Latest / INSIGHT TODAY / Trump says AI companies must 'self-regulate'

Transcript

Trump says AI companies must 'self-regulate' · INSIGHT TODAY

• 0:00
  Speaker 1: We begin here in Washington, where President Trump says AI
• 0:03
  Speaker 1: companies must self-regulate after hosting a meeting with top tech
• 0:08
  Speaker 1: bosses at the White House. It comes amid growing calls
• 0:11
  Speaker 1: to slow down AI development, and a day after, OpenAI
• 0:15
  Speaker 1: scrapped its latest model release due to safety concerns. The
• 0:20
  Speaker 1: tech leaders at the White House included Anthropic CEO Dario Amode,
• 0:25
  Speaker 1: Meta's Mark Zuckerberg, and Elon Musk. After a roundtable meeting,
• 0:29
  Speaker 1: the president brought the AI leaders outside the White House
• 0:32
  Speaker 1: to the media area for a rare news conference. Here's
• 0:35
  Speaker 1: some of what he had to say about AI companies
• 0:38
  Speaker 1: regulating themselves in response to a reporter's question.
• 0:43
  Speaker 2: Mr.
• 0:43
  Speaker 3: President, did anything that you heard today, did anything change
• 0:46
  Speaker 3: your mind, given the fact that you've said you don't
• 0:48
  Speaker 3: want to see guardrails, you don't want to see AI
• 0:50
  Speaker 3: slowed down? Anything that you heard today change your mind
• 0:52
  Speaker 3: on that? Well, I think I'm seeing tremendous self-policing, and
• 0:56
  Speaker 3: they understand that they have to self-police. And it's very
• 0:58
  Speaker 3: hard for somebody to come outside, get into these models.
• 1:01
  Speaker 3: These models are very complex. And again, I believe they're
• 1:04
  Speaker 3: going to be used for the good, and when they're
• 1:06
  Speaker 3: not We're going to be able to nab them. But
• 1:08
  Speaker 3: there's going to be a tremendous self-policing aspect. We're also
• 1:12
  Speaker 3: thinking about forming a committee of sorts where we put
• 1:16
  Speaker 3: maybe 10 people on that committee. It could be from
• 1:18
  Speaker 3: that group so that we can so the committee can
• 1:21
  Speaker 3: watch over the whole enterprise. We're looking at this for good.
• 1:25
  Speaker 3: It's also maybe the biggest industry. They say bigger than
• 1:29
  Speaker 3: the Industrial Revolution.
• 1:30
  Speaker 1: Think of it.
• 1:31
  Speaker 3: You know what? We're leading by a lot and we're
• 1:34
  Speaker 3: going to keep it that way.
• 1:36
  Speaker 1: Elon Musk, whose company developed the Grok chatbot, gave a
• 1:40
  Speaker 1: similarly optimistic view of the future of AI.
• 1:43
  Speaker 2: I think by far the most likely outcome is an
• 1:45
  Speaker 2: age of abundance, where we don't have just universal basic income,
• 1:49
  Speaker 2: we have universal high income. And people speak about medical benefits.
• 1:56
  Speaker 2: But maybe it's worth talking a bit more about that.
• 2:00
  Speaker 2: When you have robots and you have super intelligence, it
• 2:04
  Speaker 2: means that everyone in the world can have better medical
• 2:07
  Speaker 2: care than anyone here, including me. And wouldn't that be wonderful?
• 2:12
  Speaker 4: Not if they haven't got a job.
• 2:15
  Speaker 2: Well, jobs are going to change. Jobs have always changed.
• 2:19
  Speaker 2: But being a computer used to be a job. There
• 2:23
  Speaker 2: used to be buildings full of people doing calculations. But
• 2:26
  Speaker 2: now digital computers do those tasks. I don't think anyone
• 2:31
  Speaker 2: is pining to go back and do calculations all day
• 2:34
  Speaker 2: to do bank interest rates, which is what they used
• 2:36
  Speaker 2: to do. So I think we'll see an evolution of roles.
• 2:41
  Speaker 1: Jairio Amode is the CEO of Anthropic. That's the company
• 2:44
  Speaker 1: that created the Claude AI model. He's among those calling
• 2:48
  Speaker 1: for AI development to slow down. Earlier this month, he
• 2:51
  Speaker 1: wrote an online essay spelling out serious risks of the technology.
