Mary's room

What it tests Does Mary learn something new when she first sees red, and if so, does that show that physical facts are not all the facts?

The case

Mary has never seen colour. She lives in a black-and-white room, reads black-and-white books and watches a black-and-white screen, and she has become the world’s greatest expert on colour vision. She knows which wavelengths stimulate which cones, how the signals are processed, and which words people utter when they see a ripe tomato. Suppose she knows every physical fact there is to know about seeing red.

Then she is released, and sees a red tomato.

The argument

Frank Jackson, who published the case in 1982, put the intuition plainly: it seems she learns something, namely what it is like to see red. If so, the argument runs:

1. Before her release, Mary knows all the physical facts about colour vision.
2. On her release, she learns a new fact about colour vision.
3. So not all facts about colour vision are physical facts.

If that is right, physicalism, the view that everything there is is physical, is false. Jackson’s target was the claim that a complete physical science would be a complete account of the mind.

The replies

The ability reply. David Lewis and Laurence Nemirow argued that Mary gains no new fact but new abilities: to recognise red, to remember and imagine it. Knowing what red is like is know-how, not knowledge that something is so.

The acquaintance reply. Mary becomes acquainted with a property she already knew everything about. Acquaintance is a different relation from knowing a fact, so nothing non-physical follows.

The phenomenal concept reply. Mary learns a new way of thinking about an old fact. She acquires a concept that can only be gained by having the experience, and with it she can think a thought she could not think before, about a physical state she already knew. This is the most widely held physicalist response.

Deny the intuition. Daniel Dennett argued that we cannot really imagine knowing all the physical facts. If Mary did, she might recognise red at once, and the sense that she must be surprised is a failure of imagination.

Jackson himself later became a physicalist and rejected the conclusion, though he continued to regard the argument as one physicalism must answer.

Mary and the zombie

Mary’s room and the philosophical zombie are the two most discussed arguments against physicalism, and they work in different ways. The zombie argument asks whether the physical facts could be just as they are without consciousness. Mary’s room asks whether knowing the physical facts gives you knowledge of what consciousness is like.

Why it matters for AI

A language model is, in one respect, a Mary who never leaves the room. It is trained on what people have written about red, pain and grief, and it can describe them fluently. Whether that descriptive knowledge is all there is to know, or whether there is a further kind of knowledge, from the inside, that such a system lacks, is exactly the question Jackson’s case was built to ask. The debate over machine understanding, taken up in the Chinese room and in Do large language models know what they say?, turns on related questions.