Re: answers I'd like from an SI

From: Norman Noman (overturnedchair@gmail.com)
Date: Wed Nov 14 2007 - 22:40:04 MST


On Nov 14, 2007 1:03 AM, Stathis Papaioannou <stathisp@gmail.com> wrote:
> On 14/11/2007, Norman Noman <overturnedchair@gmail.com> wrote:
>
> > Yes, I do think a godlike intelligence could improve on these methods,
> > tremendously. The performance you can squeak out of even simple
> > systems is often mind-boggling. The way a superintelligence would
> > teach a kid to read, if efficiency was its only goal, would be like a
> > videogame speedrun, an antenna produced by evolutionary algorythms, a
> > busy beaver turing machine, and a langford basilisk all rolled into
> > one. It would appear to be insane. And it would be impossibly fast.
>
> Having done that trick, could it then teach a dog to read, or tell a
> joke so funny that everyone who hears it dies laughing?

I don't think you could reprogram a dog's brain to be able to read by
speaking words at it, but I would guess that it still might be
possible through some kind of audiovisual input. It would require a
more fundamental change than teaching a human to read, because you
would have to make the dog smart enough to use language in the first
place, but you've got a moderately large lump of gray matter, and to
an extent it can be reprogrammed.

If a foothold can be established whereby greater and more nuanced
control can be exerted, then the dog can be made into more than a dog.

I get your point though, at some point limits must be reached. I doubt
an AI could talk a ladybug into reading.

As for the joke so funny it makes people die laughing, yes, I
absolutely think this is possible.



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT