The Most Human Human (29 page)

Dave Ackley offers very similar confederate advice: “I would make up words, because I would expect programs to be operating out of a dictionary.”

My mind on deponents and attorneys, I think of drug culture, how dealers and buyers develop their own micro-patois, and how if any of these idiosyncratic reference systems started to become too standardized—if they use the well-known “snow” for cocaine, for instance—their text-message records and email records become much more legally vulnerable (i.e., have less room for deniability) than if the dealers and buyers are, like poets, ceaselessly inventing. A dead metaphor, a cliché, could mean jail.
⁵

In his 1973 book,
The Anxiety of Influence
, Harold Bloom argues that every poet has to, aesthetically, bump off their greatest teacher and influence to become great. To think of language this way brings huge implications for the Turing test. Take even the bots that learn from their human users: Cleverbot, for instance. At
best
it mimics language. It doesn’t, as Ezra Pound said, “make it new.”

As Garry Kasparov explains in
How Life Imitates Chess
, “In chess, a young player can advance by imitating the top grandmasters, but to challenge them he must produce his own ideas.” That is, one can get almost to the top of the chess world—the top two hundred players in the world, say—by merely
absorbing
opening theory. But to crack into the ranks above it requires a player to be challenging that very received wisdom—which all of those players take as a given. To play at that level, one must be
changing
opening theory.

Pound referred to poetry as “original research” in language. When I think about how one might judge the world’s best writers, I keep gravitating to the idea that we’d want to look at who changed the language the most. You can barely speak without uttering Shakespeare coinages, like “bated breath,” “heart of hearts,” “good riddance,” “household words,” “high time,” “Greek to me,” “live-long day,” the list goes on.

I wonder if bots will be able to pass the Turing test before they make that “transition from imitator to innovator,” as Kasparov puts it—before they begin not merely to follow but to lead. Before they make a
contribution
to the language. Which most of us don’t think about, but it’s part of what we do. “The highest form of warfare is attacking strategy itself,” says Sun Tzu. The great chess players change the game; the great artists change their mediums; the most important places, events, and people in our lives change us.

As it turns out, though, you don’t have to be Shakespeare to change your language. In fact, quite the opposite: if meaning lies even partially in usage, then you subtly alter the language every time you use it. You couldn’t leave it intact if you tried.

“Retarded” used to be a polite word; it was introduced to replace “idiot,” “imbecile,” and “moron,” which had once, themselves, been polite terms. Linguists call this process the “euphemism treadmill.” Ironically, to use “retarded” as a way of disparaging a person or idea is more offensive than to use “imbecilic” or “moronic”: terms ditched—for being too offensive—in
favor
of “retarded.” This lexical switch obviously has not succeeded in the long term. When White House Chief of Staff Rahm Emanuel in a 2009 strategy meeting dubbed a proposal that displeased him “retarded,” it prompted calls among prominent Republicans for his resignation, and (in lieu of resignation) a personal apology to the chairman of the Special Olympics. In May of 2010, the Senate’s Health, Education, Labor, and Pensions Committee approved a bill called Rosa’s Law that would strike “retarded” from all federal language, replacing it with “intellectually disabled.” The treadmill continues.

A similar process happens in reverse—the “dysphemism treadmill”—with offensive words; they gradually lose their harshness, and have to be replaced every so often with new abrasives. Some words that are today considered perfectly acceptable, even endearingly inoffensive, or quaint—e.g., “scumbag”—were originally quite explicit: in its original usage, the term meant “condom.” As recently as 1998 the
New York Times
was still refusing to print the word, as in, “Mr. Burton’s staff today defended his comments, including the use of a vulgarity for a condom to describe the President.” But increasing numbers of readers, unaware of the term’s etymology—in fact, only a handful of modern
dictionaries
include reference to condoms in the word’s definition—were left scratching their heads. By 2006, just eight years later, the paper nonchalantly included the word in a crossword puzzle (clue: “Scoundrel”), prompting outrage—but only among few. The word’s origins were news even to puzzle editor and
renowned word-guru Will Shortz: “The thought never crossed my mind this word could be controversial.”

Other treadmills exist: for instance, slang terms and baby names. Slang invented by an insider group gets picked up by outsider groups, creating the persistent need for new slang to reinforce the insider group’s cohesion. In
Freakonomics
, economist Steven Levitt chronicles the process by which baby names percolate through society, from the upper economic classes to the lower ones. Parents often want their child’s name to have a ring of success or distinction, and so they look to the names of slightly more successful families; however, this very process begins to deplete the name’s cachet, and so the demand shifts gradually and perpetually onto new “high-end” names.

Linguist Guy Deutscher charts two others in
The Unfolding of Language
. The first is the perpetual pull of eloquence and push of efficiency. As he notes, the phrase “up above” has been compacted and elaborated so many times that its etymology is the hopelessly redundant “up on by on up,” and likewise some French speakers now say “au jour d’aujourd’hui”: “on the day of on-the-day-of-this-day.” The second is the constant invention of new metaphors to capture new facets of human experience in language—meanwhile, familiar metaphors are passing, through sheer use, from aptness to popularity to cliché. From there, the fact that it
is
a metaphor is slowly forgotten, the original image at the heart of the term becoming a mere fossil of etymology. For instance, Latin speakers needed a term to describe the relationship they had with their dining partners, the folks they broke bread with: the custom of simply calling such people one’s “with-breads,” or (in Latin) “com-panis,” caught on, a phrasing that eventually became our word “companion.” Likewise, as misfortunate events were believed in the sixteenth century to have astrological roots, speakers of Old Italian took to calling such an event a “bad-star,” or “dis-astro”: hence “disaster.”

