From: Gene Michael Stover
When: 2021 December 12
This file is https://cybertiggyr.com/fe5fb902-d2f1-47db-a5e9-d6d4ef447c98.html.
For months, I've been saying that A.G.I. feels like it's just a few discoveries away.
I still maintain that even a narrow problem solver in a box could do a lot of good work, but I've since thought of this...
Ask it to solve the D.B. Cooper mystery.
It's a hard mystery & a lot of people have worked on it without solving it, so we don't actually expect a solution.
The point of my thought is that I realized how hard it would be to make a problem solver in a box to work on the D.B. Cooper mystery. Things it might need to know, things it might need to be able to think include...
I realize now that a problem solver in a box that works on the D.B. Cooper mystery must be an A.G.I. and that it's tough to make such an A.G.I.
I get it now. We're more than just a few insights away from A.G.I. (But I am still inspired by The Master Algorithm & recommend it.)
That's how I think of a useful artificial intelligence, whether narrow or general.
I'd like for us humans to setup one & ask it to solve a problem such as a cure for a disease, or what's up with “dark matter”, or whether Trump winning the 2016 election implies that we really do live in a universe so insane that Douglas Adams might as well have created it.
We'd probably have to start our problem solver in a box by telling it everything we know about the problem's domain (which we might do by hooking it up to databases). It would be a disembodied mind, so it'd have to ask us to do things for it such as run experiments or gather more information. Eventually, it'd tell us a solution. Along the way, it might tell us some new insights.