@Jamie,

I don't entirely agree with you. He clearly speaks about a situation where the robots wouldn't be "against" us, but simply wouldn't consider us a factor in their calculations, just like the human vs chimpanzee/gorilla example he cites.

So I think you may not be interpreting their thesis correctly, although I agree with you that for an article with a link to science, I don't see a lot of science, more speculation.

But it's wrong to use the language of "values" to describe that. A machine doesn't have values or morals.

It definitely has values: binaries, integers, floating point decimals...

Those aren't the "values" we're talking about here.

I thinnk Alex knows that Jamie ;-)

Redhouse is right. But, in the end, are our "values" really more than just an ever-evolving set of rules that could be programmed into a computer/machine?

I liked the part where they discussed software coding new software. Does anybody know anything about this topic? That just seems fascinating. (And of course I'm not talking about software cleaning up your badly written software, but actually inventing stuff)

great article red.

I thought it was a rather nonsense article. A lot of flashy words that those in the know would realize are either misused or irrelevant. There really is nothing to worry about. We aren't going to get caught off guard my some sentient doomsday computer. And, even if we are, it will take down one University or Research Campus and that will be the end of it.

@YJ,
TY
@abgemacht,
I disagree that this article is aimed at doom-scenarios and scaring people though, which you seem to suggest. It simply explores in understandable language what kind of phenomenon we should be thinking about. I thought some of their comments were thought-provoking, personally.
Also, what about the software program that is smart enough to design computer viruses? Then a whole lot more is targeted than just one University or Research Campus. I agree that there's nothing to worry about, but it's worth philosophizing about (which they're going to do).

Again, I think most of the article is very misleading.

For instance, they claim that there are machines that "trump human performance in chess, driving, etc", but that's not really true. Those machines weren't using genetic algorithms, they were made my humans. That's a very big difference. All these machines are essentially people who can do math very quickly. That's very different from a machine that developed its own means of solving a problem (which there are examples of, but not nearly as far along).

@abge,
With that I completely agree. Nothing needs to be "learned" for either of these tasks, which does seem the key word here. But, I agree with the article that a hundred years ago, we would never have thought a computer capable of driving, playing chess etc. (or maybe we could have extrapolated calculating power to being able to play chess, but modern chess games actually being able to give a decent analysis on your play I believe would have been considered unexpected).

I think the biggest problem is that people don't understand how incredibly specialized computers are. Watson may have won Jeopardy, but it couldn't beat a child at Chess. A computer may create a new virus, but it will never infect more than what it was designed for (a datamining bot isn't going to infect the power infrastructure). I do think we're on the verge of having some overwhelming sophisticated machines, but it will be for a specific task. They will be useless at everything else. We are very, very far away from having machines that excel at many tasks.

"discussed software coding new software..."

this happens all the time. It's called "auto-code generation" in which engineers provide specs to a system which auto-generates new code. Albiet, the initial design concept is still human-derived, the idea of machines creating their own code has been around for a long time.

@krellin,
Interesting. Never heard of it. Guess that could be an intermediate step between mere execution of instructions and some semblance of creativity.
@abge,
That's true, but what if the computer is equipped with a camera that registers the world around him and when someone puts a chess-board in front of him, he'll recognize it as chess and play against you and when somebody puts on jeopardy, he does that?
You say "We are very, very far away from having machines that excel at many tasks." In a way, my computer does just that :-) (and yes I fixed last week's bug :-))

"But, I agree with the article that a hundred years ago, we would never have thought a computer capable of driving, playing chess etc."

I've heard this argument a lot, and I don't think it really holds water.

We as Humans past a threshold over the last 100 years with 1) Ubiquitous Electricity 2)The Transistor and 3) The Internet. I can communicate instantly with anyone in the world and I understand what a computer is. I think that gives us a much better ability to predict the future that people in the past could. I could, of course, be wrong, but I think it will be a very long time before anything new is discovered. We're going to just be building off of the three big inventions I listed above. It is still impressive, but I don't think there will be that many surprises.

Hey Krellin.

@red

"That's true, but what if the computer is equipped with a camera that registers the world around him and when someone puts a chess-board in front of him, he'll recognize it as chess and play against you and when somebody puts on jeopardy, he does that?"

No. First, image recognition is one of the hardest fields of EE. Even machines with very good image recognition can only recognize what they've been programmed to. Second, both Watson and Deep Blue are each the size of a small room. To do both games, you'd need 2 rooms, one for each game, essentially. You haven't accomplished anything. You've just plugged two giant computers together.

