Different types of evidence
Some popular ways to categorise different types of reasoning/evidence/views:
Inside view vs outside view.
In this post, I’ll give my own (still work-in-progress) take on how to think about different types of evidence, and how they relate to the above typologies.
I like to separate reasons for believing something into the following 3 categories:
Deference: Trusting an epistemic process that is somewhat opaque to you, but that you nevertheless have good reason to believe is reliable. This is most commonly just another human, perhaps an expert on the subject.
Induction: You’ve observed some fact to be true in the past, so you will assume that it keeps being true.
~Deduction: Some facts that you believe to be true about the world imply that another fact is probably true.
The tilde is there because in philosophy, “deduction” is often used to mean logically valid inference, and I want to include probabilistic epistemic moves here, too.
(I considered adding “raw intuition” to this list, since that’s often a very significant source of disagreement, but I’m not sure if it makes sense to treat as separate. In practice, some amount of intuition is necessary for all these kinds of reasoning — to decide which deductions are plausible, what reference class you can use induction with regards to, and who seems like they should have relevant expertise. )
Comparing with the types of reasoning from above:
Independent impression vs all-things considered view:
The independent impression takes into account induction and ~deduction, but doesn’t use deference.
The all-things considered view is willing to use all 3 types of evidence.
Inside vs outside view:
As have been noted, EAs and rationalists have started using these terms in terribly ambiguous ways.
In their original usage, I would say that there’s a clear correspondence between induction and outside view, and a (slightly less obvious) correspondence between ~deduction and inside view.
In modern parliance, I think some people will use inside view as synonymous with “independent impression”, including both induction and ~deduction.
An important thing to note about these kinds of reasoning is that they often rely on each other:
When deciding who to defer to, you may use induction to look at their track-record.
When using induction, you may need to use deference or ~deduction to figure out what happened in the past.
But most importantly, ~deduction almost always works by ~deducing things from already known facts, and those facts could have come from anywhere!
I think this last point is most important. I like to imagine a final deductive guess to be the root node in a tree, where each node is deduced from its children (and possibly partly influenced by some relevant induction- or deference-move), and where each leaf-node is justified just by induction and/or deference.
An important implication of this is that the concept of your “indepent impression” or “inside view” will often be ambiguous, because it will depend on where in the tree you stop including deference/induction, because it’s not feasible to construct a tree completely without it. (Though in practice, it’s often useful to talk about an “independent impression” as just being independent from a few people — e.g. people in your community or in the current conversation — in which case it might be feasible to construct a view that totally ignores those people’s views.)
I’m not totally happy with this categorisation system, and feel like I still have some things to say about the issue, so I might write more about this later. Or not. We’ll see.