Proper Knowledge [CF Article]

[Elliot]

@ActiveMind

Confidence Levels with Binary Epistemology

If we can reach high rational confidence in any knowledge, like how we’re confident in identifying apples, then there must surely exist different levels of confidence that you can have about a piece of knowledge. For example you have some exam questions where you don’t feel confident about but you have some answer which you think is your best bet. You write down these answers and across many exams you see that you have scored higher than what would be expected by chance though you scored less than on the questions where you were confident. So clearly you were right to have some confidence in those answers even though you weren’t fully confident.

The question then is how is this compatible with binary epistemology. After all in binary epistemology we can only evaluate ideas as either true or false. So shouldn’t either be in a state where you don’t know and can’t judge an idea at all, or in a state where you’re fully confident in the idea because you have evaluated it one way or the other?

The answer is that you can have partial knowledge. An idea can consist of sub-ideas.

Confidence is an emotion. Just like happiness, it can come in degrees. It’s not an evaluation of the truth or goodness of an idea (which shouldn’t come in degrees). Confidence has no primary role in epistemology.

For risk management, more importance than confidence is whether you have a refutation of proceeding with an idea now or not.

For learning and evaluating mastery, more important than confidence are error rate and being able to explain concepts.

@ActiveMind

Not being used to binary epistemology, how confidence levels fits in isn’t as intuitive as how it would fit in with degree based epistemology. If you believed in Bayesian epistemology it’s quite straightforward. If you have supporting evidence to make something 80% probability of being true, then your confidence level should be 80%.

No. Non-emotional “confidence” is related to variance. It’s different than probability/credence. If you just found out about an issue, and didn’t investigate much yet, you could have an initial 80% conclusion but very low confidence it’s correct since you’ll probably change your mind as soon as you look into the matter more. This “confidence” is related to margin of error around your answer like you could be 20% confident that the correct evaluation is in the 70-90% range. This is also called error bars.

You’re wrong to give Bayesian epistemology credit as “quite straightforward” or intuitive on these topics. Harry Binswanger didn’t even understand confidence intervals for measuring devices like rulers or scales. In general, Bayesians often talk about how some of their stuff is unintuitive and requires training, and they debate with other schools of thought about probability like frequentists.