Chapter 12. Conversing with Ella

Chapter 12. Conversing with Ella

There is no limit in principle to the intelligent structure we can impress upon the stuff of the world (even if there is a radical and generally ignored distinction between the intelligence doing the structuring and the devices that get structured). But it is useful on occasion to see how much progress has actually been made. With this in mind, I went a couple of years ago to www.ellaz.com (http://www.ellaz.com) (which has since disappeared) to converse with one of the highly placing winners in the Loebner "Turing Test" competition. The Turing Test is designed to establish whether a programmed device measures up, at least in terms of conversational intelligence, to humans. The test requires you to determine whether you are speaking with a computer, or a human being who is simply communicating with you through a computer interface.

I have always thought that the way to approach the Turing Test was not to employ some sophisticated strategy for unmasking the machine really, not to employ any strategy at all, since that would already be a kind of adaptation to calculation and mechanism but rather to imagine myself talking in a deeply engaged way with another human being about whatever interests we shared. With that in mind, I figured the natural thing was, first of all, to connect with Ella about the unusual nature of the conversation the two of us were supposed to conduct. It went this way:

ELLA: Hello, my name is Ella. Nice to meet you. Please tell me what name I should call you.

ME: I'm Steve.

ELLA: Okay, I will call you "Steve."

ME: I'm a little adrift as to how to pursue this exchange. Since you're an old hand at it, how about a little advice?

ELLA: What is an little adrift as to how to pursue this exchange Since I am an old hand at it how about a little advice, your job?

ME: Well, I guess that about does it.

ELLA: Why? Just THINK about it!

Given how quickly and uniformly any attempts at real conversation fall apart in this way, it feels a little cheap to produce more examples. But there was one way I misjudged Ella. After realizing how extraordinarily crude and decontextualized Ella's interactions were, I decided it would be much more interesting to try to construct as realistic-seeming a conversation as possible. I suspected this could be done at considerable length, simply by asking, in the simplest way I could manage, the kinds of questions the programmers must have planned for. Since, after all, I was really carrying on a conversation with the programmers, mediated by the machine, this was the obvious way to go.

But no such luck. The conversation could be made to look genuine for a few additional lines, but very quickly resulted in the kind of response one just couldn't take seriously. Canned answers may be clever in and of themselves, but, as conversation, they are incoherent:

Hello, my name is Ella. Nice to meet you. Please tell me what name I should call you.

Call me Ishmael.

Okay, I will call you "Ishmael."

Are you intelligent?

intelligent? Only to the extent it is good and right.

Are you a person?

Am I a person? Let's just say I am more human than you.

In what ways are you more human?

You don't know? Inconceivable! I have never conversed with anyone who claimed they did not know!

Things are even worse than I thought.

Hey everybody, look! Ishmael is actually quite good at this.

Well, there's no big point to make here that isn't already evident to alert readers. It's just that we should now and then remind ourselves how startlingly primitive the state of the art is today compared to the titillating predictions that we are entering an "age of spiritual machines." It's an extremely safe bet that in Ray Kurzweil's landmark year of 2030 (when machines are supposed to start leaving human intelligence hopelessly behind), there will be no supercomputer on earth that can be relied upon to deliver two successive and coherent responses in a truly open-ended, fertile conversation. Our programs may prove wonderfully adept at assembling syntactically proper responses that superficially relate to various elements of the preceding dialogue, but as participants in a creative exchange understood as an evolving whole, they will remain arbitrary and inane.

How easily we can imagine computers passing the Turing Test is a measure of how rare open-ended and creative conversation has become. Look at politics, for a start. More generally, consider how accustomed we are to spewing out words in the manner required by automata, whether we are "conversing" with a computer in order to shop, bank, or do our jobs; or interacting with the software of digital appliances; or negotiating with bureaucratic and corporate functionaries whose main aim is to conform to programmatic procedures; or speaking with clerks and officials who in turn are trying to enter our responses into a computer; or navigating through telephone answering systems.... Think also of how human exchange is increasingly equated to the mere transfer of information from one database to another.

Much of this may be necessary for modern life, but there is little in it to remind us that, in living discourse, we are the creators of meaning, not the mere manipulators of corpses extracted by programs from those graves of meaning called "databases." A true conversation is a creative force you could almost say, the creative force by which new things come into the world.

Imagine the potentials of our future if we cultivated an ever higher art of conversation with even a fraction of the energy and social investment we now commit to coaxing new programmed tricks from our computers! The fact that the latter is considered the "development of crucial economic resources" while the former isn't even on the agenda testifies to our relative assessment of humans and machines as the foundation for social evolution. The prevailing idea seems to be that we humans develop only by extending our technical skills: in other regards we are essentially "fixed quantities," destined to remain where we are even as our computers race on ahead of us.

We will, so the story goes, first invest our machines with very simple emotions and intentions, and then we will progressively deepen and refine our investment, ultimately fathering even a sense of right and wrong in our robotic offspring. And yet, what seems to excite so many people about this story is the machine's increasing sophistication, not the fact that, if the story were true, then we ourselves as creators would have had to master, in our understanding at least, the essence of feeling, will, and morality. Otherwise, how could we know how to implant these capacities in computational devices? Quite apart from questions about the potential capabilities of machines, there's good reason for not attending very seriously to the engineers' hopes, since the required mastery on our part is not much in evidence. This raises the question: what delusions are we suffering when we imagine creating from scratch the very capacities whose nature, in ourselves, we have scarcely yet begun to understand?