Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

money.cnn.com

Post reply on HN
Print view View original

91–98 of 98 posts

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#91

Post by shpx »

gonehome wrote: post #81
circlefavshape wrote: post #78

Earlier quoted context omitted.

Oh! Good insight! ... but we already have that, right? Undesirable outcomes from rulesets that govern the behaviour of a large number of (conscious or unconscious) actors are a pretty standard feature of civilisation

It's the paper clip maximizer idea [1]. "The goal of maximizing paperclips is chosen for illustrative purposes because it is very unlikely to be implemented, and has little apparent danger or emotional load (in contrast to, for example, curing cancer or winning wars). This produces a thought experiment which shows the contingency of human values: An extremely powerful optimizer (a highly intelligent agent) could seek…

OpenAI did a thing[0] where they used a human to help train a machine learning model by picking the best of 2 models.

> Our algorithm’s performance is only as good as the human evaluator’s intuition about what behaviors look correct, so if the human doesn’t have a good grasp of the task they may not offer as much helpful feedback. Relatedly, in some domains our system can result in agents adopting policies that trick the evaluators. For example, a robot which was supposed to grasp items instead positioned its manipulator in between the camera and the object so that it only appeared to be grasping it, as shown below.

[0] https://blog.openai.com/deep-reinforcement-learning-from-hum...

https://news.ycombinator.com/item?id=14545298 78 upvotes, 7 comments

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#92

Post by sebular »

phasnox wrote: post #84

I love Musk. He is brilliant and I think he has truly shaped the future. But he is wrong. Artificial Conscience(what he is referring to) is NEVER going to be achieved. I repeat NEVER. Its impossible. Conscience is in the form and not in the matter. In one word, there is never going to be an AI with free will. However, eventually we may build a super intelligent system with great power over our lives, that goes wrong,…

Let's put aside your absolute and close-minded certainty about the future for a minute. Unless you can provide some quotes, you're putting a lot of words in Elon Musk's mouth. He's made some goofy pop culture references during interviews, and it seems to be a calculated decision on his part to flirt with sensationalism in order to publicize the issue, but he's never seriously argued that the movie Terminator is a documentary about the future.

Call it conscience, intelligence, whatever you want. The danger isn't that some ominously calculating and murderous robotic mind is going to spring into being and hide from humans while plotting our downfall. In order to understand the danger, all you have to do is look at what the US military is already doing with autonomous weapons. Killer robots aren't hypothetical, they're historical.

In fact, you're the prime example of the danger of AI. You probably have a stronger than average understanding of computers, and maybe you know a lot about actual AI implementations, which is what makes you so self-assured that there's nothing magical about them. So you trust them, and you're a big fan of throwing an ever-increasing amount of trust into AI.

You even admit that there will be bugs (there always are) and misbehavior (now who's humanizing programs?), but you fail to see why that's a problem when the stakes are raised from "crap, an app leaked private data online" to "crap, the autonomous weaponized drone mistook backyard fireworks for an attack and bombed a family."

The way I see it, Musk is far from alarmist, and all the kidding about "summoning demons" is almost a way of coping with what's starting to seem like a terrifying inevitability. We've been building "dumb programs" for decades, and there's still a constant stream of breaking news about software that didn't do what it was supposed to. And you want to believe that there's no danger in building software that has increasingly fuzzy logic and connecting it to real-world I/O?

I'm guessing that Elon Musk saw this problem when he first started toying with the idea of self-driving cars. You say he has a limited understanding of AI, but Tesla autopilot says otherwise. And he probably became keenly aware of the stakes when he realized that people would willingly entrust their lives to the decisions made by his company's software, and that without regulation, it's entirely up to Tesla's QA process to make sure their cars don't accidentally kill people.

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#93

Post by sebular »

gonehome wrote: post #26

They're talking past each other because "AI" has become a useless term. Musk's comments are about a potential (maybe conscious) artificial super intelligence and the risks that could come from that if we enable one somehow without understanding the preconditions. Zuckerberg is just talking about "AI" in the sense that machine learning can solve a valuable class of problems. The super intelligence risk could be seriou…

I think that they're actually talking right at each other, but Zuckerberg is thinking of the possibilities, while Musk is thinking of the implementations.

"AI is the future" really means that Zuckerberg sees the massive potential of automation, and how it's going to change the face of the world.

"We should be worried about AI" really means that Musk has had to think long and hard about the fact that the software written by Tesla is navigating living human beings around obstacles at high speeds.

This isn't about some accidental super intelligence that becomes self-aware and murders all humans, it's about the consequences of applying today's standards of "let's ship this code" to technology that we're expected to entrust with our lives.

This is two things, combined: "AI" means that code will have more complex emergent behavior that's harder to debug than current code, which is already nearly impossible to completely debug. "AI" also means that this code will run very complex systems that often have powerful hardware components, unfortunately including weapons.

