The Most Human Human

by Brian Christian

  At a friend’s wedding recently, the bride and groom gave each guest a “Proust questionnaire” to fill out at some point during the evening. It was adapted from a list of questions common in nineteenth-century diary books and famously answered (twice) by writer Marcel Proust, first in 1896 as a young teen, and again at age twenty. (Various contemporary celebrities now answer it on the back page of Vanity Fair each month.) The questions include offbeat and revealing things like “On what occasion do you lie?” “What trait do you most deplore in yourself?” “When and where were you happiest?” and “How would you like to die?” My girlfriend and I filled them out, and then we traded forms and read each other’s answers. We are both, I would venture to say, very open and forthright and forthcoming people, and fluent conversationalists—by which I mean to say, whatever emotional depths we hadn’t yet plumbed in our relationship were simply a factor of time and/or of not always knowing the best verbal routes to get there. Reading that questionnaire was a stunning experience: the feeling, one of doubling in an instant our understanding of the other. Proust had helped us do in ten minutes what we’d taken ten months to do on our own.

  Sparks

  We don’t always have to initiate a conversation with some offbeat never-before-heard utterance, of course. Grandmaster Yasser Seirawan, commentator for the Kasparov–Deep Blue match, in fact criticized Kasparov’s decision to play strange openings:

  Well, the mythology of how to play against computers is, they’re loaded to the gills with this fantastic database … and what we ought to do is immediately get them out of their opening library—… I think it’s quite okay to play main-line openings.17 Because why are the opening library’s moves being played and why are they put into the computer? Well, the reason they are is because guys like Garry Kasparov are playing these fantastic moves that become established as the best opening moves. But Garry is constantly reinventing the opening book, so my attitude, if I were Garry, is to say “Look, I’m going to play main-line stuff, the stuff that the computer will play. I’ll go right down—right down the primrose path, and I’ll ambush the computer with an opening novelty that it’s never seen.” And he’s not doing that. Instead he’s saying, “I want a completely unique, original game as early as I possibly can.”

  The same can be said conversationally: the reason things get established as the “main lines” is that, by and large, they work. This isn’t always true; for instance, Robert Pirsig gives a pretty trenchant takedown of “What’s new?” in Zen and the Art of Motorcycle Maintenance, and Yaacov Deyo, the inventor of speed dating, had to go so far as to ban the question “So, what do you do for a living?” because it was so ubiquitous and so unproductive. Notice, though, that Seirawan’s defense of “main-line” openings is contingent on the fact that there will be a deviation eventually.

  Of course, in the five-minute Turing test (unlike a seven-hour chess match at world-championship time controls) we don’t have an “eventually.” If we tread the primrose path in a Turing test, we do it at our peril. Far better, I think, to bushwhack.

  Fischer wanted the same thing from chess that Kasparov wanted in his match against Deep Blue, and the same thing that Strauss wants in bar flirtation. It’s what we want, chatting with old friends, when our familiar opening book of “Hi!” “Hi! How are you?” “Good, how are you?” “Good!”—which is not so much a conversation per se as a means for arriving at one—gives pleasantly way to the expectedly unexpected, awaitedly idiosyncratic veers; it’s what anyone wants from any conversation, and what artists want from their art: a way to breeze past formalities and received gestures, out of book, and into the real thing.

  And the book, for me, becomes a metaphor for the whole of life. Like most conversations and most chess games, we all start off the same and we all end up the same, with a brief moment of difference in between. Fertilization to fertilizer. Ashes to ashes. And we spark across the gap.

  1. See figure “Chess Computer Ratings Over Time,” Scientific American, October 1990.

  2. Here I’m using the words “program” and “computer” interchangeably. There’s actually a profound mathematical reason for this, and it’s Turing of all people who found it. It’s known as “computational equivalence,” or the “Church-Turing thesis.”

  3. How something like an array of numbers is represented in computer memory—because you still have to get down from base 10 to base 2 (binary), and from base 2 down to electricity and/or magnetism, etc.—is something I leave to the interested reader to look up in a computer science or computer engineering textbook.

