[ExI] Augmentations to Science /was Re: Drugs and creativity

[Anders Sandberg]

I think there is a fundamental difference between insight and p-values. The first is about how to generate a hypothesis, the second about how to test it (after an experiment or observations ).

In creativity there is both divergent thinking and convergent thinking - come up with new things, weed out the bad things,  and build the things for real. Some people are good at some steps, but not others. Boost the weak steps, and they become more creative. Psychedelics may help the first step or even cause a transvaluation that enables new ways of evaluation. But other tools exist for other steps. Being meta enough to figure out which ones need boosting is also a skill.

P-values for all their faults is an example of a tool that helped the scientific creative process. There are many more, like bayesian methods, for that step. Maybe we need more metascience to evaluate where we need more tools for our projects.




Sent from Samsung tablet


-------- Original message --------
From: Will Steinberg
Date:2015/12/20 17:52 (GMT+01:00)
To: ExI chat list
Subject: Re: [ExI] Augmentations to Science /was Re: Drugs and creativity


Off the top of my head, maybe a candidate is what appears to be to be a paradox in statistics.  I can't really explain it so I'll give two examples:

If you vote for a candidate, that one vote is basically worthless.  But if you convince a million people to vote for a candidate, those votes gain a different kind of value.  The question is sort of:  If you're one of those million people being convinced to vote, how should you think of the value of your vote?  Does it become more important to vote now that you're part of that million?  It could almost be argued that now it's less important to vote because other people will en masse for your candidate.  I'm sure this problem can be lexically pared down to a more parsimonious statement, it's related to levels of abstraction and whether a hypothetical, future-intended action on a higher level affects the value of the lower level actions it's comprised of.  Also related:  If I make a bracket tournament of everyone on earth and the winner is determined by coin flip, I will necessarily produce a winner who has won every flip.  The chance of that is miniscule and I believe equal to 1/(people in the competition).  So I can, by performing the tournament, create precisely the theoretical distribution of winners and losers to a number of coin flips.

I think statistical questions like that, related to scale, level of abstraction, and observer position, are "psychedelic" problems.  I'll call those statistical relativity, but someone let me know if what I mentioned has been addressed already.

Another problem regards insight, and how to hypothetically measure it.  If you have an insight that gives you information much more quickly than an experiment producing the same result, how can we verify and validate the truthiness of this insight?

In my mind (psychedelic) insight has a place alongside science as an information gathering tool, and perhaps a value similar to p value could be used to show the strength of an insight.  Thus insights with low enough value could be used as data.

To me, that seems almost predicated by the creation and study of "artificial insight".  I guess it would be a computer program that made guesses using data, though I don't exactly know how to randomize that in the same way as human insight.

So yeah those are my two for now: statistical level-of-abstraction paradoxes, and the nature of insight, what it is, how it works, whether it can be incorporated into theories as it's own form of earmarked derived data, whether the scientific discovery of the nature of insight a) is possible; b) would nullify or supplant the abilities of insight.

I almost sense another paradox in the 'artificial insight box'.  It's something like, if we created that box, if it had an insight, then knowing the derivation of that insight it would immediately be able to come up with a better one, ie the insight would be superseded even as it was created.  That paradox leads me to think, almost, that we can't build the insight box without building a box that made absolute conclusions from given data which were not really insights in the traditional sense because they would be proven true completely.  I can see why this stuff occupies so much of R. Penrose's thinking.

Sorry for rambling, writing this on mobile as a one-off.  But those are my two for now, insight and statistical LoA.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20151220/a41717fb/attachment.html>