelligent at all?

    This brings to mind the sort of mystery that goes on in neural networks, where developer understands the input and the output layer, but doesn't understand why the program changes the weights in the middle layer on a level more complicated than "to get the right answer". Could there be some level of "thought" inside those layers that we don't have the vocabulary to talk about, much like we cannot describe the level of thoughts in our head? I don't know! In fact I'm very oblivious to any sort of neuroscientific discoveries made since the publishing of this book (whoops! maybe I should ask that aforementioned friend...). Another professor I've had here at university has ran some experiments vectorizing the meaning of letters in a passage of text, mirroring a study that vectorizes words. In the original study the researchers discovered that the neural network created a vector system that had its own internal rules, such that the vector for "king" minus the vector for "man" was very close to the vector it had created for "queen". My professor did the same with letters, and wants to deduce meaning from the results. Maybe those neural nets have "higher-level intelligence" that we just cannot see. Although I doubt it.

    To return to the book, Douglas Hofstadter writes in a playful style, and his Carrolian dialogues are at once groan-inducing and astonishingly clever (to be expected from a man who's spent his entire life in academia). He excels at what he sets out to do. His dives into number theory and mathematical logic are dry (how can they not be!) but more importantly they are not any kind of pop-science. As well, his musings on Zen Buddhism and experimental art are open minded and profound. Like a Zen master at one with the world, throughout this book Hofstadter is tactfully finding the braid that links all things, and treats each topic he falls upon with the same love and respect. He knows all things are connected, and I have come back from this book a little more aware of that fact.

    post-script: This is not a very political book (perhaps this following quote is the only explicit reference to politics or government throughout the whole thing) yet I want to leave you with this passage from the final chapter Strange Loops, or Tangled Hierarchies in light of the US government, the DNC, and countless other piece-of-shit politicians truly failing us during this period.

    "It is well to remember that in a society like ours, the legal system is, in a sense, a polite gesture granted collectively by millions of people--and it can be overridden just as easily as a river can overflow its banks. Then a seeming anarchy takes over; but anarchy has its own kinds of rules, no less than does civilized society: it is just that they operate from the bottom up, not from the top down."

    My response to Hofstadter is he's correct, but boy, does the government know how to build dams!

Leave your comment!

33 Comments