[ExI] fun outsider's view on ai

spike spike66 at att.net
Mon May 9 23:12:58 UTC 2016



From: extropy-chat [mailto:extropy-chat-bounces at lists.extropy.org] On Behalf Of William Flynn Wallace


>…If you want an AI to be superintelligent, why reference the neuron, Spike?  Human brains are so fallible it's just silly…


We don’t know a better way to true AI than by studying and emulating actual I.  


In any dynamic system being modeled by computer, you need super-detailed models of the subsystems if you want a good high-fidelity simulation.  For instance, if you want a jet aircraft sim good enough to train pilots, you will likely need to have structural characteristics of all the parts, since high-performance jets flex under load, things act in unexpected ways when you push to the limits etc.  You need to model things like the moment of inertia of the elevators, rudder, ailerons and so forth.  There are no shortcuts, if your sim precision is critical, as it is in some cases.


If we want to get AI, we have one really good example of I.  We have one good example of what we need, one example of a machine which can generate software.  To create a sim which can write software, we simulate the one example we have of that.  To do that, we sim every subsystem, every single one, all the way down.  Then we run it as a background process on a bunch of interconnected computers.  We might find we can simulate a second of human-like thought per hour.  


This approach requires that we understand how dendrites and glials and synapses and all the rest of it works.  Currently we don’t.  Certainly not entirely.



>…All told, we are many decades away from a good grasp of the brain, maybe 100 years.  A super smart AI will likely not function at all like a human brain.  No reason it should.  (boy am I going to get flak on this one) bill w


No flak, you might be right.  We might find an alternative path to AI.  I will note however that I took Thune’s class online, the one Stanford offered free a couple years ago.  I noticed the text hasn’t changed all that much in the last 20 years.  The insights offered in the mainstream courses is really not progressing all that much, which tells me we need to try something else.





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20160509/4f414dd3/attachment.html>

More information about the extropy-chat mailing list