Mary has a problem. She's colorblind.
This might not seem like much of an issue at first glance, but delve deeper, and you'll discover that Mary's situation is unique and paradoxical. She is a renowned neuroscientist, an expert in the field of human vision—her knowledge encompasses every aspect of how we perceive the world visually.
She can intricately describe how light travels, enters the eye, is received, and processed within the brain's complex networks. Mary can effortlessly detail the precise wavelengths corresponding to each color—red, green, blue, and beyond.
Her expertise extends to understanding the physics of color and the myriad ways our brains interpret these visual signals.
Yet, despite her extensive knowledge and her status as a leading authority in her field, Mary has never experienced color herself. Her world, up to this point, has been a monochrome canvas, devoid of the very essence she has spent her life studying.
Suddenly, a miracle (or convenient plot device, ahem) occurs: Mary can now see every color!
Now, here’s a good question for you that unveils the reason for this thought experiment in the first place: does Mary gain any new knowledge by looking up at the bright blue sky, or by seeing what a red tulip looks like?
Is there a difference between knowing something conceptually, and experiencing it?
To both ask and answer this question, I’ve brought in, who first brought the concept of Mary’s Room to my attention. Regular readers of Goatfury Writes may remember that I published Turing 2023, a brief look at the history of the Turing test and on whether we’re close to anything passing the test (fooling a judge into thinking an AI is human). On the same day, Alex alerted the community to a piece he was working on about the Turing test, but updated through a modern lens. I was all too happy to provide some feedback, but his piece speaks on its own, and its intriguing title is Can Machines Think?
Ever since then, we’ve bounced ideas back and forth, and we’ve both had a lot of fun thinking about big questions like this one.
Now, back over to Mary.
On one hand, it could be argued that Mary does gain new knowledge—the experiential, subjective understanding of what it feels like to see colors. This knowledge isn't just factual or descriptive (knowing the wavelengths of blue light), but it's a direct, personal experience of color perception. This perspective suggests that there are aspects of knowledge that can only be gained through direct experience.
On the other hand, another viewpoint might argue that Mary doesn't gain new "factual" knowledge. She already knew everything scientifically knowable about color. What changes is her personal experience, but this doesn't necessarily equate to a new type of knowledge. From this perspective, knowledge is seen more as an objective understanding, and personal experience doesn't constitute a new form of it.
This thought experiment, often related to Frank Jackson's "Mary's room" philosophical scenario, illustrates the complex relationship between scientific knowledge and personal experience, and it raises questions about whether there are limits to what can be learned from objective, third-person perspectives.
Alex points out that this is an age-old philosophical debate between rationalism and empiricism.
Rationalism is the theory that reason itself is the primary source of knowledge, and that reality can be understood through primarily intellectual and deductive processes. This school of thought emphasizes the role of innate ideas, reasoning, and logical deduction in acquiring knowledge. Like the ancient Greek philosophers, the belief is that there are fundamental principles and truths that can be arrived at just by thinking about them.
There are things that can be known purely from a rational perspective, like math theorems. You don't need to go outside in the universe and measure straight triangles to see that the Pythagorean theorem is true.
There are some things that can only be known through experience because there is no reason why the electron has to have a negative charge.
There's no a priori reason for that. It's just something that is, and you have to measure it.
Empiricism, on the other hand, argues that all concepts and knowledge are ultimately derived from sensory experience. In this view, observation, experimentation, and the practical use of our senses are the primary means of gaining understanding and knowledge. This approach is deeply rooted in scientific methods, where hypotheses are tested through experiments and observable phenomena.
Here’s where things get really interesting:
If we ever devise an AGI, an artificial intelligence that can read all of the Internet and know everything that we know, will it truly know what it means for an apple to be red, or what it means for sugar to be sweet, or what it means to love somebody, or any of those experiences? Or does it need to have embodiment? Does that intelligence need to be placed in a robot and it needs to go outside in the world? Does it need to try things?
This is a great set of questions that keep coming up for both Alex and me. What about human inputs into, say, ChatGPT? Should those count as sensory inputs? If not, why not?
After all, our senses are really just ways data can get into our brains. The nose knows what I’m talking about.
Is sight a fundamental part of the human experience? Few among us would suggest that someone with a serious vision impairment is somehow less human. If you lost your sense of smell during the early days of the pandemic, did you become somehow less intelligent or human?
There’s plenty of stuff out there that we don’t sense, but other animals do. Birds use magnetism to navigate, for instance. Dogs can hear and smell all sorts of things we can’t, and snakes can see infrared colors.
My personal take here, and perhaps a way to reconcile multiple points of view, is that I don't necessarily believe that any two humans view experience the same way. There's no way for us to know what another person sees as "red" or "cold", nor whether it's anything at all like the way we see it or feel it. I'm not sure, therefore, that anyone knows what it's like to be me, to have my experiences. They could be utterly, completely different than anyone else's.
As is often the case, Feynman deserves the last word here:
Now, what do you think? Do you “gain” anything from sensory input you can’t also gain from data? Bring your thinking caps to the comments section today, and let’s get into it!
Thanks for reading Goatfury Writes! Subscribe for free to receive new posts and support my work.