Comment by unono | Hacker News Reader

Comment by unono | Hacker News Reader

There's a lot wrong with that article. If the computer is smart enough to tell you what to do in social situations and can also read, write, and drive cars, the game is pretty much up.

There will be nothing humans can do that computers can't.

The best thing to do will be to plug everyone in to the Matrix, because otherwise millions of wars between people with really smart computers will break out. Wars that will feature weapons designed by smart computers i.e. drones with nuclear/chemical/biological weapons - deadly enough to kill everyone.

The safest route for humans in the coming years is for global scale cooperation to create a basic income, not a stipend, but self-contained self-sufficient virtual reality simulator so that we don't descend into chaos when some of us get access to hyper-intelligent computers.

It's quaint that the good professor thinks we'll just go on seeking jobs after taking coursera courses, and marketing sales letters to each other, but we hackers unfortunately know that that isn't an accurate prediction. The gravitas of the situation is far greater.

edit: Ted Talk by Daniel Suarez (author of Daemon and Freedom) on the drone threat http://www.youtube.com/watch?v=pMYYx_im5QI

I think we have two eras, each very different than the other.

The first is the immediate future where computers can not act as independent, autonomous agents. This is the era we are in today, and have been for quite some time. Some of the concepts described in the article could be used to describe years ago, from a different perspective. Free online education? Has existed for over a decade, arguably pushing two decades. And so on.

In today's era it pays to use machines as tools of great leverage. It means not only that you get ahead of other humans, but you get way ahead of them. Your wealthy, in good health, and very happy.

The second is the era where "computers" become independent autonomous agents. In movies they always either enslave or exterminate humans. I would expect a more quiet extinction, not in existence but in relevance. In that case, what a columnist writes in the New York Times just doesn't matter, because it would have zero impact on anything that was doing decision making. Perhaps there is a biological transformation in this period as well, where genetically modified humans split off not in to another species but multiple species. In that respects, homo sapiens could become relics at the zoo.

Its all conjecture, but I think one thing here is a certainty -- that machines become both independent of humans and autonomous, able to act, buy, sell, and build things all on their own.

People greatly exaggerate the progress and inevitability of AI.

On the contrary, I think we're now in 'the age of the great underestimation'.

Based on what? The only example we have of human intelligence is human intelligence, and we're still more or less at thinking-this-makes-neurons-here-light-up knowledge of how it all works.

The latest moto X has always listening chip inside. That's pretty significant. There's robocars, ibm watson, siri/gnow. The CEO of the quantum computer recently said that machine learning is progressing faster than people think, having seen some of the secret internal stuff brewing up at big G. Microsoft is developing improved versions of the kinect internally as chips.

Who cares if we don't know how the brain works in detail, we still haven't built bird-like machines that flap, but we do have machines that soar through the sky faster and higher than birds. Same is happening with AI.

Are we making great, amazing advances in pattern recognition, mechanical systems that very cleverly incorporate sensor feedback, and other novel areas? Sure. Are we making super-impressive progress in general intelligence. No.

There's two main fields of thought - one that "general intelligence" is some unified, centralized system, maybe a "symbolic system"; another that "general intelligence" is merely the composition of a bunch of pieces that do certain stuff.

So maybe our advancement in the things you mentioned are making strides towards the latter. I think that's wrong, but it's possible. Even if it is the case, we're still not making strides in the pieces of that composed intelligence that answer questions like - what's bigger, a shoebox or a mountain?

In regards to your bird analogy: we don't have very popular bird machines, but we do know the principles that enable birds to fly. Similarly, we don't need to build brains, but anything we build that is comparably "intelligent" as a brain would be quite surprising if it didn't come from insight we have into the principles that enable general intelligence.

'General intelligence' is wacko amateur thinking that reasonable people in AI don't believe in. You're shooting yourself in the foot mentioning that.

The consensus amongst competent people in AI is that of building practical systems.

I'll be candid with my sources - I took a class with Patrick Winston called The Human Intelligence Enterprise where we went over a history of AI ideas. Nowhere in that class did I get the impression that 'general intelligence' is 'wacko amateur thinking'. Maybe the former director of the MIT AI lab just failed to impress upon us the disdain for general intelligence you say is commonplace.

http://ocw.mit.edu/courses/electrical-engineering-and-comput...

Patrick Winston is a disciple of Marvin Minsky. They're considered wackos these days.

Who would you consider the leading authority?

The machine learning/statistics crowd. Some names - Judea Pearl, Tom Mitchell, Pete Norvig, Andrew Ng, Geoff Hinton, John Langford, Jeff Dean. Of course there's many more than that, but these have gained some fame through textbooks/open sources/popular press etc. More generally the approaches taken by Google, like vector space models that power search engines. If Google's doing it it's likely not wacko because they have a low tolerance for tomfoolery down there.

OK, now cite where they think general intelligence is tomfoolery.

They won't, out of politeness and not wanting to make enemies. I will, as an anonymous coward on the internet.

That statement seriously bears repeating. The best algorithms today can't beat an average human at Go; discussions like this tend to remind me of the flying car, Mars colony, hoverboard predictions of sixty years ago. We aren't there yet, people - to be honest, we're not even very close.

Reply on news.ycombinator.com