Full employment theorem | Hacker News Reader

Full employment theorem

en.wikipedia.org

58 pointsprofquail20 comments

On a meta-note: I deeply enjoy that linking to Wikipedia articles is part of HN culture. If that which is timely cannot be timeless, then it is of supreme importance that what is News to us, individually, consists of more than what is New at the time of posting.

There isn't anything that sufficiently complex algorithms can't find out that humans can, so this doesn't seen to be a terribly useful conclusion.

It's unclear what this has to do with A.I. Even humans are limited by the halting problem or other undecidable questions.

So why is this called a "theorem"? There is yet no proof that there are classes of problems inherently better solvable by the human mind. Or otherwise, that the human is something extra, computationally-wise, than a Turing machine.

>>So why is this called a "theorem"?

It's not A theorem, it's an informal name for a class of theorems. Specifically theorems that imply creativity will never be useless.

I think the point is that, for some problems, no computation can perform better than human beings. A strong AI would perform equally as well as humans on these problems, but not better.

Of course, a "strong AI" in this usage would probably have to not only be computationally equivalent to the human mind, but also be provided with the same insanely complex and chaotic input that the human mind operates on.

To truly mimic human "creativity," this strong AI would probably need all the qualitative experience and memories that human beings have. It would probably need to undergo development in a dynamic social environment, or at least have the memory of doing so.

It doesn't seem like this is really an AI vs. Humans issue. Rather, it's saying, regardless if you have some optimal algorithm X for a class of "Full Employment" problems, it is always possible to have an algorithm that performs better for a specific case at the cost of worse performance for other cases.

It states that some problems will never have a perfect solution. It would be possible, I guess, for a strong AI to pump out stronger and stronger AI, so that humans will be unlikely to solve any problem before the stronger and stronger AI solve it.

Until that happens, it will always be possible to write a better space-optimizing compiler. Afterwards, it will still be possible, but the strong AI will beat you to the punch.

The trick is that perfection is usually overrated. Just like human experts, you can be the best at what you do, but you don't need to be the best ever (especially considering future experts).

That's also the key difference between science and engineering. In engineering, you need a good-enough solution, but you are not so concerned with proving that there is possible better implementation.

Isn't this like saying that AI is impossible?

Quite the opposite, it's saying there are some jobs which are AI-complete.

I agree, but the wording is poor. "no algorithm can optimally perform a particular task done by some class of professionals" implies that general intelligence is not an algorithm.

Go forth and improve that article.

thats what confused me

More that there is no ONE single perfect AI, or even two or three or n<infinity.

Computer/algorithms cannot tell who is hot, if the input is photo of Marilyn Monroe versus Scarlett Johansson

Hot when? A better example would be Scarlett Johansson and a contemporary.

Except perhaps by recognizing Marilyn Monroe and knowing enough about her to realize she is is room temperature at the moment.

Whenever you claim no algorithm can do something that a human can, you must ask yourself: what does the computer lack that a human might posses?

In this case, there are indeed machine learning algorithms that can do this task.

http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5533...

If I wrote an algorithm that choose Scarlett Johansson, who is to say my algorithm doesn't work.

Computer algorithm cannot tell who is hotter, Marilyn Monroe or Scarlett Johansson

Reply on news.ycombinator.com