News for hackers. Stuff that matters.
Re:Machine intelligence, part 1 by graycat on Wednesday February 25, 2015 @03:40PM (#9109387)
We know how to work with fire and not burn down everything we care about.

We know how to fly at 0.80 Mach, safely.

Well, we understand fire and airplanes.

Altman is correct: So far we don't have even as much as a weak little hollow hint of a tiny clue about how to construct what he calls machine intelligence; I thought so when I was at IBM's Watson lab where our team shipped software products and published a string of papers on artificial intelligence; I still think so now; and some maximum likelihood estimation heuristics, rules, finding parameters to do nonlinear fitting to data, etc. don't change my mind.

When we understand how intelligence works well enough to construct it, then likely we will also understand it as well as we do fire and airplanes and, then, also how to have machine intelligence safely.

Just don't put the robot factory in Chernobyl or Fukushima!

So, go ahead and develop this intelligence machine. Then I will formulate a Turing test, but you don't get to work with the machine after I show you the test!

Secret: Here's the test:

Here's W. Rudin's Principles of Mathematical Analysis; hand a paper copy to some machine learning computer and have it read the book, work the exercises, read Knuth's The TeXBook, learn Emacs, use Emacs to type in the TeX for the solutions to the Rudin exercises, and post a link here to a PDF with the solutions. I will be glad to grade the solutions!

There are no comments on this comment yet.