Prediction, Theories of Truth

I was trying to explain this argument to someone a few days ago, when we both suddenly saw the problem with it:

In outline, it is the argument that a comprehensive 'predicting machine' would have to incorporate a theory of truth for the language of any people whose behaviour it was going to predict, and so it's workings would be incomprehensible - unintelligible - to them.  It would have to be a 'black box'.

The problem is this:

The machine could predict behaviour - physical movements etc. - without intentional content.  In fact, if it was a machine which conformed to the traditional deterministic metaphors, it would be restricted to doing only this.  Someone examining the output of the machine might interpret it as having content - 'I know what these sounds mean', or 'I know what is described here - it is a person who is thinking this'.  But this interpretation would be made now - in the conversation of the interpreter.  It would not be a necessary component of the raw output.

If the output were to have semantic content, the predictor could not be just a mechanical device.  A semantic predictor would need to produce interpreted output, in the language of the user.  It is this kind of predictor - an 'oracular' predictor - which would need to incorporate a theory of truth for the language of its subjects, but a predictor of this kind could not be rendered in rule based terms - because it could not be a 'machine'.  It would have to be capable of normative judgements.  A prediction such as 'Mary will fall in love with John, and will tell him' could be fulfilled in a number of physically quite distinct ways.  The oracular predictor would need to know what our interpretive decisions would be in order to reliably make such a statement.

A computer which renders its outputs in a natural language can look like an oracular predictor, but examining its internal processes partly undermines this illusion.  We would see how it 'appeared' to speak, and what the nature of our 'conversation' with it was.  Whether, in some future, people might continue to attribute interlocutor status to objects such as this will be a matter for them.

What may bring them closer to this attribution would be a 'mechanical' understanding of how biological human beings speak - and this is some way away.  It is likely that the models will be so complex that few will understand any part of them, and no-one will understand them fully.

In this kind of world the metaphor of universal determinism will seem absurd - it will seem to be based on an astonishingly naïve conception of 'science' as something that could be grasped or managed by individual human beings.  The world will seem to be made up of 'black boxes' ... maybe not such an unfamiliar world after all.