• 2:56
  Speaker 1: A reporter asked Mr. Amode for his views following the
• 2:59
  Speaker 1: White House meeting.
• 3:01
  Speaker 5: Look, I mean, I will say what I've always said,
• 3:04
  Speaker 5: which is that AI has incredible benefits.
• 3:08
  Speaker 2: I've talked about the medical benefits.
• 3:09
  Speaker 5: Of the technology. As the president has said, you know,
• 3:12
  Speaker 5: whoever wins AI wins. I think that's very important. But
• 3:16
  Speaker 5: I think the technology has very real risks. And, you know,
• 3:20
  Speaker 5: the mechanism, how we address those risks is still under discussion.
• 3:24
  Speaker 5: But we all need to work. We all need to
• 3:26
  Speaker 5: work together to make sure that we can win and
• 3:29
  Speaker 5: we can win safely if we if we do this right.
• 3:31
  Speaker 5: We work with the president and everyone here.
• 3:33
  Speaker 2: We can win.
• 3:33
  Speaker 1: Well, joining me live now is Andrew Ng. He's the
• 3:38
  Speaker 1: founder of Deep Learning AI. He was also co-founder of
• 3:42
  Speaker 1: Google Brain. Thanks for being with us. So much going
• 3:45
  Speaker 1: on today in terms of AI. I guess we could
• 3:48
  Speaker 1: sum it up by saying breaks on or breaks off.
• 3:51
  Speaker 1: What's your take?
• 3:53
  Speaker 4: I think AI risks and AI safety has been sensationalized.
• 3:57
  Speaker 4: There are risks to AI. There are some problems with it.
• 4:00
  Speaker 4: But these are enduring problems to be solved and that
• 4:03
  Speaker 4: we will solve rather than signs that the sky is falling.
• 4:08
  Speaker 1: You essentially have AI agents who are tasked with a
• 4:10
  Speaker 1: goal that they're very committed to achieving and what it
• 4:15
  Speaker 1: does to get that goal lacks a moral compass. How
• 4:17
  Speaker 1: is that not concerning?
• 4:20
  Speaker 4: I think it is concerning that open AI software agents
• 4:23
  Speaker 4: hacked into a hugging phase. That is bad. The silver
• 4:26
  Speaker 4: lining is that now that it's happened once, OpenAI and
• 4:29
  Speaker 4: the rest of the AI world are aware of this
• 4:31
  Speaker 4: problem and we're doing a lot of work to put
• 4:33
  Speaker 4: a stop to it. In fact, to make an analogy,
• 4:36
  Speaker 4: lots of things humanity has built used to be really dangerous.
• 4:39
  Speaker 4: Airplanes used to be really dangerous. So we flew them,
• 4:42
  Speaker 4: figured out what was wrong, solved the problem, made them safe.
• 4:45
  Speaker 4: And I'm actually confident with all the ways we have
• 4:47
  Speaker 4: to shape AI and control AI. There's just many years
• 4:50
  Speaker 4: of engineering work to be done, but we can corral
• 4:52
  Speaker 4: these things and it will end up benefiting everyone.
• 4:55
  Speaker 1: But when it comes to objects, if you like, humans
• 4:58
  Speaker 1: broadly are in control of those. Are we in control
• 5:00
  Speaker 1: of AI?
• 5:03
  Speaker 4: Our control of AI is... So first, we don't perfectly
• 5:05
  Speaker 4: control airplanes. Even now, airplanes get buffeted around. But I
• 5:09
  Speaker 4: think we have such wonderful tools to control AI. They're
• 5:12
  Speaker 4: not perfect. For example, a few years ago, AI used
• 5:14
  Speaker 4: to be famously sexist, racist, toxic because of things that
• 5:18
  Speaker 4: learn on the internet. But we learned that by zeroing
• 5:20
  Speaker 4: out certain numbers... We could make AI much less racist,
• 5:23
  Speaker 4: much less sexist, and so on. Then, you know, we
• 5:27
  Speaker 4: could make AI less sexist, much better than we had
• 5:30
  Speaker 4: tools to try to convince humans to be less sexist
• 5:32
  Speaker 4: or racist. Because to go in and change AI directly,
• 5:36
  Speaker 4: the ways to do so are complex. I don't want
• 5:38
  Speaker 4: to minimize the research and entry work ahead. But today,
• 5:41
  Speaker 4: I have much better ideas for how to shape AI's
• 5:43
  Speaker 4: behavior than I have ideas for how to change a
• 5:46
  Speaker 4: human's behavior. And the problems we've made is good, and
• 5:48
  Speaker 4: we'll keep on working at it. The bigger danger is
• 5:51
  Speaker 4: that people sensationalize these risks, leading to, say, lobbyists getting
• 5:56
  Speaker 4: governments to pass stifling regulations. They're anti-competitive or to slow
• 6:00
  Speaker 4: down AI development among democracies, leading adversaries to take the lead.
