Some common mechanism of cognition and science

Coming from recent personal experiences, I have asked myself quite a few times whether and then also why it is necessary for us to interpret other people’s behavior instead of just taking it at face value. For a while I really felt it would be much more factual and less limiting if we could just observe a behavior and then interact with a person based on what they’re doing instead of based on our interpretation of it. But, unless I am much mistaken, this is neither possible nor would it be practical. And that is where I draw the parallel between science and human cognition: reducing reality into as little and compressed pieces of information, which I would call prediction models.

It may seem obvious but I believe it’s yet worth pointing out that the world we live in, even from a mere physical point of view, has way too many moving parts–on all layers of observation, from subatomic particles, energy fields, and transmission phenomena to people in a crowd that could move virtually anywhere. And as much we may sometimes wish that evolution has truly made the biggest leap ever between the great apes and the human species, our cognitive abilities are still fairly limited. Our experience might sometimes suggest we perceive everything around us as pristinely as HDTV compared to the old standard, but that is probably rather an illusion than reality. So what does happen?

Each and every second, our brains are literally bombarded with data coming in through various channels. For some reason is seems to me that most people have developed a strong preference for visual data, such that seeing has become one of the most important input modalities. But seeing is not perceiving… Our brains have a multi-layered and computationally efficient set of modules that incrementally reduce the data that is being received by primary visual areas into a dense stream of information. Imagine walking through your neighborhood and then seeing a family member about 500 feet away, waving at you. A chain ensues, from a first recognition that what you are seeing is another human being to the point where you realize it’s a family member and then, in combination with some memories, remembering you had made an appointment for later in the day, and finally inferring that the waving is a means of grabbing your attention because something needs to be communicated.

At each step of this (hypothetical but I think plausible) chain, different brain areas are involved, providing necessary patterns that allow to form a consistent and seemingly “feature-rich” experience. In reality however, almost all other data–which could have become information–is rejected in favor of this one cognition: your family member wants to communicate something to you. All other possible explanations for the same reality are discarded, as are all unnecessary elements of data that do not support the conclusion that was reached. And while this is all very elegant, efficient, and beautiful, it also means there is always a chance of missing something fundamental.

And how does it work in science? Well, to some degree it usually starts with a somewhat unsystematic but usually significant observation. For instance, the fact that liquids seem to evaporate under certain conditions, and faster so at higher temperatures. However, while this observation might be made repeatedly, in itself it is only a fact and does not necessarily allow to make predictions about the future. And that is, I believe, what drives both the efficiency in cognition as well as the motivation behind scientific discovery: the desire to “know” the future, so as to enhance the chance for making life-supporting decisions.

To be able to do so, one needs not only observations but also something that links observations to a formulation, a model. And such a model can be extremely simple, such as some already fairly old physical models–which by now have been substantially updated or enhanced to incorporate new evidence that would not be explained by older models. Importantly, a model does not necessarily describe the exact mechanism of transmission of effects! As long as it is able to make correct predictions, at least under a broad range of circumstances and contingencies, it is usually accepted and preferred over not having any model. However, if a model also contains hypotheses about the presumed mechanism, it usually is easier to expand it to uncovered circumstances and test directly; otherwise it may be more difficult to explain why it succeeds or fails under those new conditions.

In cognition, there is a direct mechanism to detect inconsistencies in mental models: a neural signal usually called prediction error. Coming back to the example with the family member, it is entirely possible that the family member who is waving has not yet seen or recognized us and is waving at someone else, in which case the prediction we are likely to make that the family member will continue walking toward us could be proven wrong by subsequent events, such as him or her hugging someone else who was standing close by. Our prediction then needs updating, which is signaled to us, possibly associated with the feeling of surprise, and that is where flexibility in mental models becomes important–as well as in science.

One of the basic notions that I belief are important to keep in mind is that whatever we conclude from an observation, such as it adhering to a specific model, should be taken with a grain of salt. For one, we might have missed a crucial element of data on the input side of the equation. For another, our mental models might simply not cover the circumstances. And finally, the model we apply might be too narrow to allow extension in case of new evidence. There are probably other important cases to consider, but those three are for me the main causes of continued misunderstandings and misinterpretations.

In short, we have to interpret reality to make sense of it–it’s simply too complex to be taken as is. But I believe we should always leave room for new data, new contingencies, and new models or extensions to old models. Then, maybe, we can find a way of using this necessary function of reducing reality to a small set of models even more efficiently and life-supporting.