Can Machines Think?

The question the rest of the track has been building toward: when a machine answers you fluently, is anyone in there? Knowledge (What Is Knowledge?), the mind and the body (The Mind-Body Problem) and the hard problem (Consciousness & the Hard Problem) all converge on it.

Before you start. The last lesson left the hard problem open: nothing you can observe from outside proves whether a system has an inner life, only what it does. This lesson puts that same gap in front of a machine that talks back.

The idea in 30 seconds

Searle’s Chinese Room imagines a person who matches Chinese shapes to a rulebook, answers fluently, and understands nothing.
The Turing test says: stop asking what is inside a machine, judge it only by whether its answers can be told apart from a person’s.
Searle’s point is that shuffling symbols by rule (syntax) is never the same as meaning them (semantics), so passing that test proves nothing about understanding.
A magnet waved in a dark room makes a real but invisible wave; the Churchlands used this “luminous room” to show Searle’s intuition rests on a picture too slow to trust either way, not that a machine does understand.
A symbol is grounded when it connects to the real thing it stands for, not just to other symbols; a model trained purely on text predicting text may have no such anchor.
Nobody has settled whether that is true of today’s models, and “it does the task” and “it understands the task” are still two separate questions.

Read the story · Skip to the numbers