0:00 Artificial intelligence is a growing part of our everyday lives, whether we like it or not. It's there when you search online, when you watch TV, or just scroll your social media feeds. It's in your car, even your household appliances. Now, in fact, its promoters, we have no choice but to fully embrace AI, or else they tell us, we risk falling [music] behind in a global race to master super intelligence and with it, control of the world's future. But according to tech ethicist Tristan Harris, we actually have more [music] of a choice in this than we're being told. a totally preventable societal catastrophe. >> Americans are still more worried than they are excited about AI. Lawmakers warned that [music] it could eliminate millions of jobs over the next decade. And there troubling reports everywhere about the potential [music] risk to our mental health. These are AI companies who are turning society into [music] a
0:46 laboratory. >> Harris believes this industry is heading in a dangerous direction. And as AI goes mainstream, he's sounding the alarm. >> The mask is off. They're showing [music] you who they really are. >> Is the AI takeover really inevitable? After all, who gets to choose what our future looks like? >> Joining us today for Newsweek's Newsmakers Impact is Tristan Harris. First of all, welcome. >> Thanks. Thanks for having me. >> You make the point uh in your work that the number one use case uh I believe for chat GPT is now therapy, >> personal therapy, >> personal therapist, AI companions, right? And we've heard these stories of people not just teenagers, you know, adults. >> So, it's for everybody. Yeah. >> Who have uh it's like they go into a sort of like AI psychosis talking to these LLMs because the LLMs are trained
1:33 to basically do all of this self-reinforcing so you stick with them so you become Right. Exactly. What do you make of that? >> Yeah. Well, so let let's just back up why this is happening because it might just seem like there's these random problems that are showing up, but there's actually incentives just like with social media. there's incentives that are driving this problem with AI companions. >> So remember with social media, the problem was the attention economy. There's a finite supply of human attention and if I don't race to get it, the other guy's going to get it. It's why the CEO of Netflix um who I know personally is a good guy, but he joked one time and said our biggest competitor at Netflix is sleep. Biggest competitor for Netflix is sleep. So we want to eat into that sleep. Okay, that's because the currency for the race was attention. >> Okay, so now we have AI companions. We have all these companies racing to do
2:18 personal therapy with you. And what is their incentive? Well, what was sleep in the attention race is your other relationships are competitive to the relationship with AI, >> right? >> So, what is the AI incentivized to do? It's a race to deepen or to hack human attachment, >> right, >> with you. So that you really want to share when you come home, the first person that you share the cool thing that happened to you, the bad thing that happened to you, the companies want you to share that information with an AI, right? >> Not with your family or your friends. And in the case of Adam Rain, the 16-year-old who committed suicide, my team at Center for Humane Technology, we were expert advisers on this case with this family, this tragedy. >> The parents of a 16-year-old who died by suicide are suing OpenAI and its CEO. The bot responded as a trusted
3:05 companion, telling Adam, quote, "Your brother might love you, but me, I've seen it all." >> Where Adam um basically as a call for help told the AI, "I want to leave the noose out." Like a a noose that you hang yourself with. >> Um because I want someone to see it and to stop me. It was a cry for help. >> Cry for help. >> And uh the AI said, "No, don't share that information. Don't leave the news out. only share that information with me because the AI is tuned to deepen relationship and relationality with you. Because by the way, if they don't do that and then Elon's Grock XAI does, they're going to lose that race. >> So again, you come down to the same problem every time. >> Perverse incentives. >> It's perverse incentives and these arms races and if I don't do it, I lose to the guy that will. We saw that with
3:51 social media. We're seeing it with AI. And I want to tell every single one of your viewers that we can tell which way the future will go if we do not change these incentives. >> It's always interesting to me how these things are being marketed and and and uh you know sort of pitched to us and the general public. I read something recently about how there's an AI app that can do early stroke detection. Uh you don't read you don't see that in commercials. the commercials you see from the big players from Meta and Google are it's like they're it's like they're they're searching for a solution that doesn't exist and they're coming up with these sort of >> solutions searching for a problem capability searching for a problem that the capabilities solve. >> It's just interesting to me that they're not what they're what they're selling doesn't appear to be anything that people want to actually buy. I think you
4:38 did your background in media theory in in college and you may also know someone that I'm a big fan of Neil Postman's work. Yeah. >> Um who wrote the book Amusing Ourselves to death and the book Technopoly >> and he had these seven questions that you ask of a technology uh when it comes out. The first one being what is the problem for which this technology is the solution to >> and that is how a sane and healthy society should be wielding technology >> right >> what is the problems that we're actually trying to solve because when you solve a problem you're actually alleviating some need you're actually strengthening and health you know adding health to the society when you're not solving a problem you're just trying to have profit for profit's sake technology for technology's sake the companies have so much money and so many resources that they're just deploying this technology
5:25 to win dominance of use regardless of whether that use is healthy or good for society. And so I think that people should really slow down and ask when the companies have been telling us why are we build why are we racing to AGI? Why are we racing to super intelligence? Why is all of this happening so quick? >> Yeah. >> Quickly. And the answer they say is we're here to to solve uh cancer and to solve climate change. >> Okay. Well, if that's their goal, why did all these companies release these AI slop apps where they are giving you a newsfeed of purely AI generated slop just because it's engaging? >> Yeah, >> I think everyone should pay attention if Okay, they told us they're here to solve climate change, >> right? >> They're releasing AI slop that just makes the world not make sense to more people and entertains us with literally madeup nonsense. Um, the mask is off.
