[ExI] Augmentations to Science

[Anders Sandberg]

I usually start by my definition of understanding: you understand X, 
when your mental model of X is complete enough that you can run mental 
simulations of it, check what-if cases, and look at the causal relations 
between parts. I understand long division and classical mechanics, I do 
not understand most of economics.

An insight would then be a jump in understanding. Ideally leading to a 
fairly complete understanding, but there are smaller insights too 
(getting the special properties of central forces and potentials were 
two key insights for me in mechanics). Some insights are unexpected, 
others gradual. What happens is that your mental model becomes (1) 
better and (2) more compressed - while before there were lots of random 
facts and examples, now they fit into a pattern.

As Jurgen Schmidthuber likes to say, cognition is all about compression. 
The problem is that compressing evidence into a model can be NP or worse 
(consider figuring out the best explanation of a bit string generated by 
a hidden computer program).

We can become better at generating some kinds of insights by learning 
the right tools, whether they are pattern-recognition and -handling 
skills, a rich set of analogies to other things, or search mechanisms 
inside or outside our minds. This will not work on all problems; the 
meta-skill of knowing what tools might help is also useful. Some 
insights may require shifting strategy radically if one's current 
strategy set is not working.

I think this might provide a reasoned way of looking at the use of 
psychedelics or other strategy-scrambling tools: viewed as an 
exploration-exploitation problem in a reinforcement learning setting, 
you want to temporarily increase the temperature of your softmax action 
selection when you have evidence that you need to explore outside the 
low-value already explored region of the world or mostly used action 
set. It wouldn't surprise me if one could prove some optimal 
temperature-regulation strategies in some world models.


On 2015-12-22 01:52, William Flynn Wallace wrote:
>
> will wrote:  Was asking WFW how he thinks insight operates.  Same 
> question to everyone, though: to propose the most parsimonious logical 
> model of insight possible, or, if not logical, what a substitute for 
> that logic might be.
>
> For me science has to start with some definition of terms. My 'model' 
> bad term is 'instinct', which is defined dozens of ways and so has 
> little meaning for me. So ask me what instinct is and I ask you what 
> definition do you want?
>
> So do the same for insight.
>
> For something very off the cuff, I'd say that insight, if it proves to 
> be true in some sense, is seeing a pattern that you, and maybe no one 
> else, has seen.  Of course we see many patterns that turn out to be 
> false, like, in the extreme, paranoid delusions.
>
> As to where it comes from, it comes from the unconscious, like 
> everything else.  So what we need to do is to discover the operations 
> of the unconscious, which might prove to be something not totally 
> dissimilar from what Freud said about it.
>
> All this focus on consciousness is misplaced.  It is just a tool of 
> the unconscious.  I won't be alive then, but would be extremely 
> interested in what Freud wrote about the scientific nature of the 
> unconscious, which will be revealed when 100 years have passed since 
> his death (2038 or close to that), according to his will (unless 
> someone has stolen a look at it and I don't know about it).
>
> Without googling it, I think it's called Project for a Scientific 
> Psychology.
>
> bill w
>
>
>
> On Mon, Dec 21, 2015 at 2:57 PM, Dan TheBookMan <danust2012 at gmail.com 
> <mailto:danust2012 at gmail.com>> wrote:
>
>     This paper might interest some of you:
>
>     http://econfaculty.gmu.edu/klein/PdfPapers/DiscoveryandtheDeepself.pdf
>
>     Very Arthur Koestler feel to it, no? Of course, it seems to say
>     it's going outside s framework is part of the discovery or insight
>     process.
>
>     Regards,
>
>     Dan
>     Sample my Kindle books via:
>     http://author.to/DanUst
>
>     _______________________________________________
>     extropy-chat mailing list
>     extropy-chat at lists.extropy.org <mailto:extropy-chat at lists.extropy.org>
>     http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat
>
>
>
>
> _______________________________________________
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat

-- 
Anders Sandberg
Future of Humanity Institute
Oxford Martin School
Oxford University

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20151222/51874f60/attachment.html>