The nature of scientific understanding is related to the nature of our general understanding. We need reasons when specific things happen because those reasons could help us predict what is going to happen next. For example, if my friend Sanjay rings the bell of my house unexpectedly, I try to find out a reason why he would have come to pay a visit. If I don't have this drive to find reasons, I cannot do any prediction at all and all events in the universe will seem random and arbitrary. Perhaps the drive to find reasons is ingrained in us via evolution in the form of [predictive processing](https://www.quantamagazine.org/to-make-sense-of-the-present-brains-may-predict-the-future-20180710/) with our brain is constantly trying to predict what happens next. Continuing the example, when Sanjay rings the bell and my brain quickly scans through a variety of reasons. Perhaps he was in the town and decided to surprise me. Perhaps he needs my help. Perhaps I'm imagining that he's here. And so on. All these reasons have additional predictions. For example, if he were in town maybe he would have posted on Instagram. Did I notice any such picture on his Instagram today? If he needed my help, perhaps he would have called me before. Have I not checked my phone today? Depending on how the actual observed data fits the predicted data from an assumed reason, we can tentatively settle onto a reason until we collect more data that either confirms the reason we've decided or forces us to change and adopt another reason. In this sense, we can see that reasons are nothing but predictive models of the sort: >A: Friends surprise >B: When distant friends are in the same city, they tend to meet >C: Sanjay is a friend who doesn't live in the same city as me >D: Ergo, Sanjay must be paying me a surprise visit for fun Note that the model above need not correspond with reality. It _could_ be the case that Sanjay wanted career advice but he is proud enough to never ask for it directly, so he architected a "surprise visit" and played along with the story I cooked in my head to explain why he would visit me without calling. To know the *real* reason, I will first have to come up with this specific reason as an alternative and try to design clever interventions in order to figure out whether this is the reason. For example, I may decide to not talk about career at all and see whether Sanjay continuously tries to steer the conversation in that direction. From the example above, we can notice following points about the nature of explanations and understanding: 1. **Reasons or explanations are predictive models under the hood**. We're attracted to finding accurate reasons because we want to have accurate predictions that help us gain an information advantage over others (and the universe) by predicting what can harm or help us several steps before it actually happens. 2. **Data or observations DO NOT supply any reasons, we have to come up with them ourselves**. Coming up with a reason is a creative act and since it is a creative act and a product of human brain, we're biased towards simpler reasons. (This explains Occham's Razor). Our brain cannot handle models that are extremely complicated or that contain many entities and relationships. Our energy efficiency reasons, our brain works with simple models. Hence, **understanding for us is discovery of a simple, easy to comprehend model that fits the observed data** 3. **All reasons are tentative**. We may never know the _real_ reason for something as the underlying phenomena may be extremely complicated or random while we can only come up with finite hypotheses (most of which are simple). That is, since reasons are a creative act of the human brain, we may never stumble upon the real reason. #### Scientific understanding Our drive to do science is an extension of our basic drive to find reasons for events in the world. In that way, science is all about making predictive models. It aims to explain a phenomena (or a set of interrelated phenomena) by building a predictive model out of it. It builds models of the nature X + Y -> Z. But for any given phenomena, there can be variety of different predictive models. So, typically, we tend to judge predictive models to be better ones depending on the following criteria: - **Experimental falsification**: does the predictive model make a new prediction which we can collect data for and observe the degree of alignment between what actually happens in reality and what the model suggests will happen? Greater the alignment, greater the preference. - **Parsimony**: how much a predictive model can explain for its given complexity. For two predictive models with similar span of explanations, the simpler one is generally preferred. Since science is about predictive models, true reality remains inaccessible to us. In the [model dependent realism](https://en.wikipedia.org/wiki/Model-dependent_realism) stance, reality can only be known through models and never directly. Consequently, two different models explaining the same reality are _both_ real. In that sense, **reality is nothing but a set of predictive models we've deemed useful for predictions.