One of the most puzzling observation that I had as a child was about colours. How does one know that the colour they see is the same colour that others see? For instance, when a person refers to an object as the colour red, does that object appear as the same colour of red in another person’s eyes, or does the second person simply agree that the object is of the colour red, but the redness they see is in fact a different colour from what the first person sees. Say, maybe, what the first person sees as the colour red, and according to the colours as defined in the first person’s perspective, the second person actually sees the colour blue, but refers to it as red because that is the standardized reference and facilitates institutions such as the traffic lights. People may agree that the light has turned red, but not necessarily what red is.
One may observe the state of confusion that I still find myself in to this day, by simply reading and re-reading the previous paragraph. Even as I wrote it down, convoluted thoughts and language still clouded my mind. Of course, as I grew older, I learned that scientifically, we define the colour red as a specific frequency of electromagnetic waves. So, in this sense, everybody can agree, mathematically, what is the colour of red, and be able to go on merrily with our lives.
But knowing redness scientifically and mathematically is not necessarily the same as knowing it perceptually. Colourblindness is one example where some may perceive a colour incorrectly, or not perceive it at all. However, this is not strictly the same as my opening question. In colourblindness, when the patient refers to a red object as yellow, that is, we presume, their visual receptors perceive the colour yellow, albeit incorrectly. Whereas in the opening question, the individual may perceive it as yellow but continue to refer to it as red, so that the misconception is undetected. In the case of colourblindness, it is an issue of anatomy or physiology. In the case of misconception, for lack of a better term, it is an issue of definition.
I never truly solved the childhood puzzle, so the conclusion that I eventually arrived at is that, one can never be certain if what one perceives is the same as someone else’s perception. Which, in hindsight, is a benign and obvious statement. However, if we cannot even agree on whether something is red, can we truly agree on the objective truth of anything?
How we know something is red has two necessary conditions. Firstly, our visual receptors perceives a specific wavelength. Secondly, we have an intuitive understanding of the colour of red. We have to have learned what red is in the past to be able to recognize it as red. For most of us brought up in Canada or the US, that might constitute our caretaker pointing to a ripe strawberry or apple or something exhibiting redness, and saying the word red to us until it integrates into our world context. Redness, and perception by extension, is as much a scientifically measurable trait as it is a social construct.
Because, you see, we can never picture a colour that doesn’t belong to our visual spectrum. Imagine a colour off the visible spectrum, say, infrared. It is a real wavelength that we can scientifically measure, but cannot perceive with our biological instruments. Some animals can, and so infrared may form part of their perception, but not ours. One may think of the infrared sensors in movies and video games, perhaps a night-vision goggle turning heatmap into visual images. But even in those cases, the infrared wave is translated into a colour within our world context, often the colour red, for us to perceive naturally.
If we are only to look at the world through our native perception, the image and understanding that we construct for the world is necessarily flawed and skewed, for the instruments with which we observe are flawed and skewed. Flawed not in the sense that it is incorrect. For all intents and purposes, to function as an average human in the modern world, we need not agree on what the colour red truly is, only that the traffic light has turned red and we should hit the brake. In that sense, our perception of red is perfectly functional and practical. Flawed, however, in the sense of objective truth.
Of course, the objective truth when it comes to red, one may argue, is that it appears generally to be between 620 and 750 nanometers. I would make a fool of myself arguing the scientific basis of wavelength measurements and definitions, but are those definitions not assuming further world contexts: that we measure in metric units, that we operate with base ten numerals, that we exist as three-dimensional beings? The whimsical novella and thought experiment Flatland comes to mind: how would a two-dimensional being see the colour red? How would they even see wavelength?
A two-dimensional being necessarily view the same object differently from our usual three-dimensional perspectives, just as different individuals, perspectives, and contexts view and frame the same thing differently. A joke goes like this: how to tell a chemist from a plumber? Ask them to pronounce the word “unionized”.
Perhaps “arriving at the objective truth” is a misguided endeavour entirely. As we have discussed, it may be unlikely that we can arrive at an objective truth, given our necessarily biased world context. In that case, perhaps the humbler and utilitarian angle is to “approach the objective truth”. That is, acknowledging that the truth is out of reach, but striving to close our gap as much as possible. Arriving at the asymptote of truth, so to speak.
This is where we find ourselves back at The Bitter Lesson. For what is an asymptote if not the convergence of data that infinitely approaches, but never arrives at, the true value. When AlphaGo won against humans, or in other endeavors where AIs have succeeded, they do so not by knowing the best strategy in the game of go or in speech recognition, but by statistically regressing through an incredibly large set of data that allow them to approach the best strategy for solving the given problem. The programers behind the AI did not define all the moves in the winning game, or perhaps any move at all. The only instruction, to put it in oversimplified but illustrative terms, may be to “go figure it out yourself”.
Generally, the act of figuring out could be resolved in two manners. The first is to reason through it. As a human masters the game of chess, first they learn of the rules then they learn about best practices, examples from the past, and so on, and internalize them through understanding. A chain of reason similar to “moving my rook to X square may elicite Y response and result in my victory” can constitute figuring out.
On the other hand, one could play an infinite number of games, learn what worked and what failed. The player in this case may not have any reason behind their moves, in the sense of predicting responses and future moves, but rather “moving my rook to X square yields 7.5% chance of losing a piece but 78.9% chance of winning”. Observation and statistics replaces reasoning. This would be consistent with Dr. Sutton’s position on the “brut force” method for training AI.
We could call modern AIs asymptotic, and perhaps borrow that idea for ourselves. The same way large language models learn to produce the most likely next token by observing an enormous amount of data, we may be able to arrive at the asymptote of truth if provided with near infinite perspectives. We may not definitively know what the colour of red is, but if we can measure what people see when they see the colour red, we can arrive at a reasonable approximation of what red is.
Maybe an ordinary person is not interested in debating the true colour of red. But there are other subjects whose truth we do care about. Politics, economics, culture, education are just a few examples. Perhaps we already hold a view on such a subject. For example, one may view that a free market encourages economic growth. But that view encapsulates one’s assumptions about market economics specifically and the world broadly. It’s only if one has integrated the opposing view, or a middling view, and ideally all angles on this matter, that these assumptions can be neutralized, even if not perfectly, allowing one to approach a more objective description of the relationship between free markets and economic growth.
On the other hand, the asymptote that approaches the truth, constructed via a multitude of perspectives, or lens as I have called it in other discussions, can also complicate our view of the world. Sure, the more lens we have on a subject, the more clearly we may be able to converge on a truth value. But additional clarity also brings more details that we may not have observed, thus more questions and complexities. Perhaps we could save these for a future discussion.