Machine Learning, Knowledge, and Understanding


This is a good illustration of why I have always disagreed with the view that knowledge is justified true belief (or that knowledge is a repertoire of facts).

You can store the facts in a reference book. You don’t need to learn them. The important part of knowledge is our ability to connect the facts to reality. And that’s missing in the JTB characterization. And it is mostly missing in machine learning.


I don’t know how your conception of JTB can exclude connection to reality; my understanding of JTB certainly does include acting and reality:

Beliefs include claims about the nature of the world.

Truth involves one of the following, depending on how one conceives of truth

  • that which is fated to be agreed to by a community of inquirers following empiricist, scientific processes of inquiry (what I think truth is)
  • representing the world as it is
  • corresponding to the world as it is
  • the result of successfully acting in the world

Justification for empirical claims must include proper connection to the world. (Whether a knower has to understand that connection or whether a reliable connection alone is sufficient is a philosophical controversy).

Gettier cases make justification complicated by showing how the usual accounts of justification can sometimes be fooled by luck.

More general concerns with JTB question the assumption that belief is more basic than knowledge, and instead claim that the right starting point is to take knowledge as the fundamental concept.

Gary Marcus has an ongoing critique of Deep Learning and its limitations:

1 Like

Beliefs are generally taken to be intentional (have aboutness). Roughly speaking, you could rephrase my point as saying that knowledge is intentionality. But much of the literature sidesteps serious discussion of intentionality, and instead concentrates on logical structure and justification.

I’m not trying to defend epistemology as practiced by armchair philosophers who see value in artificial thought experiments. I don’t see much value in that approach to philosophy. I prefer an epistemology that is continuous with and engaged with the related sciences, like neuroscience, psychology, sociology in the case of knowledge.

I had a different reason for my comment on your post as it stood. Namely, if one wants to engage in good faith criticism of an issue or idea in philosophy, then one should take the time to engage with the related philosophy first.

Many commenters here are justly criticized for failing to engage with the details of the science of biology before criticizing some aspect of biological science.

I think the same standard for engaging in good faith criticism should apply to philosophy. Theology too, for that matter.

I recognize a separate type of criticism which applies to a whole field of inquiry. One can claim that all of science is misguided eg because it relies on methodological naturalism. Or that philosophy is pointless because it has no way to progress through eg experimentation. Or that theology it is dogmatic and hence not a reputable type in inquiry. I’m not saying these are correct critiques. Only that criticizing the basis or presuppositions of a whole field of intellectual study is different from engaging in good faith with an issue within that field.

1 Like