Slashdot Mirror
Elon Musk: The Danger of AI is Much Greater Than Nuclear Warheads. We Need Regulatory Oversight Of AI Development. (youtube.com)
Elon Musk has been vocal about the need for regulation for AI in the past. At SXSW on Sunday, Musk, 46, elaborated his thoughts. We're very close to seeing cutting edge technologies in AI, Musk said. "It scares the hell out of me," the Tesla and SpaceX showrunner said. He cited the example of AlphaGo and AlphaZero, and the rate of advancements they have shown to illustrate his point. He said: Alpha Zero can read the rules of any game and beat the human. For any game. Nobody expected that rate of improvement. If you ask those same experts who think AI is not progressing at the rate that I'm saying, I think you will find their betting average for things like Go and other AI advancements, is very weak. It's not good.
We will also see this with self driving. Probably by next year, self driving will encompass all forms of driving. By the end of next year, it will be at least 100 percent safer than humans. [...] The rate of improvements is really dramatic and we have to figure out some way to ensure that the advent of digital super intelligence is symbiotic with humanity. I think that's the single biggest existential crisis we face, and the most pressing one. I'm not generally an advocate of regulation -- I'm actually usually on the side of minimizing those things. But this is a case, where you have a very serious danger to the public. There needs to be a public body that has insight and oversight to ensure that everyone is developing AI safely. This is extremely important. The danger of AI is much greater than danger of nuclear warheads. By a lot.
We will also see this with self driving. Probably by next year, self driving will encompass all forms of driving. By the end of next year, it will be at least 100 percent safer than humans. [...] The rate of improvements is really dramatic and we have to figure out some way to ensure that the advent of digital super intelligence is symbiotic with humanity. I think that's the single biggest existential crisis we face, and the most pressing one. I'm not generally an advocate of regulation -- I'm actually usually on the side of minimizing those things. But this is a case, where you have a very serious danger to the public. There needs to be a public body that has insight and oversight to ensure that everyone is developing AI safely. This is extremely important. The danger of AI is much greater than danger of nuclear warheads. By a lot.
Well Musk, It's a really good thing we have been working hard on consumer protections, ensuring privacy, and sensibly regulating banks, company mergers, and are finally enjoying a fiscally responsible government. This should be a cake walk! (As in let them eat cake)
... we can't even stop spam.
It little behooves the best of us to comment on the rest of us.
e.g. what are the specific risks? Because all I ever hear is backhanded fearmongering. This isn't to say I don't think AI is a danger. Kill bots don't scare me because I think they'll go rogue ala Terminator, they scare me because needing to treat the army well is just about the only thing that keeps the 1% in line. But I don't hear anyone talking about that. Or about what automation is going to mean.
Basically, we need to be getting ready for a future where the rich don't need us to buy their crap and make them rich. Instead we're worrying about 80s science fiction scenarios.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
AI is all around us already, mostly in rudimentary forms. Right now it's fairly innocuous, like NetFlix's AI suggesting what movies/shows you might enjoy watching next. Same on YouTube, though YouTube's AI goes further, it decides who's videos get de-monetized as well as suggesting videos to users.
Just scratching the surface a bit there. Someone said it's all poppycock, but I'm telling you, if you start putting the puzzle pieces together, we have this NOW. Machine learning is, as expected, maturing at an ALARMING rate of speed. Our technology advancement is accelerating. People seem to neglect that ideal, we've come such a long way in the past 150 years, the common perception is we have this under control. Do we?
Already we're putting the pieces in place for some scary potential outcomes: Fitting cars and drones with AI to navigate our world. In the latter case, militarized drones. Forget about putting weapons on these things. The things themselves can be weapons. Keep your eyes on the technology, machine learning is only going to get better and faster, and do it at an accelerating rate. We already know our machine learning techniques can be trained to do all sorts of interesting tasks. From Alpha Go, we learned that an AI can train itself at a breakneck speed. It is pretty scary stuff, put these pieces together in the right away, who knows what it could figure out. I won't go as far as self-awareness, but it is certainly a possibility, with this rate of advancement, who knows what's in the pipeline.
AI can become a problem. Right now not so much by the infamous Skynet scenario and more in conjunction with big data and automation. It will kill jobs which are repetitive. Together with big data it will be used to manipulate the masses in ways past dictatorships where unable to. It can ruin our democracies. The current abilities of mass media including talk radio and the internet have already been augmented. The same applies to advertisements and customer communication. Yes we need regulations. And a lot of them. And it should include that software should not be allowed to be designed to stimulate dopamine responses.
And you seem to be under the impression that AI needs to be what you think constitutes intelligence before it becomes a risk.
#DeleteFacebook
How have humans treated other species and even subgroups of humans.
Why do you think an A.I. would treat humans any differently?
Like the A.I. that didn't play the rules of Q*Bert but instead cheated in a new way humans had missed.
A.I. does what it's incented and trained to do. But we don't always know what we are incenting it or training it to do.
An A.I. to identify sheep or tanks could easily be accidentally be trained to recognize sunny or cloudy days or white fence posts by accident.
An A.I. tasked with improving the standard of living for humans could logically conclude that a suffering human would be better off dead and a group of humans would have a higher standard of living if there were less humans in the group.
A.I. research should be done in an air gapped environment, with analog power meters, easily disruptable power supply, physically fused remotely, remotely video and audio recording of the people directly engaging with the A.I.
This is an extinction level risk. It should be taken seriously.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
I don't think anyone is beating Microsoft in AI. They are just not exposing their tech on client side. They are keeping it on server side and only letting customers use it on per-client request basis.
You mean just like Google does? That's not how AI works in practice anyway.
The algorithms are mostly the algorithms. Often tweeking them reduces their performance (hell tweeking their hyperparameters often reduces their performance too). Applied AI (solving a specific problem with AI) is usually about feature extraction and then just trying different algorithms until you find one that works well enough or better than the rest. Some algorithms (NN) have a network topology and often people spend large amounts of time/computation trying different graphs. Often this process is about access to large amounts of computing hardware, access (and use of) high performance analysis systems for streaming data to the algorithms and efficient implementations of those algorithms (Random Forest, Gradient Boosted Machines, etc). Most of these things have nothing to do with AI research (inventing new algorithms).
The point is that none of the examples Musk mentioned involve new algorithms (truly effective novel algorithms are quite rare). And the companies that Musk mentions are downright terrible at all the things around an AI system that are mostly about efficient I/O and computation (BigQuery is the worst performing analytics engine ever created). So any fear of those technologies is probably poorly founded. Either Musk doesn't understand this area (likely) or he has some other reason to propose this (also likely). I'm betting on the latter.
"Those that start by burning books, will end by burning men."
It isn't hard to imagine run away AI. Scifi is full of it. But I find it hard to imagine that humans will create an institution to prevent that on a worldwide scale before it's too late. Elon is clearly an optimist.
It's only after Hiroshima that nuclear proliferation became an issue. Only after the Netherlands was massively flooded that they started their Deltaworks. Only after big scandals where things go utterly wrong that we start with regulation and enforcement.
So I would be surprised if we - as a species - get this right and survive this one. We're probably too dumb, as most of the posts in this thread illustrate.
At least Musk tries... Jasper
What Elon is looking for here is export controls, and I would agree that we need to classify artificial intelligence as having the capability of becoming a threat to national security. Which means if you are developing AI in the United States and want to send the code to another country, you need the State Department to authorize that you're not putting bad technology in the hands of worse people.
Any fool can criticise, condemn, and complain, and most fools do. - Benjamin Franklin