I saw this video posted on reddit somewhere a couple of days ago, and didn't bother to comment, but this video makes the fallacious equivalency that "computers are faster" with "computers can do more things".
A lot of current techniques work especially well when the objectives are well defined ("capture the king" or "make the most money possible"), but these are basically techniques discovered 50 years ago, it's not like there's been quantum leaps in AI research in the past decade or so (at least not to my knowledge, apologies to researchers who might have done things).
But since these techniques revolve around defining objective functions, when the notion of "best" is fuzzy or subjective none of this is going to be super useful. And search problems will game the system: if you ask a computer to design the fastest plane possible it will probably design one that has no space for a pilot, unless you specifically ask it to.
I used to think that the best example of fuzziness is that people are way better than computers at deciding whether something is a chair, but then I found out about LEVAN.
Most of the big breakthroughs that will fuel the next round of humans being automated out of jobs are on analysis of data in fields such as computer vision and human speech recognition and such. Once a car can "see" its surroundings, it doesn't need real AI to drive, it just needs to know how to walk a graph and apply a simple set of rules based on what it sees.
General purpose AI, if we ever do achieve it, is still probably a long way off, but far more jobs than most people tend to think can (and likely will) be automated away long before we ever get there. A lot of the things we do for work just don't require actual creative problem solving most of the time.
A lot of current techniques work especially well when the objectives are well defined ("capture the king" or "make the most money possible"), but these are basically techniques discovered 50 years ago, it's not like there's been quantum leaps in AI research in the past decade or so (at least not to my knowledge, apologies to researchers who might have done things).
But since these techniques revolve around defining objective functions, when the notion of "best" is fuzzy or subjective none of this is going to be super useful. And search problems will game the system: if you ask a computer to design the fastest plane possible it will probably design one that has no space for a pilot, unless you specifically ask it to.
I used to think that the best example of fuzziness is that people are way better than computers at deciding whether something is a chair, but then I found out about LEVAN.
Then I saw what it thought were beanbag chairs : http://levan.cs.washington.edu/?state=fetchNGrams&concept=ch...
I'm not too scared of the computer apocalypse yet.