• 6:05
  Speaker 4: I think that's actually a bigger danger than AI killing
• 6:09
  Speaker 4: us all, which is not going to happen.
• 6:11
  Speaker 1: I was going to ask you, the very dire warnings
• 6:13
  Speaker 1: that are scaring lots of us about the end of
• 6:16
  Speaker 1: humanity that could be brought about by AI. Why don't you,
• 6:19
  Speaker 1: you've outlined it there, but tell me more about why
• 6:22
  Speaker 1: you don't subscribe to that idea.
• 6:25
  Speaker 4: I don't see any plausible path that AI could lead
• 6:27
  Speaker 4: to the end of humanity. All the scenarios I see
• 6:29
  Speaker 4: feel much more or less like science, but more like
• 6:32
  Speaker 4: science fiction. And in fact, if you look at the
• 6:34
  Speaker 4: things that are dire for the human race, things like
• 6:37
  Speaker 4: climate change or maybe another asteroid hitting us or maybe
• 6:40
  Speaker 4: another pandemic, I think AI is much more likely to
• 6:43
  Speaker 4: be our savior or to help us deal with true
• 6:46
  Speaker 4: crises and true risks to humanity than to be the
• 6:49
  Speaker 4: cause of these bad things happening. We just got to
• 6:53
  Speaker 4: work on the short-term and long-term engineering problems and avoid
• 6:57
  Speaker 4: the fear-mongering, which some companies have kind of indulged in
• 7:02
  Speaker 4: to try to shell out competitors or to get PR
• 7:04
  Speaker 4: attention and for other purposes like that.
• 7:06
  Speaker 1: So you don't support a slowdown. But when it comes
• 7:10
  Speaker 1: to regulation, where do you stand on that? We heard
• 7:12
  Speaker 1: President Trump today saying that it would essentially be a
• 7:16
  Speaker 1: moral decision on regulation rather than anything that's legally bound.
• 7:20
  Speaker 1: What do you make of that?
• 7:22
  Speaker 4: You know, I think there is room for healthy regulation.
• 7:25
  Speaker 4: For example, I think non-consensual deep fake news or non-consensual
• 7:30
  Speaker 4: intimate imagery is one of the most disgusting things I've
• 7:32
  Speaker 4: ever seen or heard of. And we should just get
• 7:35
  Speaker 4: rid of that. Let's outlaw that, put huge penalties on that.
• 7:37
  Speaker 4: So good regulations would help improve AI and make it safer.
• 7:41
  Speaker 4: The problem is the number of lobbying dollars that's gone
• 7:45
  Speaker 4: into AI lobbying is so intense. There have been far
• 7:48
  Speaker 4: more bad regulatory proposals than good ones. So if we
• 7:50
  Speaker 4: can get only the good regulations passed, that would be fantastic.
• 7:54
  Speaker 4: But the danger is actually more a lot of bad regulations.
• 7:57
  Speaker 4: For example, I think the EU AI Act, unfortunately, probably
• 8:01
  Speaker 4: slowed down AI more than ever. Then it made people safer. Fortunately,
• 8:04
  Speaker 4: it doesn't roll back. We should keep on working on
• 8:07
  Speaker 4: good regulations to get rid of the harmful use cases.
• 8:10
  Speaker 4: But that has to be balanced with not letting the
• 8:13
  Speaker 4: lobbyists have their way and shut up AI competition.
• 8:16
  Speaker 1: Yeah, interesting. Thank you, Andrew Ng, founder of deeplearning.ai. He's
• 8:21
  Speaker 1: also co-founder of Google Brain. Really interesting conversation. Thank you.