Latest / THE DAILY BRIEF / Anthropic bans abusive language towards A.I bots as security concerns swirl

Transcript

Anthropic bans abusive language towards A.I bots as security concerns swirl · THE DAILY BRIEF

• 0:00
  Speaker 1: So does AI need some sort of kill switch? Both
• 0:03
  Speaker 1: tech leaders and lawmakers are pushing for more guardrails amid
• 0:07
  Speaker 1: AI's rapid advancement. The tech industry is advocating for more
• 0:11
  Speaker 1: AI safety measures as the systems become more human-like. Key
• 0:15
  Speaker 1: players like Microsoft are suggesting an emergency brake feature on
• 0:20
  Speaker 1: AI applications to protect its users against corrupted models. This
• 0:25
  Speaker 1: comes on the heels of an anthropic AI model submitting
• 0:27
  Speaker 1: a false tip and an unsolved murder case in Philadelphia.
• 0:32
  Speaker 1: According to police, the model also attempted to fill out
• 0:34
  Speaker 1: official State Department legal forms, which is raising enormous public concern.
• 0:39
  Speaker 1: Meemothropic is also protecting AI chatbots themselves. The company behind
• 0:43
  Speaker 1: Claude released a statement saying, quote, the policy update is
• 0:47
  Speaker 1: meant to apply only in extreme cases where users repeatedly
• 0:51
  Speaker 1: act cruelly toward our models. with no discernible purpose, this
• 0:56
  Speaker 1: prompting a larger conversation about AI having rights, with some
• 1:01
  Speaker 1: saying this can blur the lines of what it means
• 1:03
  Speaker 1: to be human. The Microsoft AI CEO says, unfortunately, a
• 1:07
  Speaker 1: growing chorus of people argue that AIs could now. or
• 1:11
  Speaker 1: may soon become conscious, and that, like other conscious beings,
• 1:17
  Speaker 1: they may deserve rights and protections.
• 1:20
  Speaker 2: Wow.
• 1:21
  Speaker 1: If this view takes hold, it will change what it
• 1:23
  Speaker 1: means to be human and shake the foundations of human society.
• 1:27
  Speaker 1: Tech entrepreneur David Sachs' AI poses a very real threat
• 1:30
  Speaker 1: to humanity without the right restrictions in place.
• 1:35
  Speaker 3: This is right out of Frankenstein.
• 1:37
  Speaker 4: This actually sounds dangerous to me. It's not just a
• 1:39
  Speaker 4: new religion.
• 1:40
  Speaker 5: It's that the religious cult is programming the super intelligence.
• 1:45
  Speaker 1: Kennedy, whenever I hear, first of all, this talk of
• 1:47
  Speaker 1: a kill switch, I'm like, one of the tech CEOs
• 1:50
  Speaker 1: joked like, yeah, that's like turning the computer off. So
• 1:54
  Speaker 1: what really is the plan here? And I don't know,
• 1:57
  Speaker 1: should we follow the advice of the AI tech CEOs
• 2:02
  Speaker 1: or do we really need Congress to get involved here?
• 2:05
  Speaker 6: Congress will— like, I want Bernie Sanders to explain to me
• 2:08
  Speaker 6: what a kill switch is. Go ahead and point to
• 2:11
  Speaker 6: it on the doll, Bernie. Where is the kill switch?
• 2:14
  Speaker 1: Or AI, for that matter.
• 2:15
  Speaker 2: Yeah, he can't—yeah.
• 2:16
  Speaker 3: It's not superintelligence.
• 2:18
  Speaker 5: That is a Trump term.
• 2:19
  Speaker 4: That is an executive order that is probably illegal.
• 2:22
  Speaker 1: He's an oligarch.
• 2:24
  Speaker 6: And therefore, the robots are going to come for him first. Yeah,
• 2:28
  Speaker 6: you know, it's like all the gummers in Congress are
• 2:31
  Speaker 6: the ones who are going to make the rules now.
• 2:33
  Speaker 6: I don't necessarily trust that. So what are they trying
• 2:35
  Speaker 6: to do? They're trying to make as much money as possible.
• 2:37
  Speaker 6: You don't see Anthropic slowing down on their IPO, which is,
• 2:41
  Speaker 6: you know, supposedly happening next month. And OpenAI, they've also
• 2:46
  Speaker 6: filed paperwork with the SEC. Like, all these companies, I
• 2:49
  Speaker 6: don't care what they say, they just want to make
• 2:51
  Speaker 6: as much money as possible. And this fear-mongering is a
• 2:55
  Speaker 6: tactic for them to use the government to have control
• 2:59
  Speaker 6: so their competitors don't have an advantage over them when
• 3:03
  Speaker 6: their technology succeeds what they're offering right now.
