Why Science is Always Bounded Rationality

Why Science is Always Bounded Rationality I recently came across a critique of science that actually makes a lot of sense. Science, as the empiricist methodology supposedly grounded in objective facts, has always had fundamental logical flaws. First, there's inductivism: inferring from past facts to predict the future. This is easily refuted by the black swan argument, because the logic is fundamentally circular. The cognitive method (how we determine if something is true or false) is: it has always been... therefore it must always be... But the hidden logical assumption is precisely: 'has always been' can imply 'must always be.' Because the problems with inductivism were recognized early on, falsifiability emerged as a new method. If the problem with induction is that objects usually cannot be exhaustively enumerated (incompleteness), then the black swan refutation is at least logically valid: Finding one black swan allows us to say 'all swans are white' is false. But the problem is that in reality, there are far too many things that could potentially be falsified. For example, we observe that when an object recedes at near light speed, it appears redder. We want to prove that 'approaching light speed' is indeed the cause of 'appearing redder.' But if we use counterexamples to test this, the workload becomes endless: Why isn't the reddening caused by temperature? Or proximity to a massive object? If approaching light speed doesn't cause reddening, does that prove speed isn't the cause? (Was temperature controlled? Was gravity controlled?...) Such questions can actually continue forever. Moreover, a complex prediction often involves theory, assumptions, models, and boundary conditions. When predictions don't match reality, without perfectly controlling the other three factors, it's hard to falsify the theory. And what we usually test is a combination of all four. Some might say this level of scrutiny is wrong. But if we're concerned with logic and rationality, endlessly testing infinite questions is actually the logical endpoint; Stopping too early is actually unscientific in the falsifiable sense. So we gradually discover that the common understanding of science = absolute objective empirical verification is actually incorrect. Science is just one of humanity's methods of understanding, which happens to have more accurately captured the structure of reality so far; People choose science more for practical reasons than for having found objective truth. The reason science can quickly find useful rules among endless facts and hypotheses is precisely because humans presuppose theories/intuitions from the start (explanations that feel right without rigorous logical basis). For instance, the discovery of relativity came from Einstein's insistence that Maxwell's equations were correct; If experiments required time itself to stretch, then perhaps time itself really could stretch. So Einstein, without complete evidence, built an entire theory based on belief. And a large part of scientific discoveries are actually like relativity: someone insisting on a direction that seems wrong under existing understanding, then either through intuition or Frankenstein-like stitching, producing something that initially can't clearly explain cause and effect, but whose predictions match reality better than existing theories (like Planck explaining the discrete energy phenomenon in black body radiation). There's always been the idea of using AI to compress data and directly extract all possible laws, but thinking about it now, infinite questions quickly become computationally intractable, and even if we could isolate predictive mechanisms, selecting meaningful explanations from infinite possibilities would still be extremely difficult. Using all current predictive ML models as the sole criterion for determining truth is rather tenuous. This isn't something causal inference can completely solve—causal inference is more like using randomization, experimental design, and sensitivity analysis to minimize the influence of non-critical, inexhaustible variables on conclusions. For science to progress/be useful, it requires people/AI to make decisive calls at critical moments to force a halt, rather than simple data collection + computing power + gradient descent leading to all knowledge.