[ExI] What might be enough for a friendly AI?.

[John Clark]

On Nov 19, 2010, at 1:52 PM, Dave Sill wrote:

>> There would be absolutely no point in building an AI if you just lock it up
>> in a box with no way for the outside world to interact with it.
> 
> Sure there would. It could solve problems and teach us.

A teacher astronomically smarter than you would manipulate you like a puppet. 

>> A much better analogy would be Einstein in charge of weapons development and
>> production, world monetary transfer, electric power generation and distribution, worldwide communication lines, air traffic control, nuclear power plants and pretty much the entire economy.
> 
> If you hand it that kind of control, the game is over. You're its slave, whether you realize it or not.

Exactly, and although not super- intelligent computers already run much of that sort of stuff; and if you refuse to let a super-intelagence make weapons or run you economy somebody else will and they will have better weapons and a stronger economy than you and take over.

>> When you ask Einstein why he made one decision rather than another he tries to tell you but after
>> about 20 seconds of his explanation you become totally lost and confused.
> 
> The AGI is a supergenius but unable to explain itself?

Can you make your dog understand calculus? 

 John K Clark


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20101119/791f9e17/attachment.html>