author : Douglas R. Hofstadter

The Pulitzer Prize winning tour de force: if you haven’t read it, read it; if you have read it, you won’t need to be told to read it again.

A collection of essays, including Raymond M. Smullyan's marvellous "Is God a Taoist?" and Thomas Nagel's classic "What is it Like to be a Bat?", each accompanied by Reflections of the editors.

In particular, John Searle's infamous Chinese Room argument, here explained in his "Minds, Brains and Programs" essay, gets treated to a long, and opposed, Reflection.

A collection of his Scientific American columns, plus some other essays. Most of these are on self-reference, consciousness, cognition, and patterns, enlivened with puns and word play, as one would expect.

Includes:

• "Nomic: A Self-Modifying Game Based on Reflexivity in Law". Peter Suber's game where the rules (at least initially!) are about how to change the rules.
• "A Person Paper on the Purity of Language". A little gem of an essay. It illuminates sexist language by 'rotating it 90 degrees' onto an analogous 'racist' form of language (with man=>white, woman=>black). The result is so deeply shocking and offensive (intentionally) that it illuminates brilliantly why the original should be, too. This is a much cleverer approach than the usual 180 degree rotation (man<=>woman), which often results in (unintentional) humour, instead.

Douglas Hofstadter and his research group, FARG, have a particular approach to modelling human thought processes, and this book collects together some of their work over the last two decades. They believe that perception is a key part of cognition, that analogy is a key component of perception, and that AI cognition research programmes that start from pre-perceived data input are missing the point: much of the hard work has already been done by the (human) pre-processors.

FARG conducts its research in various stripped-down micro-domains. The book covers (in chronological order) Seek Whence (what is the next number in the sequence), Jumbo (anagrams), Numbo (arithmetical puzzles), Copycat (analogous changes to alphabetic strings), Tabletop (analogous regions on a coffee-house table), and Letter Spirit (alphabetic grid-font design).

The applications are getting more ambitious, but they are all based on a common underlying architecture. Very small tasks are carried out by codelets; successful codelets get to produce further related codelets and to increase the importance of associated relationships. Codelets are weighted by urgency, but precisely which of the many active codelets is run next is determined randomly. The result of this architecture, once various weights have been suitably adjusted, is a program that exhibits the desired behaviour as an emergent property. In order to ensure that the approach has a chance of scaling up from micro-domains to 'real' cognition, it is important to make sure there are no 'brute force' searches, because combinatorial explosion soon makes such an approach infeasible; randomness is used so that not all paths, but only the currently most promising paths, are searched. The degree of fluidity -- how much the program is willing to explore 'weird' paths -- is controlled by a 'temperature' parameter: the hotter the temperature, the more random the search. But unlike 'Boltzmann' techniques such as simulated annealing, here the temperature is controlled by the program itself, by how 'happy' it is with the partial solutions it has built so far: the happier it gets, the more it concentrates on its current choices. Hofstadter calls this search technique the parallel terraced scan.

The book consists of some FARG research papers (despite being published academic papers, these are eminently readable), interleaved with Hofstadter's extensive commentary. I think it is best read in a rather short period of time, in order to get immersed in the philosophy, and to internalise the connections between the separate descriptions; I read it spread over about a fortnight. Since the individual papers are self-contained, this meant I ended up skimming the nth description of the architecture. But there's plenty more here to make it worthwhile. By the end of GEB I felt that, yes, cognition could be an emergent property of many 'mindless' agents, but still felt that Ant Hillary couldn't really be conscious. By the end of this latest book, I have a much clearer picture of the mechanisms that might make this possible.

Hofstadter's latest tour de force, executed with his customary panache and élan, is all about translating Clément Marot's 28-line, 84-syllable, 450-year-old French poem "A une Damoyselle malade" into English.

It is also a semi-autobiographical look at languages, linguistics, poetry, pattern, music, puns, playfulness, constraints, creativity, cognition, culture, AI, life, love and death.