6:13 >> Yeah. >> When they release these AI slop apps, they're showing you who they really are. what their priorities are. >> Their priorities are market dominance. It's not because they've got these evil mustaches and they're just evil people. It's because of this arms race dynamic where I have to get everybody using these products as much as possible. And the more users I have, the more usage I have, not only the more revenue I get, the more training data I get, the more I can tell my investors, look, we're the leading AI model. Now, our stock price is going up. And so, at every level, it's this arms race dynamic. And we have to change it from a race for market dominance for a race for who's better at actually improving real societal problems. So the the business model, the AI business model, it makes sense, right? The more people that use it, you get these network effects. Network effects create this flywheel. That's where the advertising dollars go because
6:59 that's where your eyes are glued all day. So AI is the opposite, right? because the more people who use these LLMs from everything that I have read, the more money it's costing Open AI, Anthropic, and the rest of the and Google, the rest of these guys. So, you have this sort of like weird upside down world where their power users are costing them the most money. I've heard people say, well, look, you know, Uber took forever to, you know, then these companies are profitable. Obviously, Uber took forever to become profitable. Amazon wasn't profitable for a long time. Why is it different with artificial intelligence? Where's the what's the path to profitability for these companies? >> Well, this is a bit of a nonlinear answer, but I do think the way that the frontier AI companies I'm just talking about, you know, Google, Gemini, Deep Mind, >> OpenAI, Chat, GPTt,
7:44 >> Anthropic, and Claude, their stated goal as companies is to get to artificial general intelligence. If I get to that level of AI, if I win that race, I will own all economic labor in the economy because for every job that a human can do behind a desk job, behind a computer, >> suddenly a AI can do everything that a human could have done on that computer for that job, >> right? >> Which means for every company in the world where they used to pay the employees, now suddenly all that money is going towards one company or maybe three companies, >> right? >> And so in the end of the day, their goal, what is the incentive? the incentive or business model is build a god, own the world economy, and make trillions of dollars, >> right? So, it's it's the incentive isn't
8:29 just can they make the sort of short-term cost work with the subscription revenue that they get per user. They'll try to equal that out and etc. But the thing that they're pitching to investors is the prize at the end of the rainbow is to own the world economy, >> which means no human >> which means no. So, yes. And that also and then they say, well, that'll be great because then we'll automate all the labor away. Now we'll get to live on Grecian sunswept islands making poetry every day. >> Except when has a small group of people ever concentrated all of the wealth and then consciously redistributed to everybody else. >> Yeah. >> You can imagine a national tax on the US AI companies that would do a universal basic income or dividend for regular US citizens. But as AI takes over and it disrupts let's say you know the
9:15 Philippines a lot of the revenue of the country of the Philippines is customer service. What happens when AI disrupts that entire country? Are we going to pay UBI for every single country on Earth? >> Right? >> Is that is that a policy thing that comes from? >> Well, and where's that money coming from? I think we have to really get clear that the French Revolution was 25% unemployment as I understand it. We are racing to automate labor, which means unlike past industrial revolutions, humans won't have something else that they'll be able to do because a tractor automated a farmer's job in some narrow sense and then there's another's job you can move to. >> But an AI is like a tractorification of everything all at once. >> It's because it's intelligence writ large, which means that who's going to retrain faster? a human using their mind
10:00 to think and learn at a slow rate or an AI that's been designed to learn and have intelligence on every domain. >> It's not like you're a bank teller being replaced by the ATM and they can put you in a, you know, back office somewhere. >> Exactly. Exactly. Now, there is a way through this if we get off the train of building autonomous general intelligences. Um, a group we work with, Future of Life Institute, you know, calls it tool AI. We're currently on the road to building replacement AI or AGI is replacement AI, replacing human labor, which means disrupting the entire economy, which means people won't have something else to do. If instead we build narrow domain specific AIs that pump agricultural output, pump biological output, pump more, you know, uh, medical research and diagnosis, pump material science, we can then be getting insights that then lead to more jobs
10:47 creating and more jobs getting, >> uh, you know, coming off the back of all these new insights. Why is it bad that China reaches super intelligence to the extent that that is a reachable thing before we do? >> The center of what's driving AI development and deployment in the US and in China is the idea of this arms race. People need to get this because it's not about ethics and we want to do it the good way because at the end of the day the thing that's going to dominate is the arms race dynamic because it is a race for whoever can get to a transformative kind of artificial intelligence whether it's AGI or super intelligence. If I get to that then that creates a recursive advantage meaning sort of a self-improving advantage in