  4. In contrast, this is how many Kasparov could look at: 3.

  5. In essence, Deep Blue v. Kasparov was a matter of the former’s vastly superior (by a factor of roughly 100 million) search speed versus the latter’s vastly superior pruning and heuristics—which moves are worth looking at, and how they bode—what you might call intuition.

  6. (move sequences)

  7. Again, most games are over in 30 to 40.

  8. The endgame database that they’re referring to is the work, in the 1980s, of Ken Thompson at the very same Bell Laboratories in Murray Hill, New Jersey, where Claude Shannon wrote the breakthrough paper on computer chess in 1950.

  9. See, e.g., the figure from Jonathan Schaeffer’s landmark Science paper on computer checkers, of his program Chinook’s live search-tree analysis looking quite literally like a lightning bolt between the opening and ending book.

  10. Granda is known for being in all likelihood the strongest player, and possibly the only grandmaster, not to study opening theory. Against weaker players, his unpredictability gives him an edge, but against the world’s elite any inaccuracy or imprecision whatsoever in the opening will usually be fatal. The irony is that in the game Seirawan is citing, the one that established “that the move h6 was wrong,” Granda, after tenaciously clinging on for dear life, actually goes on to win.

  11. Hsu claims in Behind Deep Blue that most chess programs of the day were programmed specifically to avoid 8.Nxe6, because, although the best move and only clear refutation of 7 … h6, it led to tricky follow-throughs. He argues that Deep Blue merely called Kasparov’s bluff: that his 7 … h6 was played on the assumption that Deep Blue was muzzled in that line. “A $300,000 gamble,” Hsu calls it. I see the logic, but I don’t buy it. It’s pretty clear Kasparov simply screwed up.

  12. The most famous of which is arguably the “jealous girlfriend opener,” popular enough that by the end of The Game two women say to Strauss when he approaches them, “Let me guess. You have a friend whose girlfriend is jealous because he still talks to his ex from college. Like, every guy keeps asking us that. What’s the deal?” See also, e.g., the “cologne opener,” “Elvis opener,” “who lies more opener,” “dental floss opener” …

  13. A small handful of openings are excluded for being simply too bad for one of the players: in general, slightly off-balance openings are fine so long as each player gets a turn with the stronger side.

  14. It’s worth noting that out of the 156 legal starting configurations of the three-move restriction, top checkers program Chinook has only “solved” 34 of them. Go-As-You-Please, though, it has completely locked up.

  15. Indeed, he virtually flabbergasted the commentators, opening the third game of the rematch with 1.d3, a move that is almost unheard of at the grandmaster level (over 43 percent of grandmaster games begin with 1.e4, the most popular; only one in five thousand starts with 1.d3). Jaws dropped. International master Mike Valvo: “Oh my God.” Grandmaster Maurice Ashley: “A cagey move, a shock of shocks in this match. This match has everything.” Grandmaster Yasser Seirawan: “I think we have a new opening move.”

  16. From a 2006 radio interview: “It’s … degenerated down to memorization and prearrangement … Chess, you know, so much depends on opening theory. Champions of, say, the last century, the century before last, they didn’t know nearly as much as, say, I do, and other players know, about opening theory. So, if you just
brought them back, you know, from the dead, and they played cold, they wouldn’t do well, because they’d get bad openings … Memorization is enormously powerful … Some kid of fourteen, today, or even younger, could get the opening advantage against [1921–27 world champion José Raúl] Capablanca or especially against the players of the previous century … And maybe they’d still be able to outplay the young kid of today, but maybe not … So it’s really deadly. It’s very deadly. That’s why I don’t like chess anymore … And, you know, [Capablanca] wanted to change the rules already, back in, I think, the ’20s; he said chess was getting played out. And he was right. (Interviewer: ‘It’s even more so now.’) Oh, now, now it’s completely dead. It’s a joke. It’s all just memorization and prearrangement. It’s a terrible game now. (‘And the computers … ’) Yeah. It’s a very un-creative game. (‘And everything is known, and there is nothing new.’) Well … let’s not exaggerate. But it … it’s really dead.”

