[ExI] AI

[Anders Sandberg]

William Flynn Wallace <foozler83 at gmail.com> , 29/9/2014 2:56 PM:


Just recently glial cells have been found to not only supply neurons with nutrition and serve as the blood brain barrier, but actually affect the functioning of them.  We have no idea how or why now.  Given that there are 100 billion neurons with up to 25K connections and up to 5 trillion glial cells, it will be a very long time, maybe never, before we can begin to understand just what is going on up there except in a fairly general way.I did some analysis of this in my WBE roadmap and was not too worried. Glial cells are slow to update, so they take orders of magnitude less computing to simulate than neurons. If they really matter we still may have a higher memory demand, but throwing around large numbers is not a good argument (especially when Moore's law works; see http://www.aleph.se/papers/Monte%20Carlo%20model%20of%20brain%20emulation%20development.pdf ). And we are already doing specific simulations of the brain, it is just that they lack actual scanned connectivity so they are either generic circuits or small well studied microcircuits. Brain emulation is not the road to AI for the next few decades (again, see my paper).




So when we say that we want a computer to think or create we are going to have to specify exactly what we meanYup. This is why de novo AI is hard to do. The wild success of machine learning recently has been because we found a way of getting software to figure out roughly what we mean in some domains by giving them lots of examples. 


Anders Sandberg, Future of Humanity Institute Philosophy Faculty of Oxford University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20140930/f6f8cdb8/attachment.html>