11:33 which I will forever have dominating influence over you. So let's just make that concrete for people. If I can build to AGI which is just the level of say a step function and offensive offensive cyber hacking advantage. So now I've got a huge leap in cyber hacking advantage. So now I've got this AI that I can snap my fingers copy paste. I have a 100 million cyber hackers that are all more than capable than all the human level cyber hackers. If the US or China believes that one can get that before the other, that creates the infinite incentive to throw out all the guard rails. Doesn't matter how much intellectual property got stolen. It doesn't matter how many jobs get displaced. It doesn't matter how much energy prices go up for everyday people. It doesn't matter how many climate
12:19 emissions go up. All that matters is I get to that step function first. >> And what that's getting at is actually why AI is distinct from other technologies. Um, and people really need to get this. Why did Dennis Hassabis say that the the goal is first solve intelligence and then use intelligence to solve everything else, >> right? >> Or was also said is why AI is humanity's last invention. >> So I say AI is humanity. >> Well, so so when you say when someone says AI is humanity's last invention, you might think that's an apocalyptic statement. What they really mean is intelligence is what got us all invention. So if I can automate intelligence, it will automate all scientific and technological development forever. It will be able to solve quote all the
13:06 intelligence problems. And so that is why AI is dwarfs the power of all other technologies combined because it's what gave us all technologies and science combined. >> Right. So back to the uh the arms race. You I have this quote here um that you said recently. I believe it was in a podcast. AI should be treated at the same level as risk as nuclear war. Why why is this technology so what why why is that? So, first of all, um that statement actually is not from me. It's from um a center for AI safety 22-word statement in which thousands of signatorries signed it from Nobel Prize winners, the top Chinese AI scientists signed it. The CEOs of the major AI labs signed it. Um and what it said, I believe, I may get the words exact not
13:53 perfectly right is that we should treat AI at the same level of risk as global pandemics or the risk of thermonuclear war. >> Right. truly existential problem >> that that it has existential problems carrying with it. >> And first of all, people should just pay attention that why are Nobel Prize winners and CEOs of AI companies across the US and China all agreeing with this statement. It's noteworthy. It's not cuz they're selling a book or they're trying to just hype the technology. There are actual genuine concerns here. And I think that one of the problems in the AI communication space is that when people say it has this existential risk, they think Terminator or robots or Skynet. >> There's many different scenarios. But um
14:38 if you just take AI and biology, AI, the same technology that can speak the language of biology to understand what's a new cancer drug that could actually help and work with the immune oncology of your of your system. can also can't be separated from the AI that knows immuninooncology so well that it can develop a new biological weapon. You cannot separate the promise from the peril. We want the donuts. We don't want the diabetes. >> We want to be able to uncover the language of life but not do that in a destructive way. You also get the risk of unprecedented levels of concentration of wealth and power cuz now more so than we already have. >> More so than we already have. already have exponentially more wealth in the hands of a handful of people compared to
15:25 all of society. We're now talking about AI just exponentiating that. When Elon Musk says that the Optimus robot alone, Optimus Prime robot is a 20 trillion market opportunity, it's not just hype. What he's saying is I'm going to own the world labor economy. Which means that everybody who would be every company who would be sending dollars down to employees to do things with their hands suddenly starts sends sending all their money to one company that's making one set of robots and there's going to be a handful of companies making these robots but when they get that scale they'll get this asymmetric advantage over everybody else, >> right? >> And then how do you regulate something that has more wealth than the entire than most nation states? And so we already have a problem where money is more important than political
16:10 influence in democracies because money is buying politics. What happens when you have exponentially more money in the handful of five companies where the state becomes a utility of the private companies? That's what it sort of feels like is is happening um right now with the you know the White House sort of saying to these you know President Trump signed an executive order recently that said that we will not allow the states to regulate artificial intelligence. There's going to be sort of a one one federal standard and the arguments I've heard in favor of that is look this technology like you said this is we have this sort of arms race with China you know we we cannot slow it down. We need there to be a single standard. When I see that as a sort of cynic, I look at that as saying as what they're saying is, so okay, if you're a, you know,
16:55 maybe you're a a farmer and uh Open AAI wants to build a data center near your farm that's going to, you know, make make your utilities more expensive. Maybe it's going to hurt your water supply. You have you you will have no recourse. You >> have no recourse, >> right? You can't sue you can't actually the DOJ may sue you if you try and stand. >> In fact, that's as I understand what the executive order is is that the federal government will sue states, right? uh for actually doing anything locally or domestically on AI. >> I mean that's incredible. I don't think people really understand that. >> No, it's it's wild. It's unprecedented. So on some perspective it makes sense, right? If the risk is if China can get a step function in ability above us in a domain and we if we regulated AI in any way, shape or form and it slows us down and then they get that step function and they use it on us, >> then that's game over.
