Friday, June 3, 2011

Brian Christian's "The Most Human Human"

This was an amazingly entertaining book. It was a grand tour of knowledge dressed up as reportage about competing at the Turing Prize as a contestant trying to convince judges you are human while computers do their best to convince the judges that they are in fact "the humans". Like a good Hollywood move, the guys in the white hats win, Cristian comes out... spoiler alert!... triumphant and wins the "most human human" prize.

As a tale of prepping to present himself as a "genuine human" in the Turing test for the Loebner Prize, this book presents an interesting and entertaining tale. The author delves into a richness of subject matter that is unexpected. He explores what it means to be a real person and just how you could use language to try to convince a blind judge that your are really a "human". Getting there involves many detours into arcane specialized knowledge.

What I liked about the book was the amazing depth of knowledge of many fields served up as a charming story in the context of prepping and competing in the Turing test. I actually learned many new things in philosophy and computers, my two fields of expertise, from reading this book. I also learned interesting bits about psychology, linguistics, chess, poetics, and poetry. All this learning while being cleverly entertained!

I learned of entropy, lossless and lossy compression, of aposiopesis, synecdoche, and enthymemes. of computability and complexity theory, of the chatterbots ELIZA and PARRY, and so on. Delightful stuff.

By the way, he points us at the "Shannon Game" and invites us to play it (click here). I can honestly say it is one of the most frustrating things I've ever done. It humiliates you because you think you know language and what letters to expect when, but as confident as I was, I came out a loser, a humbling experience.

Like a delightful musical composition, this book ends with a coda, an epilogue where he looks at another field of computers -- computer graphics -- and in less than a dozen pages riffs a variation on the theme of the book how the effort to present "real stuff" makes you much more critical and appreciative of the reality:
Being a computer graphics person brings with it, as most jobs do, a certain way of looking at and of noticing the world. My own poetry background, for instance, gives me an urge to read things against the grain of the author's intended meaning. I read a newspaper headline the other day that said "UK Minister's Charm Offensive." This to me was hilarious. Of course they mean "offensive" as a noun, as in the tactical deployment of charm for diplomacy purposes, but I kept reading it as an adjective, as though the minister's creepy unctuousness had really crossed the line this time. My friends in the police force and the military can't enter a room without sussing out its entrances and exits; for the one in the fire department, it's alarms and extinguishers.


Reflection and refraction are difficult to simulate on a computer. So is water distortion. So-called "caustics," the way that a glass of wine refocuses its light into a red point on your table, are particularly hard to render.

Reflection and refraction are also fairly computationally nasty because they have the habit of multiplying off of each other. You put two mirrors in front of each other, and the images multiply to infinity in no time flat. Light travels roughly 200,000 miles per second: that's a lot of ping-pong,and way beyond the where most rendering algorithms tap out. Usually a programmer will specify the maximum acceptable number of reflections or refractions and cap it there, after which point a kind of software deus ex machina sends the light directly back to the eye: no more bouncing.


I love these moments when the theory, the models, the approximations, as good as they are, aren't good enough. You simply must watch. Ah, so this is how nature does it. This is what it looks like. I think it's important to know these things, to know what can't be simulated, can't be made up, can't be imagined -- and to seek it.
Try the book. I think you will be surprised at how enjoyable it is.

No comments: