I have written before about the constructivist viewpoint that each person creates their own reality. This is sometimes illustrated by the philosophical argument that I don't know whether the way you perceive the color "red" is exactly the same as what I see when I look at something red. Maybe the visual experience that you have in that situation is one that I would call "blue" if it were happening to me, but we have both learned to call the experience by the same word "red."
There's a famous thought experiment in philosophy, called the "knowledge argument" about personal experience, from Australian scholar Frank Jackson (1982). In this story, a scientist named Mary has lived her entire life in a black-and-white room with no opportunity to see colors. (If that's too big a stretch for you, alternate versions of the story posit a neurological problem of some sort to account for her lack of color perception). Nevertheless, color is Mary's area of scientific expertise. She uses detailed measures of wavelength characteristics and retinal cells to understand everything there is to know about color perception by the human brain -- yet she has never herself experienced color. The story posits that Mary literally knows everything it is possible to know based on physical characteristics. Mary then gains the ability to see color, either through release from her monochrome prison or by means of neurosurgery. The question is whether Mary learns anything by finally seeing colors. If so, then her knowledge of color was actually not complete beforehand. Jackson uses this to argue for an element of existence separate from physical reality -- our experiences, what philosophers often call qualia.
Qualia are the level at which people might have different ideas of "red" even though a red apple has the same physical characteristics. Technically, the idea that everyone might have a different experience of color arises from the problem of intersubjectivity -- we can never get fully inside of another person's mind. Philosophers dating back at least to John Locke in 1689 have argued that someone could have a completely different subjective experience of color, but as long as they used the same words in the same situations as we did, we would have no way of knowing that to be the case.
But researchers this year apparently found evidence for shared reality at the neural level, using AI tools to data-mine fMRI data. They found that by knowing only the neural firing pattern in someone’s brain, they could predict what color the person was looking at with slightly better than chance accuracy (34-40%, in a task where chance accuracy was 33%). That means that to at least a slight degree that's not attributable to chance, we do all see red in the same way, or at least with the same firing pattern of neurons in our brains.
Is the problem of intersubjectivity therefore solved? Are 350 years of philosophy rendered obsolete by modern neuroscience? Now for the caveats ... First of all, you will notice that the researchers' model was accurate about 40% of the time, compared to a 33% base rate. A 7% improvement (maximum) is statistically significant in this context, but it's not all that impressive in practical terms. Perhaps we can say that there are some commonalities across people in their experience of the color red, but those commonalities are outweighed by (a) chance -- the initial 33% base rate, and (b) unique and unexplained variability in people's brains -- the 60% of neural activity not predicted by the model. Instead of saying that everyone sees the color red in the same way, it might be more accurate to say there is some common neural substrate to color perception, which each individual brain builds on in vastly different and individualized ways.
Second, even in the same person’s brain, knowing the person's prior neural responses only gave the researchers 42-67% accuracy in predicting the color that was being reviewed (again, compared to a base rate of 33%). This is consistent with some of the neuroscience research that I summarized in my post on mind-uploading: Not only are the same concepts encoded differently across different brains, they are also encoded differently over time within the same brain. AI models can pick up just enough consistency to correctly identify what a person is thinking about, for example allowing for gross motor control of robotic arms. But they don't account for all the nuances of thought. Instead, they gloss over them in models that are just good enough to be functional.
Finally, it's worth noting that the commonalities across participants were in areas of the primary visual cortex -- the large area in the back of the brain where input from the eyes is converted to interpretable signals for the rest of the brain to work with. The fact that some of the same brain areas were lighting up in response to the same stimulus is not necessarily surprising. These are not the areas of the brain that assign labels to colors (the Narrative Mind -- generally language is localized in the left temporal lobe), and they are not the parts of the brain most strongly associated with subjective conscious experiences (generally connected to midbrain areas like the cingulate cortex). One might think, therefore, that the common brain activation across participants in this study was connected to perception or differentiation of color stimuli, but not to the subjective experience of color. In other words, the AI model can predict (to a degree that's modestly better than chance) which color word a person is going to use when they see an apple. But they can't tell us whether all of the participants actually mean the same thing when they say that same word, "red."
Comments
Post a Comment