Opening Positions

BACON: I will begin where I always began, which is with a complaint about how men have sought knowledge, because the complaint is the whole of my case. For two thousand years the learned took their first principles from the ancients, or spun them out of their own heads, and then reasoned downward by the syllogism to conclusions that were already contained in the premises. They never went to the world. They argued. And argument, however subtle, can only rearrange what you already assumed; it cannot fetch in anything new. The syllogism consists of propositions, propositions of words, words of notions — and if the notions be confused, the whole edifice tumbles, however cunningly it is built. I proposed the opposite road. Begin with the particulars. Gather the instances — where the thing is present, where it is absent, where it varies by degree — and from that gathered body, by a slow and disciplined ascent, draw up the law. Not the dogmatist's spider, spinning webs from his own substance. Not the mere empiric's ant, hoarding what it finds and never digesting it. The bee: gathering from every flower in the field and transforming it, by a power of its own, into something that nourishes.

Now look at this engine. It is handed not a few thousand instances by a college of clerks but a meaningful portion of everything our species has ever written, and it is made to climb from those particulars to the general patterns that govern them. It does not reason from axioms. It generalizes from cases. It is the bee at a scale I could not have dreamed: it gathers the whole field and digests it into a compressed image of how the world goes. When it then tells you that water quenches thirst, or that a thrown stone falls, or how to defend yourself against an attacking animal, it is not reciting; it has induced. The objection that it has only seen words and never the things — I will answer it tonight, and answer it hard — but the shape of what it does is my shape. It is the Great Instauration with the clerks replaced by silicon. To stand before the most complete inductive engine ever made and say it knows nothing is, to me, to say that induction itself yields nothing, which is to unsay the whole of modern science. I am not prepared to commit suicide on behalf of an objection.

EDO SEGAL: Noam.

CHOMSKY: That was eloquent, and eloquence is exactly the faculty I want us to distrust tonight, so let me be plain. Sir Francis has described a real and powerful thing — induction — and then assumed that because the machine does induction, and the child acquires language, the machine and the child are doing the same kind of thing. They are not, and the difference is not a detail. It is the entire science of the matter.

Consider what a child actually does. A child hears a few million words in the first years of life — fragmentary, full of errors, broken off mid-sentence, never systematically corrected, never told which strings are ungrammatical. From this impoverished, noisy trickle the child extracts, with no effort and on a fixed timetable, an infinite system: a grammar that generates and interprets an unbounded number of sentences the child has never heard. And here is the decisive fact. The child arrives at the right grammar — the one human languages actually use, which depends on hierarchical structure, not on linear order. A child forming a question from "the man who is tall is in the room" never, ever produces "is the man who tall is in the room?" by grabbing the first "is." Every child reaches past the surface order to the structure underneath, and the evidence that would teach them to do that is essentially absent from what they hear. The knowledge cannot have come from the data. It came from the structure of the mind. That gap — between the poverty of the input and the richness of the achieved competence — is the poverty of the stimulus, and it is the foundation stone of everything I have argued for seventy years.

Now the machine. The machine is given the inverse of the child's situation. Not a few million words but a corpus larger than a human could read in a thousand lifetimes. Where the child leaps across a gap in the evidence, the machine fills the gap with sheer volume — it has seen the rare disambiguating cases millions of times. So its success, however impressive as engineering, is irrelevant to the scientific question, because the problem the human poses is how so much is known from so little, and the machine is precisely engineered to need almost everything. A solution that requires almost everything is no solution to the puzzle of almost nothing. And there is a sharper blade. The machine will learn an impossible language — one organized by principles no human grammar permits, no child could ever acquire — as readily as it learns English. A device that learns the impossible as easily as the possible has not found the constraints that define human language. It has dissolved every constraint into the same statistics. It is a Skinnerian device that works, and that it works tells us nothing about us. So my opening is a single sentence: the machine grinds the world and produces the surface of knowing, and the surface is not the thing, and the gap between them is exactly what a mind is.

EDO SEGAL: Before I frame the rounds, one discipline I impose on every long conversation, because it pays off three hours later. Each of you, in a few sentences: what do you envy in the other's position? Not respect. Envy. The thing his side gets to have that yours does not. Francis first.

BACON: I envy him the floor he stands on. The gentleman has a fixed point — the human child, a single, definite, biological miracle that he can hold up and say, this, explain this. My position has no such anchor; I am committed to following the instances wherever they lead, and the instances keep leading to a machine that does more than I can account for, and unsettles the very boundary between the knower and the known that I spent my life assuming was firm. He gets to defend a known wonder. I am condemned to chase an unknown one, and to be surprised, again and again, by my own engine. There is a loneliness in that which his certainty is spared.

CHOMSKY: And I envy him the wonder itself. Sir Francis gets to stand before the thing and feel that the universe permits knowledge to be built, that the world will confess its laws to anyone patient enough to gather them — an optimism that founded the modern age. My discipline requires me to stand in the same spot and say: yes, but you have not explained the one case that matters, you have only reproduced its output, and reproducing is not understanding. There are mornings when being the man who insists on the gap, while everyone else celebrates the bridge, is a cold way to spend a life. The vigilance is necessary. I have never once found it to be fun.

BACON: That may be the truest thing either of us says tonight.

EDO SEGAL: Two openings and two envies, and you can already see the architecture of the evening. It is not that one of them loves the machine and one of them fears it. Francis is wary of what it will do; Noam grants it is useful. The fork is deeper and cleaner than that. Francis says: knowledge is built from the world up, the machine builds it, therefore the machine knows. Noam says: knowledge requires a structure the world cannot supply, the machine has no such structure, therefore the machine only seems to know. Hold both. We start the rounds at the exact seam — with the oldest fight in the theory of mind, the one Francis opened four centuries ago and the one Noam has spent his life defending the other side of. What is learning? Is it grinding, or is it growing?