How do you translate poetry between languages, between cultures? Should you strive for a literal translation, sacrificing rhyme and metre to get across the exact semantics? Or should you "try harder", and capture the music as well as the meaning? Hofstadter falls firmly in the second camp: poetry's "esthetically restricted medium" forms an essential part of the message, and by not translating the medium, you lose that essential part. He doesn't reserve his distaste just for rhyme-less, metre-less translations; he's not that keen on original works in blank verse, either:

When the poet is writing the original composition, the constraints of rhyme and metre can, paradoxically, help creativity. Being forced to make a rhyme, or have a certain number of syllables, or a particular stress pattern, can make the poet come up with a word or phrase that might not have been thought of otherwise, which can in turn lead to new ideas. The translator, however, is working under a double constraint, because the source text is fixed, and rhymes found in the new language have to reflect that. Some translators wimp out and use partial rhymes, some, like Nabikov with Pushkin's Eugene Onegin, scorn any attempt to translate the rhyme and metre, insisting that the meaning is all. Hofstadter demonstrates time and again that such an approach is neither necessary nor in any sense 'better', by exhibiting several different translations of classic poems, many of which have managed to preserve the music as well as the meaning.

TbdM isn't just about translating poetry. Hofstadter also discusses translating prose, and how much the cultural context should also be translated. This is not too difficult in technical prose, where there is little such context. But what about novels? Should an incident be translated 'literally', which might make the translated incident sound unintentionally exotic, or into an 'equivalent' incident in the new culture? And what if an incident is autobiographical? These questions are illustrated with examples from translating Hofstadter's own GEB into French and Chinese.

He also reckons that it could well be nigh on impossible to translate TbdM itself. That's not due just to the fact that it is about translation, and that there are snippets discussing English translations of foreign poetry (how do you translate the translations?). TbdM itself is 'poetry' in places, because of the medium constraints Hofstadter imposes on himself. Some passages have rhyme and metre, some are written without using a particular letter of the alphabet, some have a particular internal structure, some are complicated puns on other parts, the first words of each chapter are in a larger font so cannot have descenders; even the layout on the pages is strictly controlled so that there are no ugly page breaks in the middle of poem fragments. [When I quoted the passage above, I forced the line-breaks to occur in the same places, respecting Hofstadter's original choices; I didn't force a 'page break' after "marvelous", however.]

TbdM is written in what I believed to be my native language. But I had it made clear to me how much translation is about culture as well as language from the very first sentence: "Picture Holden Caulfield all grown up, now a university professor, writing a book about translation." Holden who? I read on, hoping for clues, until I got to "This sounds like poor, poor Salinger." Aha -- Salinger -- so it's a literary reference then -- presumably Catcher in the Rye? Pull a reference book off the shelf -- yes, Caulfield is the protagonist. Apparently, Catcher in the Rye is deeply part of American culture, one of those books nearly every American has read; but it's not part of my, British, culture. I'm vaguely aware it's about American teenage angst, but that's all. So the very first sentence doesn't translate for me! Then, later on, we get to the phrase "topless zoning laws" as an example of something a native reader would obviously understand, but would take a long sentence, or even short paragraph, of explanation to translate into any other language -- well, I needed that explanation in order to translate it into my English. I'm not a native speaker of my own language, it seems. So, this is rather fascinating: I'm learning something from this book, where that 'something' is what the book is all about, that a native speaker of American English would not!

Hofstadter goes on to contrast the curious 'invisibility' of translators, despite the enormous creativity and skill they bring to their job, with the almost 'cult of the individual' of musical performers [and conductors] where the identity of the original composer can definitely play second fiddle. When we applaud wildly and shout encore, we are asking for more from that performer, not more from that composer. [Although I do remember once hearing an interview with a famous cellist, where they said they preferred playing a particular concerto that wasn't recognised as first rate (possibly a Saint-Saens concerto) because then they knew any applause was for their performance, rather than for the composer. (Since I recall neither the cellist nor the concerto, obviously I'm using 'remember' in rather a loose sense here.)]

