Monday, 7 September 2020

‘Mary’s Room’: A Thought Experiment

Posted by Keith Tidman
Can we fully understand the world through thought and language—or do we only really understand it through experience? And if only through experience, can we truly communicate with one another on every level? These were some of the questions which lay behind a famous thought experiment of 1982:
A brilliant neurophysiologist, Mary, knows all there is to know about her academic specialty, the science of vision: the physics, biology, chemistry, physiology, and neuroscience. Also how we see colour.

There’s a catch, however: Mary has lived her entire life in a totally black-and-white room, watching a black-and-white screen, and reading black-and-white books. An entirely monochromatic existence. Then, unexpectedly, her screen reveals a bright-red tomato.

What was it like for Mary to experience colour for the first time? Or as the Australian philosopher Frank Jackson asked, who originated this thought experiment, ‘Will [Mary] learn anything or not?’ *

Jackson’s original takeaway from his scenario was that Mary’s first-time experience of red amounted to new knowledge—despite her comprehensive scientific knowledge in the field of colour vision. Jackson believed at the time that colour perception cannot entirely be understood without a person visually experiencing colour.

However, not everyone agreed. Some proposed that Mary’s knowledge, in the absence of first-hand experience, was at best only ever going to be partial, never complete. Indeed, renowned philosopher Thomas Nagel, of ‘what is it like to be a bat’ fame, was in the camp of those who argue that some information can only be understood subjectively.

Yet, Mary's complete acquaintance with the science of vision might well be all there is to understanding the formation of knowledge about colour perception. Philosopher and neurobiologist Owen Flanagan was on-board, concluding that seeing red is a physical occurrence. As he put it, 'Mary knows everything about colour vision that can be expressed in the vocabularies of a complete physics, chemistry, and neuroscience.

Mary would not have learned anything new, then, when the bright-red tomato popped up on her screen. Through the completeness of her knowledge of the science of colour vision, she already fully knew what her exposure to the red tomato would entail by way of sensations. No qualities of the experience were unknowable. The key is in how the brain gives rise to subjective knowledge and experience.

The matter boils down to whether there are nonphysical, qualitative sensations—like colour, taste, smell, feeling, and emotion—that require experience in order for us to become fully familiar with them. Are there limits to our comprehension of something we don’t actually experience? If so, Mary did learn something new by seeing red for the first time.

A few years after Frank Jackson first presented the ‘Mary’s room’ thought experiment, he changed his mind. After considering opposing viewpoints, he came to believe that there was nothing apart from redness’s physical description, of which Mary was fully aware. This time, he concluded that first-hand experiences, too, are scientifically objective, fully measurable events in the brain and thus knowable by someone with Mary’s comprehension and expertise.

This switching of his original position was prompted, in part, by American philosopher and cognitive scientist Daniel Dennett. Dennett asserted that if Mary indeed knew ‘absolutely everything about colour’, as Jackson’s thought experiment presumes, by definition her all-encompassing knowledge would include the science behind people’s ability to comprehend the actual sensation of colour.

To these points, Mary’s factual expertise in the science of colour experience—and the experience’s equivalence and measurability in the brain—appears sufficient to conclude she already knew what red would look like. The experience of red was part of her comprehension of human cognitive functions. Not just with regard to colour, but also to the full array of human mental states: for instance, pain, sweetness, coldness, exhilaration, tedium—ad infinitum.

As Jackson ultimately concluded, the gist is that, given the special particulars of the thought experiment—Mary acquired ‘all the physical information there is to obtain about what goes on when we see ripe tomatoes, or the sky, and use terms like red and blue’—Mary did not acquire new information upon first seeing the red tomato. She didn’t learn anything. Her awareness of redness was already complete.

* Frank Jackson, 'Epiphenomenal Qualia', Philosophical Quarterly, 32, April 1982.


Thomas Scarborough said...

Well, talk about an article being topical. This piece was published just hours before the BBC published 'Can artificial intelligence create a decent dinner?' You saw it first on Pi.