** #### Beyond scientific understanding There are certain questions like: "why does the universe exist?" or "why red feels like read" that seem to be beyond the ability of science to answer. To gain a better handle on these questions, let's try elaborating on what would an answer to these questions even look like. Since our understanding is limited to finding hypotheses or reasons that help us make predictions, our perfect answer for these questions would also be of similar (predictive) nature. It would assume some reasons and then try to see whether those reasons imply the existence of universe or redness of the color red. But coming up with such a reason is difficult because: - **Either a reason can be of non-sensory or abstract nature**. In which case, mathematics is the tool of our choice. - Example, black hole dynamics. We have a hypothesis of a model which gives certain consequences (such as light bending from behind the black hole) which we can observe and increase our confidence in the model. - **Or, it can be sensory**. In which case, our _direct_ sensory observations are the tool of choice. - Example, close friends pay surprise visits. We have a hypothesis that close friends pay surprise visits while other people don't. When we observe confirmatory examples of both cases (for close friends and other people), our confidence in the model grows. For the question "why does the universe exist", we can come up with all sorts of hypotheses but the difficulty is in collecting evidence supporting those hypotheses. For example, if the answer is that universe exists because God created universe for humans because it loves them, the presence of hate or evil in the world doesn't collate with the prediction from the reason that only love should exist. Also, in theory, this question seems answerable even though a satisfactory explanation will generate additional questions. So, for example, if we predict that the universe exists because some entities wanted to observe which conditions lead to more suffering or less suffering and for that they created many universes, out of which ours is one. This answer simply pushes the question ahead where we ask why those entities exist. So, in practice, it's difficult to get a self-bootstrapping *final* predictive model. The other question "why red feels like red" is a different beast altogether. For this question, both modes of reasons fail. Since we require a predictive model that predicts the color red (a sensory quality), we cannot use a reason that's non-sensory or abstract. So mathematical model is out of question since mathematical model give mathematical outputs and sensory qualities aren't that. Moreover, since what we need to explain is sensory quality itself, we can't use sensory quality as an input (it has to be the output, the prediction). Unlike the case of "why does the universe exist", it seems we're unable to even imagine what would a predictive model for this question will look like. This is the "hard problem" of consciousness. #### The science of brute facts One way out of the quandary above regarding the questions "why does the universe exist" and "why does red feel like red" is to assume their existence to be brute facts. So our hypothesis becomes: - Reason for "why does the universe exist" => It just does - Reason for "why does red feel like red" => It just does To make the analogy to initial example, perhaps my friend Sanjay paid me a surprise visit because he just did that. Brute fact explanations are unsatisfactory in cases where the underlying data could have been different than how it was observed. For example, if it was possible for Sanjay to not pay me a surprise visit but he did, a brute fact explanation seems lazy cop out. But if the data could NOT have been anything different, then brute fact need not be a lazy one. Indeed, it could be the only explanation that makes sense. For example, if for the question "why does red feel like red", we answer: "what else could it feel like?", we see that we're unable to imagine a counter-factual. Similar for the question: "why does the universe exist" we're unable to imagine a counter-factual (it's hard to imagine a non-universe). In these cases, **where counter-factual cannot even be imagined, brute facts might be a reasonable explanation**. (Although, unsatisfactory). By itself, this brute existence of an event isn't particularly interesting. **What is interesting is to see what can the brute existence suggest to us which cannot be directly observed?** That is, can we use brute existence as a data point to generate new questions or use it as an input to another explanation? For example, what else could "the universe exists because it simply exists" suggest? Some ideas: - Since universe seems to be animated according to mathematical laws, **does this imply all mathematical relationships and entities simply exist as a brute fact? ** - Ask what is more likely: - all mathematical entities exist and we find ourselves in a region of mathematical space capable of generating intelligence that asks such questions? - Or, just our specific universe exists as a brute fact? - It seems like the former case where all mathematical entities exist is more likely than a our specific one existing as a brute fact. How do we formalize/defend this intuition? What else could "red feels like red because it does" suggest? Some ideas: - **What is the equivalent of "all mathematical laws" for sensory qualities?** - This will help answer which all sensory qualities exist (even the ones we don't have direct access to) - Like mathematics, we can't enumerate sensory qualities beyond what we have an immediate access to. Are we stuck? - **If all mathematical entities and relations exist as universes, what sensory qualities exist in universes different than ours?** - **Is there a relationship between what specific universe an intelligent being finds itself in and the type of sensory qualities it has access to?** - Example: say some fundamental property of standard model is isomorphic to color-space (or sound space and so on) and conscious beings sense that property itself. - Is this a worthwhile project to attempt?