Vitez Engineering Blog
Science Isn't Irrational: On Hume's Problem of Induction
Hume's problem of induction has long troubled the philosophy of science. Hume's problem states that there is no rational reason to believe in induction as there is no non-circular argument to believe that the future will resemble the past. As a result of Hume's problem, there is no rationale to believe almost all of science. The argument presented proposes that we should view science as a process of aiming to develop a model, or models, which explains the entirety of the universe. Viewing science in such a way allows scientists to justify choosing the models which are simplest, giving insight into Hume's problem.

Van Fraassen in The Scientific Image describes the framework scientific realisists believe:
Science aims to give us, in its theories, a literally true story of what the world is like; and acceptance of a scientific theory involves the belief that it is true.
I argue instead that we should consider that science aims to develop a working model about the world rather than "a literally true story" of what the world is like. I will justify this new framework by considering two universes, one where all knowledge is knowable and one which contains fundamentally unknowable knowledge.

Let's first consider a universe where everything is fundamentally knowable. In such a universe, for a model to be perfect it must, by definition, perfectly model the literal true story of the universe, thus the perfect model is no different from the literally true story. The perfect model of that universe is possible to create as everything about that universe is knowable (and nothing is fundamentally unknowable). Thus, viewing science as the pursuit of this perfect model is justified in a universe where everything is fundamentally knowable.

Now let's consider the universe where not all knowledge is attainable.Consider this universe as a black box where scientists input experiments and the box outputs the results of the experiment with some inner-workings of the box unknowable. Philosophers such as Van Fraassen would argue that science tries to perfectly replicate the exact inner-workings of the box in such a way to predict the results of experiments. Consider a "faux-box" that has different inner-workings than the black box but for any given inputs the faux-box yields the exact same outputs as the black box. To the outside scientists trying to discover the inner-workings of the universe, there exists no difference between the faux-box and the black box. Thus, it doesn't matter if scientists aim to develop the literal true story (black box) of the universe or an alternate model of the universe (faux-box). Both box's function identically from the perspective of the scientist. Furthermore, science has no way to distinguish a faux-box from a black-box; even if science did aim to know the "literally true story", there is no way to know if science got the unknowable inner-workings of the universe correct or rather if science developed an adequate model of the unknowable inner-workings.

Another analogy might be useful in order to justify why science can be viewed as aiming to develop a model of the universe. Consider a game with a 50% probability of outcome A. From the perspective of a player of the game it doesn't matter how this probability is assigned. Maybe the game uses a fair die and if the number is odd, outcome A occurs. However, to the scientist trying to understand the inner-workings of the game, it doesn't matter if the scientist uses the heads of a fair coin, black cards in a deck, or the odd numbers of a fair die in order to determine if outcome A occurs. In fact, a scientist with no more information about the game has no way of determining how this probability is "assessed". Thus, the scientist exploring the game can model this probability as a coin toss rather than as a die roll, without any loss of fidelity in understanding the underlying game

Consider the theory that the sun will rise tomorrow. Using our model based framework we as scientists might construct multiple models describing which days the sun will cease to rise. As theories are undetermined by evidence, there are infinitely many models consistent with all the past evidence we have observed. Then how do scientists decide which model to use? Well, scientists have the freedom to select any model, but for ease of calculations they should choose the simplest model that is consistent with all the evidence that has already been observed. This demonstrates the advantage and convenience of viewing science as the process of developing a model. Rather than answering the question which theory is the literal story of the universe consistent with all past evidence, we instead answer the question which model is the most convenient model consistent with all the evidence observed. Viewing science from a model framework gives scientists the freedom to choose the simplest model which fits all the evidence. It is impossible to select which theory on the rising of the sun is true, as there are an infinite number of theories consistent with all evidence observed so far. However, it certainly is possible to select a theory which is most convenient.

Consider the Principle of Uniformity of nature (PUN), the idea that the future will resemble the past. Normally, any argument to justify PUN is circular in nature. Let's examine how a model based scientific framework might decide a model of PUN to believe. Consider the infinitely many models that describe when PUN holds true which are consistent with all the past observations related to PUN. As scientists, it is simply an impossible task to know which version of PUN is consistent with the ''literally true story'' of the universe, as theories are underdetermined by evidence. However, if our aim is to make an ever improving model of the universe, we can simply select the most convenient model of PUN that is consistent with all the past observations. In this case, scientists accept the model that PUN is always true, not because they believe it is consistent with the true story of the universe, but instead because it is the most convenient model of PUN consistent with all the past evidence of PUN.

If the argument for viewing science as a model development process is to be believed, there is the question on why the convenient model seems to have such widespread success predicting phenomena. I speculate that the success of selecting the simplest scientific models is primarily due to selection biases and human psychology. Consider a country where schoolchildren learn calculus before learning how to read. These children would likely believe calculus is simpler than reading. On an Earth where the Einsteinian Theory of Relativity was long discovered and understood before Newton's Theory, the local scientists would likely believe the Einsteinian theory was simpler than the Newtonian. Likewise, in a universe in which things which are green change color to blue in the year 2030 (colloquially called grue) scientists would think grueness is less complex than greeness or blueness. I speculate that the nature of the universe is what generally informs our idea of what simple is. This means, in general, simplicity would be a good metric for selecting theories as our idea of simplicity is "tuned" to our very expectations of the universe.

Philosophers of Science should frame science as the process of developing an ever improving model for describing what the universe we live in is like. This framework makes sense to believe in both universe's where there is something fundamentally unknowable and those where everything is fundamentally knowable. As a result of this new framework, there is a new insight towards solving Hume's problem. Science develops infinitely many models of how the universe operates consistent with all the past evidence observed. As the aim of science is to model the universe rather than know the "literally true story" of it, scientists can justify choosing a model based on simplicity rather than based on what is the true story of the universe. This means that one model can be selected for approximating the phenomena by simplicity, rather than by being fully true and correct. Applying this model selection process to the Principle of Uniformity of Nature, leads science to justify selecting the most convenient model of PUN consistent with past evidence, namely that PUN is always true. Admittedly, applying this new framework towards Hume's problem still needs more philosophical development. In particular, it is important to more completely explain why choosing a simple model is justified. Scientists never choose a more complicated model than necessary to explain phenomena, as doing so would just complicate their calculations. But defining "simplicity" and "complicating", falls under a new riddle of induction. I speculate that simplicity is fundamentally tuned to how we perceive the universe to operate. Regardless, it still holds that viewing science as a framework for developing models is in itself a powerful framework towards solving Hume's Problem of Induction.