Keith said...

Although the BBC item was somewhat playful, I agree, Thomas, there exists a more ambitious tie between the ‘Mary’s room’ thought experiment — aka the ‘knowledge argument’ — and artificial intelligence. Many of the factors surrounding AI will increasingly borrow from and/or adapt aspects of the knowledge argument — as framed or inspired at least in part by the ‘Mary’s room’ discussion — to suit AI-related considerations: put simply, the what and how of knowledge. Also with respect to AI and the knowledge argument, longer-term considerations, I suspect, might increasingly bear on the still-thorny issues related to consciousness, ‘personhood’, and the full range of what falls under the rich rubric of cognition, including experience and learning. The stuff of both philosophy and neuroscience.

Keith said...

You might have seen this already, Thomas, but germane to your point and BBC post, here’s a whimsical op-ed written, of all things, by an AI language generator for The Guardian. The title of the op-ed is ‘A robot wrote this entire article. Are you scared yet, human?’ Quite convincing. The URL:

Martin Cohen said...

Re. that Guardian opinion piece, like much in this area, it is rather breathless and rushes to declare human intelligence matched by machines. That claim was made back in the 1960s when computers 64k things barely capable of returning user input. And this Guardian essay, it turns out in the small print, was re-edited by humans, who merged elements from six computer versions to make "a good one".

But back to this essay, and I do think it is very fine one, just the stuff a computer could be proud to have done! Maybe I don't quite get the argument, but it seems to me that experience and knowledge are not the same things. I may be told that the wavelength of red light is x, and that of green light is y, but still not be able to distinguish between the two when I look at a traffic light...

Keith said...

I suspect that not even AI’s most enthusiastic cheerleaders pretend AI yet matches human intelligence. Rather, what’s going on is people tipping their hats at the wow factor of each step in AI’s advance, across many sectors of our lives. Also, I’d venture there are gulf-like differences between the AI of the 1960s and the AI of the 2020s.

That progress will predictably continue, as AI experts, computer scientists, neuroscientists, physicists, biologists, engineers, industrial researchers, and experts in other fields determinably keep moving that needle. So, just as the AI of the 1960s and the 2020s don’t compare, the AI of the 2020s and the 2100s likely won’t, either.

Keith said...

I agree that knowing the ‘wavelength of red light is x, and that of green light is y’, is insufficient to fully understand the difference between the two at a traffic light. The personal experience, that is, of red and green as one sits in her car. But that’s only because the ‘wavelengths’ of red and green don’t represent the totality of what distinguishes the two colours.

I think that the Mary of this thought experiment, her hypothetically endowed as a brilliant neurophysiologist with all knowledge of the physics, biology, chemistry, and neuroscience of colour vision, would by definition have complete knowledge (beyond just wavelength) of the experience of red and green traffic lights — including of what brain events occur in our experiencing red and green at a traffic light.

Thomas Scarborough said...

I think there is a mistake in the experiment. There is a suppressed inference: the bounds of information in both knowledge and experience are basically the same, which in this case concerns the narrow information and experience of a colour.

Our experience deals with things as they relate to the universe, while our knowledge (or what it has become) deals with stripped down concepts. My experience tells me what, say, a woman is in all her inexhaustible fullness, while a definition tells me she is 'an adult human female'.

Mary knows 'absolutely everything' about colour, says the post. Yes, as a stripped down concept. But that description of Mary's knowledge is deceptive. It is not 'absolutely everything' that she knows. It is 'absolutely everything' as a theorist decided it.

Tessa den Uyl said...

That last comment above by Thomas sounds plausible to me. The mind, if trained by discriminating, might presumely become that rigid that it al-ready knows by sticking to the acquired theories. Though orange-red or purple-red, how to be sure of red? Nobody sees colors alike, like nobody experiences a cut leg alike. There is more in the game than just the knowledge.

Post a comment

Recent Comments