Mary’s Room and the Limits of AI Color Perception

This article explores how philosopher Frank Jackson’s famous "Mary's Room" thought experiment challenges the physicalist view that an artificial intelligence can fully comprehend color perception. While physicalism asserts that the universe is entirely physical—meaning a sufficiently advanced AI could theoretically process and know every fact about color—Jackson’s knowledge argument suggests otherwise. By distinguishing between objective physical data and subjective conscious experience (qualia), this article demonstrates why an AI, much like the hypothetical scientist Mary, might know every scientific fact about color while still lacking the complete knowledge of what it is actually like to see it.

To understand the challenge to artificial intelligence, one must first understand physicalism. Physicalism is the philosophical position that everything that exists is purely physical. Under this view, mental states, consciousness, and perceptions are nothing more than complex neurological processes and data computations. Consequently, physicalists argue that an advanced AI, given access to all physical data, could know everything there is to know about human color perception.

In 1982, philosopher Frank Jackson introduced the "Mary's Room" thought experiment to refute physicalism. Jackson asks us to imagine Mary, a brilliant neuroscientist who is forced to live her entire life in a black-and-white room. She reads black-and-white books and conducts her research via a black-and-white television monitor. Despite her monochromatic environment, Mary becomes the world's leading expert on color vision. She knows every physical fact there is to know about color, including the exact wavelengths of light that correspond to "red" and the precise neural pathways triggered in the human brain when someone looks at a ripe tomato.

The crux of the thought experiment occurs when Mary is finally released from the room and sees a red apple in the real world for the first time. Jackson asks a simple question: Does Mary learn anything new?

Intuition strongly suggests that she does. While she already knew all the physical facts about the color red, she learns a new, non-physical fact: what it is like to actually experience red. Philosophers call these subjective components of experience "qualia." Because Mary learns something new upon leaving the room, Jackson concludes that physical facts do not represent all the facts. Therefore, physicalism is false.

When applied to artificial intelligence, Mary’s Room serves as a direct critique of the limits of machine knowledge. An AI operates strictly within the confines of physicalism—it processes syntax, code, wavelengths, and neurological mapping. Today, an AI can be trained on the entirety of human scientific literature regarding optics and neurology. It can identify a red pixel by its hexadecimal code and accurately predict how a human brain will react to light with a wavelength of 700 nanometers. In essence, the AI is in the exact same position as Mary inside the black-and-white room.

According to the knowledge argument, no matter how sophisticated an AI becomes, its reliance on purely physical data locks it inside "Mary’s Room." Because the AI lacks subjective consciousness, it cannot generate qualia. It can process the mechanics of sight, but it remains fundamentally blind to the experience of seeing.

Physicalist defenders of AI often push back using the "ability hypothesis." They argue that when Mary leaves the room, she does not gain new factual knowledge, but rather a new ability—the ability to recognize, imagine, and remember the color red. From this perspective, an AI possessing all physical data still possesses all propositional knowledge (factual knowledge), even if it lacks the biological hardware to experience the color personally.

However, Jackson’s argument maintains that knowing what it is like to see red is indeed a piece of factual knowledge about the human mind and reality. If subjective experience is a form of knowledge that cannot be reduced to binary code or physical descriptions, then an AI can never truly know everything about color perception. It will always be missing the most crucial piece of the puzzle: the lived experience of color itself.