The language is constantly dying, and constantly being born. English poet John Keats asked that his tombstone read simply, “Here
lies one whose name was writ in water”: a comment on the ephemerality of life. In the long run,
all
writing is in water: the language itself changes steadily over time. All texts have a half-life of intelligibility before they must be resuscitated through translation.

Language will never settle down, it will never stabilize, it will never find equilibrium. Perhaps part of what makes the Turing test so tricky is that it’s a battle on shifting ground. Unlike chess, with its fixed rules and outcomes, language—ever changing—is not amenable to being “solved.” As ELIZA’s creator, Joseph Weizenbaum, writes, “Another widespread, and to me surprising, reaction to the ELIZA program was the spread of a belief that it demonstrated a general solution to the problem of computer understanding of natural language. [I have] tried to say that no general solution to that problem [is] possible, i.e., … even people are not embodiments of any such general solution.”

You can’t take the temperature of a system without the thermometer itself becoming part of the system and contributing its
own
temperature, in some degree, to its reading. You can’t check a tire’s pressure without letting some of that pressure out—namely, into the gauge. And you can’t check a circuit without some of its current flowing into the meter, or vice versa. As Heisenberg famously showed, measuring an electron’s position, by bouncing a photon off of it, perturbs the very thing you sought to measure through the act of measurement. Scientists call this the “observer effect.”

Likewise, you can’t ask a friend if they’d like to go out to dinner without implying the extent to which
you
want to go out to dinner, and thus biasing their answer. Polling studies and eyewitness testimony studies show that the wording of questions biases someone’s response—“About how fast were the cars going when they collided into each other?” produces lower estimates than “About how fast were the cars going when they smashed into each other?” Asking
“Do you approve of the job the president is doing?” receives many more affirmatives than asking “Do you approve of the job the president is doing, or not?” The
order
of questions matters too: asking someone about their overall life satisfaction and then their financial satisfaction produces some limited degree of correlation, but asking
first
about their finances and
then
about their life overall magnifies that correlation tremendously.

Computer programming is largely based on the “repeatability” of its responses; as most programmers can attest, a bug that is unrepeatable is also for the most part unfixable. This is part of why computers behave so much better after a reboot than after days of continuous use, and so much better when first purchased than after several years. These “blank-slate” states are the ones most encountered, and therefore most polished, by the programmers. The longer a computer system is active, the more unique its state tends to become. By and large this is true of people too—except people can’t reboot.

When I debug a program, I expect to re-create the exact same behavior a number of times, testing out revisions of the code and undoing them where necessary. When I query a computer system, I expect not to alter it. In contrast, human communication is irrevocable. Nothing can be unsaid. (Consider a judge laughably asking the jury to “forget” a piece of testimony.) It is also, in this way, unrepeatable—because the initial conditions can never be re-created.

“Hold still, lion!” writes poet Robert Creeley. “I am trying / to paint you / while there’s time to.” Part of what I love so much about people is that they never hold still. As you are getting to know them, they are changing—in part due to your presence. (In conversation with one of these PARRY-style personalities, or reading Racter, or watching a video demonstration of a bot in action, I have quite the inverse feeling. I can’t get the damn thing to
move.
) In some sense my mind goes to Duchamp’s
Nude Descending a Staircase, No. 2
—a series of quick, overlapping sketches of a thing in motion, creating a kind of palimpsest, which scandalized a public accustomed to realism in portraiture. “An explosion in a shingle factory,” wrote the appalled
New York Times
critic Julian Street, and the piece became a lightning rod for outrage and mockery alike. Somehow, though, there seems something profoundly true (and “realistic”) about a human subject that refuses to sit still for the painter, who must aim to capture their essence via
gait
rather than
figure
.

Part of what we have invented the Turing machine–style digital computer for is its reliability, its repeatability, its “stillness.” When, in recent years, we have experimented with “neural network” models, which mimic the brain’s architecture of massive connectivity and parallelism instead of strict, serial, digital rule following, we have still tended to keep the neuron’s amazing plasticity in check. “When the [synaptic] weights [of a network of virtual neurons] are considered constant (after or without a process of adaptation) the networks can perform exact computations,” writes Hava Siegelmann. Virtual neurons can be controlled in this way, with strict periods of time in which they are permitted to change and adapt. The human brain has no such limits, owing to what neuroscientists call “synaptic plasticity.” Every time neurons fire, they alter the structure of their connections to one another.

In other words, a functioning brain is a changing brain. As Dave Ackley puts it, “You
must
be impacted by experience, or there is no experience.” This is what makes good conversations, and good living,
risky
. You can’t simply “get to know” a person without, to some degree, changing them—and without, to some degree, becoming them.

I remember first coming to understand that owing to the electrically repellant properties of the atom, matter can never actually
touch
other matter. The notion brought with it the icy chill of solipsism: the self as a kind of hermetically sealed tomb.

Sort of
. Someone may not ever quite be able to get to the outside of you. But it doesn’t take much—merely to be perceived, or thought of, alters the other’s brain—to make it
inside
, to where the self is, and change something there, however small.

Another way to think about it, as you levitate and hover around the
room on an angstrom-thick cushion of electromagnetic force, is this: You will never touch anything, in the sense that the nuclei of your arm’s atoms will never knock against the nuclei of the table’s—for whatever that would be worth. What
feels
like “contact” is actually your body’s atoms exerting electromagnetic forces on the table’s atoms, and vice versa. In other words, what appears to be static contact is actually dynamic interaction, the exchange of forces.

The very same forces, by the way, that your body’s atoms are exchanging with each other—the ones that make you whole.