Hofstadter weaves his tales of translation into a story about human cognition. Thinking about good translations can help us understand some of the deep questions about cognitive patterns, about how we as humans understand and create, and about whether computers could ever do the same. There's yet another devastating attack on Searle's Chinese Room argument -- by building a lovely mental image of the sheer scale of a real such room, and the minute relative size of the human CPU in it. Yet he also attacks the hype and over-simplicity of some AI researchers 'dictionary' approach to machine translation. These programs still show no understanding of what they are translating, and so make ludicrous errors. [My own litmus test for AI is when a computer can correctly understand the newspaper headline SHELL FOUND ON BEACH, and notice that it's funny. This requires not only common-sense understanding, but 'meta-common-sense', to realise that the 'obvious default' meaning here cannot possibly be the intended meaning.]

And there's much more, too. For the most part, because of Hofstadter's playful style and wit, TbdM is a light-hearted read. But because of the overshadowing fact of the death of his wife Carol, it is also almost unbearably poignant in places.

Hofstadter thinks TbdM is the best book he's ever written. Personally, I would have to disagree -- I still reserve that place for GEB, which I found a sheer mind-blowing experience. TbdM is more anecdotal, reflective and contemplative, and I didn't feel I came away with my paradigms quite so severely shifted. (Or maybe that says more about me and the passage of nearly 20 years between reading the two?) But second best from Hofstadter is still streets ahead of much of the competition.

Hofstadter is back exploring the perennial theme of consciousness. His main theme is that consciousness emerges in a system that is powerful enough to have "strange loops" of a certain kind: a sort of self-referential, self-modelling, self-referential capability. The first such strange loop is Gödel’s discovery that integers are powerful enough to be used to represent (at a different level) statements about themselves. Hofstadter builds on this as an analogy to the way conscious beings can think about and represent themselves. And, in his usual way, he weaves in lots of word play, music, and conversations along the way.

The death of his wife Carol is as prominent here as it is in Le Ton beau de Marot, but now it is being used as a way to discuss how the patterns that in some sense are people can also exist (as low resolution copies) in other people’s heads, and so how "I"-ness is subtly distributed.

Towards the end, discussing the locus of "I"-ness, he engages in "intuition priming" fantasies about teleportation, and copying people. He mentions the philosophical literature in the shape of Derek Parfit’s Reasons and Persons (a book now added to my wish list: one book read resulting in one more to read; at least I haven’t fallen further behind!), but none of the extensive recent science fiction literature. I get a much better feel for what this all might be like from reading Wil McCarthy’s Queendom of Sol series, and Charlie Stross’s Accelerando universe tales, among others, with their characters happily copying and remerging, backing up and restoring. And this all fits with Hofstadter’s own discussion that "I"-ness is an emergent function of one’s past experiences. It is difficult to understand how "dying and being restored from backup" might feel from a philosophical thought experiment involving a single experience. But in these stories, backups and so on are routine: maybe if you lived in a society where this happened frequently, you really wouldn’t feel that "you" are dying and that an "other" will be restored and continue "their" separate life.

I find it refereshing to read something about consciousness that uses Gödel’s "loopiness" as a requirement for consciousness (and not a hard one to achieve, if integers can manage it!), not that uses Gödel’s proof as an argument against machine consciousness. Nevertheless, I haven’t given this as high a ranking as earlier Hofstadter works. It’s true as he says, that although there is a degree of continuity, the "I" of today is not the same person as the "I" of yesterday or the "I" of tomorrow. I today am not the same person who read GEB 25 years ago, and Hofstadter today isn’t the same person who wrote it. We’ve both come a long way, and things are more complicated. But today’s Hofstadter’s work is still definitely worth reading – and thinking about.