yes. what else would people be doing when they think but computation?
I’m surprised by this conclusion.
what is your view?
would the algorithm just be a fixed procedure that outputs judgments from inputs, while still being fallible?
I hadn’t thought about the matter before. I thought “knowledge isn’t automatic”, although what is meant by that is that you have to put purposeful effort into thinking, i.e., computation, to get knowledge. I don’t see a way to reject that this means knowledge creation is algorithmic other than invoking supernaturalism/mysticism/dualism, which I wouldn’t do.
what do you think? you try analyzing.
I wondered the same as Oracle. I also wonder whether it would be the same algorithm for everyone and would it give the same results, or is it dependent on the knowledge of the person applying the algorithm. In that case does that knowledge go into the context of the IGC, and so the people would get different answers because they’re actually evaluating different IGCs.
I can think about it myself.
i’m not confused about a fixed procedure still being fallible. that part seems fine.
what i’m trying to pin down is the stronger implication: if truth discovery could in principle be automated, then this sounds like there is in principle a general judgment procedure, which doesn’t seem plausible. what would such an algorithm even look like?
By automated do you mean run on a computer?
yes.
the potential contradiction part to me is: does your view mean there is in principle some fixed computational procedure for deciding which idea to reject when criticisms conflict?
I don’t know where “fixed” is coming from. People adjust at least some parts of their judgment during their lives. Computers can run self-modifying code, not just fixed code.
my bad. i confused myself here. i mixed up fixed with deterministic
i went back and relistened to deutsch. i think i was framing this incorrectly.
the thing i actually want to check is whether you buy deutsch’s regress about explicit knowledge. his argument is that explicit statements always depend on background understanding that isn’t itself explicit. like dictionaries need more definitions, rules don’t specify how to interpret them, etc. and that bottoms out somewhere inexplicit.
do you buy that? or do you think a fully explicit AGI is possible in principle, where every level of the process is specifiable in explicit rules or representations?
The same sort of regress concern applies to inexplicit ideas too. Inexplicitness (basically ideas that aren’t in words or math symbols) isn’t a solution to the regress issue.
Deutsch says that inexplicit knowledge does break the regress because it’s stored in physical configurations rather than in more words or symbols. like meaning bottoms out in the physical structure itself, not in more explicit statements. do you think that solves the problem?
i’m linking the interview here just in case i’m putting words or misunderstanding david
Words and symbols are also stored physically, and the regress is about the meaning not the storage.
Deutsch isn’t open to debate and violated my rights, so I’m not going to look through what he said. It’s OK for you to post about the ideas. Just learn them yourself so they are your own knowledge/understanding, so I can engage with you rather than indirectly engaging with Deutsch. (Citing Deutsch is OK.)
fair. here’s my own understanding: explicit rules don’t seem to interpret themselves. applying a rule already uses background knowledge about relevance, meaning, context, exceptions, etc.
so when you say judgment is computational/formalizable, are you including that background interpretive machinery too, or only the explicit evaluation layer?
It must all be computable b/c humans think using their brains which are computers. The computations have to deal with context, background knowledge, etc.
i think your storage vs meaning point is right. i was being sloppy there.
the part i’m still unclear on is what you think solves the meaning regress. if explicit rules/statements need interpretation, can that interpretation in principle be supplied entirely by more explicit rules/statements? or does it have to bottom out in some non-explicit computational/physical organization that makes the explicit layer meaningful?
i’m not denying computation. the question is whether explicit representations can be self-sufficient all the way down.
The standard non-Popperian regress solution is some sort of foundations.
The Popperian answer is: incompleteness isn’t error. Knowledge starts in the middle. It’ll always be incomplete with more you could explore and add. It doesn’t have to be complete to conjecture and criticize about it, use it, make progress, etc. You can add more details as needed as they’re relevant to the problems you’re working on, rather than trying to deal with all the infinitely many details from the outset.