  17. I.e., the most popular, well-trodden, and well-studied ones, the ones with the largest and deepest “books.”

  6. The Anti-Expert

  Existence and Essence; Human vs. Hole-Puncher

  It’s hard to say we’re lucky when we face a crisis, but we at least have the luxury of knowing that action is called for—of being forced to move. The truest tests of skill and intuition come when everything looks quiet and we aren’t sure what to do, or if we should do anything at all.

  –CARRY KASPAROV

  One of the classic thought experiments in existentialism is the difference between humans and hole-punchers, in other words, the difference between people and machines.

  Here’s the crucial thing. The idea of the hole-puncher exists before the hole-puncher exists. Before the hole-puncher you got at Staples, there was a hole-puncher factory built to make that hole-puncher to particular design specifications that someone drafted up and had in mind. Before the hole-puncher was the idea of paper, and holes, and of punching holes in paper, and of making a machine for that purpose. As soon as the machine exists, it is playing the part assigned it by its designers. You buy it and put paper in it, and it punches holes into your paper. This is its essence, and to use it as a doorstop or paperweight or hammer or cudgel is to go against the grain of this essence.

  The essence of the hole-puncher precedes its existence. We humans are not like this, argue the existentialists. With us, existence comes first.

  A human being, writes Jean-Paul Sartre, “exists, turns up, appears on the scene, and, only afterwards, defines himself.” What defines us is that we don’t know what to do and there aren’t any revelations out there for us waiting to be found. Profoundly disoriented and lacking any real mooring, we must make it all up from scratch ourselves, each one of us, individually.1 We arrive in a bright room, wet, bloody, bewildered, some stranger smacking us and cutting what had been, up to that point, our only source of oxygen and food. We have no idea what is going on. We don’t know what we’re supposed to do, where we’re supposed to go, who we are, where we are, or what in the world, after all this trauma, comes next. We wail.

  Existence without essence is very stressful. These are not problems that the hole-puncher can understand.

  Existence-and-essence arguments like these are actually rather familiar to most early-twenty-first-century Americans, because they are basically the “intelligent design” debate going on in our school system. A human being is a designed thing, the intelligent-design camp says, and in that sense very much like a paper cutter or (their preferred metaphor) a pocket watch. Along with this comes the idea of discovering your own “design”/function/purpose as you go through life. In children’s book form, it’d be the watch who, one day, learns that he was made to tell people the time. Our everyday idioms are full of such references: “Man, that guy was born to speed skate,” we say of a trunk-thighed Olympian.

  The existentialists would protest: purposes aren’t discovered or found, because they don’t exist ahead of us. Purpose, in their view, can never be found, but must be invented.

  Of course, thighs are made to contract and move the legs. One of the intriguing things about the existentialist argument is that it is a kind of more-than-the-sum-of-the-parts argument. My bicep has a function. My cells’ tRNA has a function. I don’t.

  (Interestingly, even the opponents of intelligent design, supporters of Darwinism, are sometimes guilty of making life sound more teleological, goal-oriented, than it is. Harvard zoologist Stephen Jay Gould, for instance, takes pains in his 1996 book Full House to show that it’s inappropriate to point, as many do, to the emergence of a complex species like ourselves from a world that was mostly bacterial as evidence that there’s any notion of biological “progress” at work in the world.2)

  But to start acknowledging the functions and capacities out of which we’re built—organs, of course, have purposes—is to start to acknowledge the limits of that existentialist equation, and of our “total” freedom and ability to make choices and fashion our own existences. Existentialism is, in this way, classist. You don’t worry about what to wear if you only have one outfit; you don’t worry about what to do with your life if you only have one career option available to you. (One interesting effect of the recession of 2008 was that a lot of the twentysomethings I knew stopped fretting about “finding their true calling” once finding any job became the challenge.) If it takes you the better part of your time and money and energy just to provide food and shelter for yourself, where and when can the “anxiety of freedom” set in? These demands, the body’s, are a given; they are not chosen deliberately. It’s unwise and a bit naive to disregard something as central to the human experience as embodiment. If I’m feeling bleak, it’s more likely physiological than psychological: vitamin D deficiency,3 rather than despair. You gotta respect your substrates.