"You say "We are very, very far away from having machines that excel at many tasks." In a way, my computer does just that :-)"

No, your computer does exactly what it was programmed to do (add and multiply) and nothing else. Anything clever that your computer appears to do as a result of adding and multiplying very quickly is due to clever programmers, not clever computers.

@abge,
I also only recognize what I've been programmed to right?

I'm not arguing that Humans are anything more than very well-programmed machines.

Programmed by who / what?

Does it matter?

Ever watch a car park itself or drive down the road? The automotive industry is working hard and fast to instill your vehicles with radar and lidar based vision systems to interpret the world around them for collision avoidance and other "driver assist" functions. Rudimentary...but still, the goal is to work towards and living "vehicle network" in which all cars communicate with the purpose to developing a live, interactive image of the road system and eliminate vehicle-to-vehicle collisions, autonomous driving, etc.

If GM, Ford, etc are doing this, what do you think DARPA has already put in the field???

Redhouse -- the idea of taking a spec and turning it in to code is really not that impressive, given that any "high level" language you use (basic, c, whatever) is really run through a compiler which turns it in to code (assembly). The difference now if the idea of using design software on a system level in which you spec out functionality with visual block diagrams, etc and it will generate what you want, instead of old school system design to auto-code generation of a decade ago where you were still writing rudimentary code in your block diagrams, etc. Cool...but hardly novel.

Ultimately, machines are still "designing" nothing, and are only automating the detailed, laborious work of spitting out code. It is no different than a man hand-fashioning a piece of metal in to a gun, versus a man programming a machine with arobot arm to pcik up the next piece of metal stock from a pile, load it in to a machine, and lathe it in to a gun. Impressive...but ultimately the machine was still built by a man.

I know of no "invention" ever created by a computer...and AI will not exist until such event occurs.

@Abge "I'm not arguing that Humans are anything more than very well-programmed machines. " Seriously?

@krellin

Yeah, I've taken Radar courses. It's cool, but those cars aren't telling you *what* is there, just that there's something in your way.

"@Abge "I'm not arguing that Humans are anything more than very well-programmed machines. " Seriously?"

Yes? Our code is in the form of proteins and acids, rather than Assembly, but I don't see the difference.

@Abge: I Agree (we are machines...clearly, when the hardware is broken, the human stops..." It is the "well programmed" that causes the debate.

There does seem to be some firmware within us for certain functionality, but how we are programmed becomes the debate, I suppose. Exterior input modulated by interior pre-dispositions??

@abge, I think you're underestimating the power of recognition software. It's still in an early stage, but I think if a lot of people start working in the field for applications like the one krellin mentions, you'll see computers recognizing all sorts of things and situations within no time.
I disagree with you though that a human being is just a programmed machine. Well, in the sense that somehow a human being is more than the sum of its parts (consciousness for example).

I've already addressed Krellin's post on car radar. It's collision avoidance, no more.

"those cars aren't telling you *what* is there..." YET...no need to, at this time. But slapping in some off-the-shelf image recognition software, etc is down the road.

Rright now you have independant groups creating all sorts of the Pieces to the AI pie. Vision systems, pattern recognition, "real language" engines...getting then in to more sophisticated stuff of "values" determination....someday it will start to get pieced together, and I don't think it will be too long from now.

It all provokes the fascinating question:

"What is creativity?"

Well, yeah, of course we'll get there. My point is that it isn't as easy as just plugging in a bunch of different machines together.

No -- we need some centralized beurocracy to determine and define requried interfaces and set the communication standards for our Future Robot Overlords.

IEEE and such have done a find job of standardizing data for things such as vehicle emissions, etc. Everyone knows USB, Wi-Fi, etc...but until we standardize AI research, the disparate groups of researchers can't create the "plug and play" emotional-control box I really need for my guard dog robot.

Who says intelligent machines need brains made of metal? Rat neurons grown in a petri dish with electrodes hooked up to a flight simulator were able to learn to keep the plane steady, and that's just a tiny amount of cells compared to a brain. We also have artificial limbs that can be controlled by a persons thoughts. Simply scale it up with an artificial environment for the brain housed in a machine, hooking up some other computer parts for faster mathematical calculations may be possible.

Unethical? Do it in China!

Hyper - I would argue that what you described is not intelligence, despite using Nuerons from rats, any more than a learning computer made of silicon. neural nets are nothing new.