To put it another way: Both men agree that the code of the near future is going to make a massive leap in effectiveness and real-world applications. But Musk has to grapple with being held morally accountable for that code, while the only thing Zuckerberg's ever had to care about is performance. Say "AI" to Mark Zuckerberg and he sees all of the money in smarter data mining and fun home appliances. Say "AI" to Elon Musk and he has nightmares of deploying a wireless update that causes Teslas to confuse people wearing green clothing with green traffic lights.

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#94

Post by rspeer »

hndamien wrote: post #76
rspeer wrote: post #68

Earlier quoted context omitted.

Musk's view of AI relies on AI being magic, too. He's using the definition of "AI" that exists only in hyped-up promises about the future, and evaporates when applied to present technology. Musk dismisses AI experts who tell him that his claims bear no resemblance to real technology, because Musk isn't talking about the same AI as them. He's talking about sci-fi AI. There are no experts in sci-fi AI because, if there…

Don't underestimate the exponential.

And don't overestimate the S-curve, which looks the same when you're in the first half of it.

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#95

Post by MarkMMullin »

Facebook is capable of monetizing current ML capabilities to do things like constantly tune individual news feeds so that its profit is maximized. Certainly counterproductive to some degree for modern society, but it is profitable. Straightforward goal. straightforward outcome.

Elon's claimed that Tesla will have level 5 autonomous vehicles in 2 years. Yeah, in your dreams Elon, that's a massive cognitive stack you can't fit in the vehicle, one we've never built out to such a degree, and your car's autonomy level is going to be a function of its bandwidth to the cloud anyway. The mean part of me wonders if Elon plans on covering this wild assertion with 'Well, we can't now. Regulations and stuff' - if he's just worried about putting a weapon on a solar powered drone that patrols some area and shoots baddies, yeah me too. But that's just human stupidity, it ain't AI, it ain't intelligent, it's just a dumb program with effectors. All ML really is these days is real big boundary relaxation systems, calculating the massive number of coefficients we need to make the equations work out the way we want. But come on, it's stupider than a rock. It's an equation solver and its brittle as hell. As a tool I love what we can now accomplish, but there's a long road from here to Hal. It's not bloody magic, it's just math.

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#96

Post by mercer »

convexfunction wrote: post #80
circlefavshape wrote: post #78

Earlier quoted context omitted.

Oh! Good insight! ... but we already have that, right? Undesirable outcomes from rulesets that govern the behaviour of a large number of (conscious or unconscious) actors are a pretty standard feature of civilisation

Yes, we do have that already, and it's kind of odd that people who see no point in thinking hard about risks to humans from this particular type of inhuman optimization process would readily agree that other types of inhuman optimization processes (with much better understood risk profiles and limitations) can be very harmful. :)

Perhaps the 'singularity' type stuff makes a lot of people disengage?

So far I'm inclined to think that the whole 'sentient AI mucking things up' is jumping the gun (to put it mildly). I might very well be wrong though.

However, it's only because of HN that I even bother to make the distinction between that particular fear (realistic or otherwise) and the much more directly useful discussion of how AI-type stuff (even just facebook feed algorithms) can be dangerous or harmful.

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#97

Post by circlefavshape »

convexfunction wrote: post #80
circlefavshape wrote: post #78

Earlier quoted context omitted.

Oh! Good insight! ... but we already have that, right? Undesirable outcomes from rulesets that govern the behaviour of a large number of (conscious or unconscious) actors are a pretty standard feature of civilisation

Yes, we do have that already, and it's kind of odd that people who see no point in thinking hard about risks to humans from this particular type of inhuman optimization process would readily agree that other types of inhuman optimization processes (with much better understood risk profiles and limitations) can be very harmful. :)

I don't think it's odd. I agree that optimising for the wrong thing (or, probably, for any particular thing) can be harmful, especially when it's done on a very large scale. AI-as-we-know-it doesn't introduce a new class of risk, it just makes some of the optimisations faster. I don't see any point in thinking about AGI risk until we get some experimental evidence that AGI is possible

Re: Elon Musk says Mark Zuckerberg's understanding of AI is 'limited'

#98

Post by convexfunction »

circlefavshape wrote: post #97
convexfunction wrote: post #80

Earlier quoted context omitted.

Yes, we do have that already, and it's kind of odd that people who see no point in thinking hard about risks to humans from this particular type of inhuman optimization process would readily agree that other types of inhuman optimization processes (with much better understood risk profiles and limitations) can be very harmful. :)

I don't think it's odd. I agree that optimising for the wrong thing (or, probably, for any particular thing) can be harmful, especially when it's done on a very large scale. AI-as-we-know-it doesn't introduce a new class of risk, it just makes some of the optimisations faster. I don't see any point in thinking about AGI risk until we get some experimental evidence that AGI is possible

Maybe. I think the idea is that it's preferable to have wasted a lot of effort on coming up with irrelevant abstractions addressing a nonexistent problem, than to be surprised by something weird and poorly-understood presenting novel and not-clearly-bounded risk and not have effective tools at hand to address it.

I won't claim to know what the right amount of effort directed toward this is though, nor do I really think AI safety research has a >epsilon chance of accomplishing its goal if it turns out to be relevant at all. ¯\_(ツ)_/¯

Post reply on HN