An interview with myself; power of the dialectic:

Q: You seem a little distracted.

A: It’s spring, you know. April is the something month. The whole world is going nuts, blooming and flowering and going places, and especially here in California, one feels a little bit behind the times, even if things are going well, just because everything starts so damn early.

Q: Are things going well?

A: Yes, and no. I mean, I don’t really have anything to complain of. I have a car, a place to live, plenty of clothes, food, friends. Work to do during the day that’s certainly a lot more interesting, to me at least, than working in an office or doing some menial task. All the same, there’s that sense of confinement, the knowledge that I could be doing something more important.

Q: What would you rather be doing?

A: Assuming I didn’t just go running off after my true love, well. I’d like a lot more time to read and write. Enough to follow some kind of well-trodden path - I’m thinking of Whitehead, Hofstadter, Hilbert, maybe some Popper - and for rabbit trails. I just bought a bilingual edition of Paul Celan, and I could really imagine myself sitting on some ruin in the Mediterranean with just that book for at least a month.

Q: No dictionary?

A: No. Whenever I learn languages I tend to gravitate towards the formalism - the syntax, grammar, and dictionaries. Dictionaries can’t tell you what a word means except in terms of other words, and when the only other words you can use are in some totally different language it’s a lot like, well, forcing Beethoven to play Mozart or Shostakovitch. Of course, to some extent the formalism is unavoidable. But I wonder about mappings. If you learned German from a sailor because you got rescued off a desert island by a German crew. Or from Celan, because he rescued you from drowning in some other way.

Q: What do you mean by mappings?

A: Now you’re asking for it (laughs). The most basic kind of map is a dictionary, which lets you look up an English word and find the German equivalent. But often there’s multiple equivalents, and how do you know which one to use? That’s why machines are so bad at translation, because they don’t know what the English word means, so they can only treat it as a generic noun in some case or declination or whatever, and find the formal equivalent in the German. Everyone’s played that game with Babelfish, turning some piece of text into French and then back into English.

Q: The vodka is strong and the meat is rotten?

A: Yes, although that particular story’s somewhat apocryphal, and it certainly didn’t originate with Babelfish. Spirit gets turned into vodka, flesh into meat. The only way an MT could get really good, solve the whole class of “howlers”, would be if it could take spirit, and turn it into an unequivocal symbol for which exactly one formal equivalent exists. That’s an AI problem, which I’m interested in, but only because I think this problem with formalism indicates that the way we process language is fundamentally not formal. So as a thought experiment, take the opposite type of map. Instead of trying to find a one-to-one correspondance between all possible symbols, start the map with a singularity - a line in Celan, for instance - from which the entire set of meanings can be derived.

Q: I can’t tell if that’s at all sympathetic to what Celan was trying to do.

A: Do you mean that I can’t have Celan and my language experiment at the same time? No, not in any scientific sense. But the experiment isn’t scientific, it’s basically an existential question. It’s like… Well, look, nothing can be learned without love. If I don’t care about what Celan says, I won’t sit on an island for a month with him. I was going to say, it’s like learning a language by spending your nights with a woman who speaks it. And then it occurred to me that maybe you can’t learn anything unless love drives you. Or, at the very least, without committing yourself to something irrevocable. That makes you need to communicate.

Q: Is that an AI problem?

A: Well, I guess I’m on the record now. I’m not an optimist, I don’t think a machine that embodies a formal system can learn in the same way as humans. I’m not saying it can’t learn, mind you, and maybe it’ll be able to pass a Turing test of some kind. But whatever kind of machine it is, it’s going to be something utterly alien to us. It’s going to perceive time and reality differently, and while we humans might be able to ask it to do things or chit-chat about the weather, the important things that go on in its head or gearbox or whatever, are going to be out of our ken. That’s pretty exciting, actually.