Active Epistemology | Yonathan Fiat

Here’s a common picture of the ideal epistemic agent. He is tied to a chair without being able to move. His eyes are forcefully held open. And he is drugged such that he has no plans, hopes, or even thoughts. His entire being is about responding to what comes in.

Everyone can agree that such an agent is not going to know a lot. We know as much as we do partly because we look for evidence. But, according to most mainstream views in contemporary epistemology, this is an aside, an unfortunate complication that can be safely ignored. According to this mainstream position, the ideal epistemic agent would be like the one I just described, but attached so a more active agent who would be in charge of gathering the evidence. This way, our perfect epistemic agent won’t be sullied by such things.

There is a different possible picture of ideal epistemic agent. Imagine a rat in a maze with a hidden piece of cheese. She runs down the different routes and sniffs the air. She forms and follows a plan. She comes up with hypotheses and tests them. The rat is admirable in many ways. She is likely to achieve her goal of getting the cheese. She is a good inquirer. She works to get knowledge. She doesn’t passively attempt to make the most of what finds its way to her.

This project is about developing this second picture of ideal epistemic agents. I think that we can make progress on many of the traditional questions of epistemology when we adopt this kind of picture. I also think that when we adopt this picture, we can make better sense of our scientific practices.

For example, in “A predictivist account of significance testing” I offer a new justification for significance testing, one of the most important practices in science, by appealing to the idea that when we make a successful prediction, it makes a difference.

In “Bandits of knowledge” I argue that to keep our knowledge, we must actively maintain it. I appeal to the famous multi-armed bandit problem, and argue that it teaches us that we can lose knowledge simply by failing to maintain it by looking for evidence.

In “Knowledge by choice” I argue that in order to avoid skepticism, we must accept that what we know depend on our choices, and not only on what’s given to us. I show that this insight helps solve the lottery paradox, and it explains our inductive knowledge and our statistical practices.

Photo by Annemarie Horne on Unsplash

References

Under review

A predictivist account of significance testing

Yonathan Fiat

Under review

Abs

Significance testing (ST) is arguably the most widely used tool in statistical analysis. Since its introduction in the early twentieth century, it has led scientists to countless successes. Yet, from the beginning, it has been the target of relentless criticism. Critics - including philosophers, statisticians, meta-scientists, and practitioners - have called it irrational, unmotivated, and fallacious. Even its defenders have often struggled to explain and motivate it. We can do better. In this paper, I show how ST arises naturally from the question "how should we evaluate successful predictions?" Some predictions are more impactful than others. The success of the prediction _it will rain tomorrow_ is somewhat impressive; the success of the prediction _it will rain frogs tomorrow_ is astonishing. I show that if we sharpen and extend this observation, we end up with ST. When a scientist reports that she got a _statistically significant_ result, she reports that she made a prediction and that its success is impressive. If we add the premise that good inquirers are sensitive to impressive successful predictions - a position known as _predictivism_ - we get the conclusion that ST is a rational practice. Moreover, we also get an explanation why it’s a valuable practice, and this explanation can guide us in its application, so we will use it better.

In progress

Bandits of knowledge

Yonathan Fiat

In progress
Knowledge by choice

Yonathan Fiat

In progress

Abs

We often think that we can know depends only on what is _given_ to us; on our evidence, for example. In this paper, I argue that this is false. It also depends on our choices. I argue that our choices can change what would be an amazing coincidence, and that typically we can know that amazing coincidences don’t happen. This paper explores three applications of this idea. First, it suggests a new solution to the lottery paradox. Second, it gives us a new understanding of how we can get inductive knowledge. And finally, it shines a new light on _significance testing_, one of the most important statistical practices in science.