What might AI be good for?

Epistemology is the name for the study of truth. The word comes from the Greek words "episteme" and "logos" which are translated to "understanding" or "knowing" and "study", respectively. In philosophy this is often a very meta conversation, basically thinking and discussing how to know what is possible to call truth and how to distinguish between truth and falsehood and the range of opinions in between.

I think this is an important foundation for pursuit of truth -- you can't pursue something without agreement on what it is you're after. Beyond that though, I think the most important thing is application. Once we have a functional definition of what "truth" means, what fact patterns fit with that definition? Where we humans frequently fail in the pursuit of truth is in our biases. Current thinking on decision-making runs a bit like this:

Human brains are prediction engines that make guesses about what is likely to happen based on a framework of past experiences. The collective process that leads up to a decision draws in all sorts of information: the framework of experience, the available information in the moment, and all sorts of crazy things we wouldn't imagine -- e.g. the state of our gut biome -- that impact our state of mind and the general way all of the parts come together. These biased, subconscious guesses organically form into decisions, and all of this happens in a precognitive state.

Then, as humans, our cognitive or conscious mind sets to work fitting that decision into a belief system about ourselves. We all have opinions about what sort of person we are and what our values are, and when think back about what we have done, our consciousness is essentially the process of fitting those actions together with our idea of who we are.

This works pretty well for lots of things. It's especially good for "blink" (by Malcolm Gladwell) type situations where a quick decision needs to be made in the context of salient and immediate information and where the feedback on the quality of that decision is contiguous with the decision and identifiable. The more spurious the connection is between the decision and the outcome, the harder it is for the brain to train itself, and the more likely we are to confuse the causes for the effect, to assign irrelevant causes, and the more likely we are to let our biases drive our conclusions.

A better, longer way to read more about this would be to pick up a book by Danny Kahneman or any of his disciples. What I'm sharing here is not an original thought.

I think that AI could be useful as a way to systematically apply epistemology. With a shared view of how to define what is true, we could apply learning algorithms against this problem. When we need to explore our definition of "truth" we could change the parameters for the learning algorithms and apply again against the corpus of available information in the world. This would be valuable

This would be valuable in a way that almost every application of AI being currently explored or deployed is absolutely not.

Imagine an AI service that could help you actually learn about something in an informed and trustworthy way -- especially at real time. That would have value.

I know this article is humorless and direct. Thank you for reading this far. I'll write a second one on how I think this AI can work.