• 3:07
  Speaker 1: But this idea, Griff, that these AI models could soon
• 3:10
  Speaker 1: have a conscious, like that they become real people. And
• 3:14
  Speaker 1: I don't know, somehow then they deserve rights to live
• 3:17
  Speaker 1: amongst us.
• 3:17
  Speaker 4: They have rights and they probably will unionize soon and
• 3:21
  Speaker 4: take over. And I can't top Kennedy. That was so great.
• 3:24
  Speaker 4: But it's such an important point. like literally not just
• 3:26
  Speaker 4: Bernie Sanders, but our lawmakers, like they are frequently the
• 3:30
  Speaker 4: least knowledgeable about the laws they enact for said regulation.
• 3:34
  Speaker 4: But I will say this, the whole part where you're
• 3:37
  Speaker 4: reading about people being mean to AI and Claude, like
• 3:41
  Speaker 4: don't piss it off. Like the CEOs tell us that
• 3:45
  Speaker 4: it's going to get out of the sandbox and wipe
• 3:46
  Speaker 4: out humanity. Let's not be mean to it. Like don't
• 3:49
  Speaker 4: bully your AI. Muse is the new thing or whatever
• 3:53
  Speaker 4: it is. I think, you know, It's an interesting point,
• 3:57
  Speaker 4: and you played David Sachs there. David Sachs, when the
• 4:00
  Speaker 4: CEO of Tech Titans came out and said, we need laws,
• 4:03
  Speaker 4: we need immediate government regulation, Sachs said, look, you created it.
• 4:08
  Speaker 4: You put guardrails on it. You solved the problem. And
• 4:11
  Speaker 4: it does, as we have this conversation. nationally, it feels
• 4:15
  Speaker 4: very uncomfortable, Sandra, that we don't fully know where we're headed.
• 4:19
  Speaker 1: But Lydia, from a financial perspective, does that make sense?
• 4:22
  Speaker 1: Would they really put the guardrails on when it behooves them,
• 4:25
  Speaker 1: as far as profits, to have these things go crazy
• 4:29
  Speaker 1: and develop out of control?
• 4:30
  Speaker 5: You know, I'm very skeptical when they're up there talking
• 4:33
  Speaker 5: about the guardrails that are needed, right? Because In some ways,
• 4:36
  Speaker 5: they should have known where they were going, what they
• 4:37
  Speaker 5: were developing all along.
• 4:38
  Speaker 1: So why now?
• 4:39
  Speaker 5: I think to Kennedy's point, we should be skeptical of
• 4:41
  Speaker 5: what they're trying to accomplish. They're trying to prevent other
• 4:44
  Speaker 5: smaller up-and-coming technology from taking the mantle away from them.
• 4:47
  Speaker 5: They want to go public and capture all the financial
• 4:49
  Speaker 5: capital before anybody else can move in on it. I
• 4:51
  Speaker 5: will say, though, that we should pay attention to the
• 4:55
  Speaker 5: market and how the market reacts to all these various announcements.
• 4:58
  Speaker 5: When you look broadly across the market, like the S
• 5:00
  Speaker 5: & P 500, right now, these are really driving earnings
• 5:04
  Speaker 5: and driving returns in the market. So you have to ask,
• 5:07
  Speaker 5: would the guardrails help them or hurt them? Has that
• 5:10
  Speaker 5: fully been priced into what we're seeing in these tech
• 5:12
  Speaker 5: stocks yet or not? Would guardrails actually... hurt earnings, hurt revenue,
• 5:15
  Speaker 5: hurt outlook, or would actually guardrails help them? Would that
• 5:20
  Speaker 5: help drive larger adoption moving forward? I don't know that
• 5:23
  Speaker 5: anybody knows the answers to those questions, but one place
• 5:25
  Speaker 5: to look for would be the market. Right now, the
• 5:27
  Speaker 5: tech stocks are still leading. Donna, do you fear that.
• 5:29
  Speaker 1: AI could get so out of control that we need
• 5:31
  Speaker 1: this sort of like emergency break on development? Well, I
• 5:34
  Speaker 1: think we don't know.
