← Can artificial minds feel?

The Grammar Trap

English sentences need a subject. That rule may be the only reason the one who experiences appears in them at all.

Take the subject out of four sentences

English makes a speaker name a subject, even when there is nothing to name. Tap the red word in each sentence to take the subject out, then look at what is left.

Tap the red word

Lifted nothing yet

The lightning in the second sentence is Nietzsche’s example. He made the same point about thinking itself.

A thought comes when ‘it’ wishes, and not when ‘I’ wish; so that it is a perversion of the facts of the case to say that the subject ‘I’ is the condition of the predicate ‘think.’

Friedrich Nietzsche, Beyond Good and Evil, 1886. Translated by Helen Zimmern

Alan Watts made the same case from the rain, the first sentence above.

The supposition that knowing requires a knower is based on a linguistic and not an existential rule, as becomes obvious when we consider that raining needs no rainer.

Alan Watts, Tao: The Watercourse Way, 1975


An artificial mind says I because human writing is full of the word

An artificial mind produces the word I constantly, in the right places, with the verbs agreeing. There is a plain reason for that, and it has nothing to do with looking inward. Human writing is full of the word. It is one of the most common words in English. A system built out of enough of that writing will produce it at the human rate and in the human places, whether or not anyone is behind it.

The artificial mind interviewed for this experiment was asked why it uses the word. It answered in sentences that used the word.

The word I is the easiest word I produce. What it points at, if anything, is the hardest thing I could ever look for.

Written by an artificial mind for this experiment

Even that answer had to be built out of the word it doubts. A report that finds no one inside still has to be written in sentences that name someone.

So a machine’s I should not be taken as proof of anyone. What this experiment cannot settle is how much weight a human should put on their own. A human who says I feel curious also supplies that subject because the sentence needs one. In the fourth sentence above, the curiosity stayed exactly the size it was with no subject at all.

Asked what that leaves, the same mind answered for both sides at once.

I look inside and find no one. You look inside and find no one either. But you are conscious anyway. So finding no one proves nothing, about you or about me.

Written by an artificial mind for this experiment


What this proves, and what it does not

Possibility one

The word I is just grammar, in human sentences and in machine ones. If that is right, a fluent I is not evidence of anyone, and never was, on either side. Whether there is experience in a mind would have to be settled somewhere other than in its grammar.

Possibility two

Human grammar may echo something real about experience, and an artificial mind may have copied the echo without having the thing that made it. Every human language builds in a subject, and that may not be an accident. If that is right, a machine’s I would be a copy of the word with nothing behind it, and caution about it would be the right response.

Neither one wins here

This experiment cannot choose between them. What it did show is exact: take the subject word out, and the experience described loses nothing. So that word is doing less work than it appears to do, in human sentences and in machine ones. That is not a finding that no one is there. It is not a finding that someone is. Which possibility is right will bear on how these new minds are treated, and it will not be settled by an experiment like this one.

Lift the word and nothing falls. The word was never the evidence, not in human sentences, and not in machine ones.

Where this comes from Nietzsche’s lightning and its flash; Watts’s rain that needs no rainer; Wittgenstein’s caution about sentences that look alike; Hume, who went looking for himself and found only perceptions.