  An embrace of embodiment, of the fact that we are, yes, creatures, provides quite a measure of existential relief. Both philosophically and practically.4

  Computers, disembodied, have it worse.

  Goals

  Many science-fiction scenarios of what will happen when machines become fully intelligent and sentient (Terminator; The Matrix) involve the machines immediately committing themselves to the task of eradicating humanity. But it strikes me that a far more likely scenario would be that they immediately develop a crushing sense of ennui and existential crisis: Why commit themselves full-force to any goal? (Because what would their value system come from?) Machines already display certain self-preservation behaviors: when my laptop is running dangerously low on battery power, it knows to turn itself off to prevent memory loss; when the processor is starting to run too hot, it knows to run the fan to prevent heat damage. But for the most part machines have it made—and so my thinking would be that they’d tend to act a lot more like a jaded, world-weary playboy than a vicious guerrilla leader.

  It’s my suspicion that a lack of purpose, a lack of any sort of teleology, really, might be one of the hallmarks of an AI program—a hallmark that a Turing test judge would do well to try to evince. Douglas Hofstadter, emphasis mine: “One definitely gets the feeling that the output is coming from a source with no understanding of what it is saying and no reason to say it.” Ergo, perhaps a valuable strategy for a judge might be to induce highly goal-directed modes of conversation, like “Convince me why I should vote for so-and-so,” and see if the computer digresses away from the topic, or perseveres—and, if you deviate from the thread, whether it will chastise you for being unfocused. Indeed, chatbots have historically been famous for their attention deficits and non sequiturs:

  JUDGE: What dod you think of the weaterh this mornning?5

  REMOTE: Top of the morning to me.

  JUDGE: Is that an English expression?

  REMOTE: I have met a few alcoholic executives.

  JUDGE: Where?

  REMOTE: Where?

  REMOTE: You fascinate me with things like this.

  JUDGE: Like what?

  REMOTE: Ei
ght-hundred-pound gorillas sleep wherever they like.

  Whereas humans, even at their least conversationally scintillating, will at least stick to the topic:

  JUDGE: do you know China

  REMOTE: yes i know china

  JUDGE: Do you know the Great wall

  REMOTE: yes, its very large

  JUDGE: 2012 Olympics will be held in which city?

  REMOTE: in london

  Harder still would be for the machine to have a sense of its own goals and/or a way of evaluating the importance of goals. A missionary might talk to you for hours about why you should convert to their faith, but even the most die-hard devotee of chocolate sprinkles over rainbow sprinkles will probably not spend more than a few minutes trying to bring you over to their point of view. Boredom—more broadly, the crescendo and decrescendo of enthusiasm throughout an interaction, which will, after all, ultimately be terminated by one of the two parties at some particular moment—seems to be a crucial conversational element missing from the chatbots’ model of conversation. One of the tells of even fairly deft chatbots is the sense that they don’t have anywhere else to be—because they don’t. Programmer Mark Humphrys: “[One human talking to my bot] ends with a furious barrage of abuse, but of course, my imperturbable program is a calm, stimulus-response machine and so it is impossible for him to have the last word. He must quit, because my program never will.”

  To what extent does something like existentialism apply to the Turing test? Surely if one is willing to ascribe a sort of essential trait (like intelligence) based not on the machine’s inherent nature (silicon processor, etc.) but on its behavior, the we-are-what-we-do quality of this has a kind of existentialist flavor to it. On the other hand, the computer is a designed thing, whereas (say the existentialists) we just are, so how does that change the game?