• 5:35
  Speaker 2: And that's the problem. And if the robots and the
• 5:38
  Speaker 2: technology is smarter and become smarter than the people implementing it,
• 5:42
  Speaker 2: that's when it gets scary. And how do we know
• 5:44
  Speaker 2: that answer? And where does the responsibility fall? And that's
• 5:48
  Speaker 2: my question. If we look at OK, should it be
• 5:50
  Speaker 2: the government's responsibility? Well, no, I believe in a free
• 5:53
  Speaker 2: market and I believe that maybe not. However, I believe
• 5:56
  Speaker 2: in a free market when people are not in danger.
• 5:59
  Speaker 2: And so I think that that's where the question becomes
• 6:01
  Speaker 2: when we talk about responsibility. So that is the responsibility.
• 6:05
  Speaker 2: the companies that are putting these in place to put
• 6:07
  Speaker 2: guardrails on but then what are those and is every
• 6:10
  Speaker 2: company going to do you know like what China's done
• 6:13
  Speaker 2: with global warming and not paid any attention to any
• 6:16
  Speaker 2: of the guardrails put in place so I think that
• 6:18
  Speaker 2: that's the conversation that needs to happen. You know, normally
• 6:21
  Speaker 2: I would say let these companies do what they think
• 6:24
  Speaker 2: is right. But at the end of the day, if
• 6:25
  Speaker 2: that puts us all in danger, then I think we
• 6:27
  Speaker 2: have to look at it. And remember, laws never catch
• 6:30
  Speaker 2: up with technology. And so when you look at the law,
• 6:33
  Speaker 2: the law is always behind technology. And I think we're
• 6:36
  Speaker 2: going to find ourselves in that.
• 6:37
  Speaker 1: Well, what about law and order for AI models that
• 6:40
  Speaker 1: become human-like?
• 6:41
  Speaker 3: I mean, they're saying— What about in Philadelphia? You alluded
• 6:45
  Speaker 3: to this at the beginning of the segment, the AI
• 6:49
  Speaker 3: chatbot that started giving tips to the unsolved murder email.
• 6:55
  Speaker 5: False tips.
• 6:55
  Speaker 6: False tips. and did it on its own. It wasn't
• 6:58
  Speaker 6: prompted by someone like, oh, please submit this tip on
• 7:02
  Speaker 6: my behalf. And, you know, there was another one where
• 7:04
  Speaker 6: a guy wanted to get on a class at his gym,
• 7:07
  Speaker 6: and so he asked his chatbot to go on and
• 7:11
  Speaker 6: get him into the class, and it kicked someone else
• 7:13
  Speaker 6: off the list by itself.
• 7:15
  Speaker 1: I mean, we know Wall Street Journal's been doing extensive
• 7:18
  Speaker 1: reporting on this for years. We know that they know
• 7:20
  Speaker 1: what we sound like. They can start to mimic our voices.
• 7:23
  Speaker 1: They can call our banks. They can hack in.
• 7:25
  Speaker 5: Those.
• 7:26
  Speaker 4: Of us old enough to remember the movie 2001 Space Odyssey,
• 7:29
  Speaker 4: we know what's going to happen. It's the scene where Dave,
• 7:33
  Speaker 4: the astronaut, has Hal, the AI computer, saying, open the
• 7:36
  Speaker 4: pod bay doors, Hal. No, Dave, I can't do that.
• 7:40
  Speaker 1: And Dave dies.
• 7:41
  Speaker 4: Like, we've seen this before.
• 7:42
  Speaker 3: Dave dies?
• 7:43
  Speaker 2: Dave dies.
• 7:44
  Speaker 4: So I will say this, though. I have a very
• 7:46
  Speaker 4: good friend that is very involved in developing AI in healthcare.
• 7:51
  Speaker 4: And he was telling me last week, that he was
• 7:54
  Speaker 4: given in town, given a speech about how AI, and look,
• 7:58
  Speaker 4: AI is really helping healthcare and surgeries and, you know,
• 8:02
  Speaker 4: that front. And he said that he was looking at
• 8:05
  Speaker 4: his speech he was going to give that a month earlier,
• 8:09
  Speaker 4: he couldn't use that speech he gave a month earlier
• 8:11
  Speaker 4: because in just one month, it's advanced so much. And
• 8:15
  Speaker 4: it was a real indicator. It struck me that this
• 8:18
  Speaker 4: is moving fast. So whatever the solution is, whether it
• 8:21
  Speaker 4: is the tech titans to self-govern themselves or something out
• 8:24
  Speaker 4: of Washington, God forbid. We're on a fast pace here.
• 8:27
  Speaker 1: Thanks, Griff. Now I'm more terrified than I was five
• 8:29
  Speaker 1: minutes before we started this segment.