[ExI] Purified humanity Re: Your Genome Is a Post-Apocalyptic Wasteland

Anders Sandberg anders at aleph.se
Mon Oct 7 11:35:35 UTC 2013

On 2013-10-07 11:34, Eugen Leitl wrote:
> On Mon, Oct 07, 2013 at 11:20:43AM +0100, BillK wrote:
>> You may be mixing evolutionary computing techniques to solve specific
>> problems with evolutionary changes.
> When arguing your case, please don't invent facts. They hate that.
However, this morning I think the facts are confused about who is 
arguing what. I have a suspicion that BillK was talking about doing 
artificial selection on real genomes using selection methods generated 
by various methods, including evolutionary computing.

>> Evolutionary computing starts afresh with every new problem. Evolution doesn't.
> It depends on what you prime your initial population with.
> Starting from urslime all the time is expensive. Notice that
> none of current methods are anywhere criticality, aka evolving
> evolvability. This is computationally expensive, and likely not
> easy to replicate in a different context. So this means mature
> methods will have initial pool conservation.

I think this is an important consideration. Far too much evolutionary 
algorithms start with ur-slime because it is far cooler and more 
open-ended than playing Author and dropping in some already-designed 
scheme. Then again, most useful applications of EC tend to be finding 
weird-looking solutions in small problem spaces rather than open-ended 
invention. Small 'c' creativity, rather than the capital 'C' Creativity 
that answers a different question than you asked.

>> For example, way back, some creature in the sea found that four fins
>> were quite useful at that time. After that, evolution, building on the
>> past, was stuck with four limbs for ever after. Even when four limbs
> Hexapodia is the key insight.

I haven't had the chance to see the video. Bandwidth here in the UK is 
so limited.

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

More information about the extropy-chat mailing list