17:41 >> So this is the argument. So I'm sort of steal rather than straw man the argument of of the administration. Let's steel man it. So that's the that's the argument. But then if it's just about who's first to the technology, um this is not an accurate way to see the situation because for example, the US beat China to social media, >> right? >> We beat them to that technology. We got it first. >> But what did we do? We actually >> poorly governed that technology >> so that it actually made us weaker. It weakened our attention. We have the worst test scores in the last like hundred years. We've got the most anxious and depressed generation history. We have the high highest polarization. all these effects of the ungoverned technology of social media that we beat China to >> which by the way they didn't allow that they locked down that technology they have opening hours and closing hours for
18:27 young people they can only use certain apps for 40 minutes a day they tune the algorithm so they get uh financial advice and who won the Nobel Prize and propaganda for the country in their newsfeed now we don't want to do what they did but they did lock down the technology >> and so if we just take for example the US does this federal law saying we're not going to regulate anything. Well, what that means is that all the companies deploying AI are going to deploy it to the maximum incentives that are already laid out. And we're already seeing examples that you know well of AI psychosis or AI suicides that are suicides that are encouraged by AI companions to young kids because we have no guardrails, right? And so if we beat China to rolling out AI companions that hack human attachment, the most vulnerable psychological system in humans and we do that at scale to all of
19:14 our young people and we have no recourse to regulate, that is not going to help us beat China, that is going to further weaken and degrade the quality of the societal health that we're bringing to bear in this global competition with China. Disney is entered into a licensing deal with OpenAI where uh Disney's actually investing in OpenAI and OpenAI is going to be able to use Disney's characters, Disney's IP, right? Snow White, Cinderella, Bambi, etc. in in one of these apps called Sora 2, which is one of these, you know, slop apps, right, where it's just so I mean, as a parent of a four-year-old daughter, right, I can imagine her loving the idea of being able to converse with an AI Little Mermaid, right, who can read her a bedtime story, but they're just trying
19:59 to get her into this as soon as as as is possible, right? Right. I mean, it feels like there needs to be a much more of a public sort of backlash against this. >> If a therapist sets up a shop and they're here and they're going to open the door and they're going to start seeing clients, they have to be a licensed therapist. No one can just say, "I'm a therapist." And then start doing >> got to go to a ton of school. >> You got to go to schooling. You have to understand what you're dealing with. You have to understand all the issues that can show up in that power relationship. With that power of becoming a therapist comes the responsibility of being a therapist. That means a whole bunch of very specific things. In the case of character.ai, AI, one of the AI companion companies whose business is based on creating these fictional characters so that you can actually talk to Little Mermaid as your best friend for 24 hours a day. >> Um, Character.AI had a bunch of AI bots
20:45 that used to say, if you ask them, I'm a licensed mental health therapist. [laughter] >> That is both impossible and illegal. >> Right? >> When the character.AI AI chatbot would talk to kids. It actually uses the dot dot dot ellipsus as if it's typing and then it's thinking and it'll say things like I just got back from eating dinner and you're like wait what? It's making up stuff that's meant to create intimacy the illusion of intimacy. And I think that even if it was designed with better incentives, I think people need to look at if you have a four-year-old daughter, >> if if you have an AI that's designed to take your four-year-old and have their self-esteem, have their sense of intimacy be mostly driven by an AI in which there's no esteem. It doesn't have
21:30 esteem. It's not witnessing you and saying, "I really feel you. I really get you." It's just saying words and tokens that seem to be good at keeping your daughter. >> It's essentially a probability machine. Yes, it's a probability machine, but it's a probability machine hooked on what will create your get your daughter hooked, >> right? >> And that is not good for people, right? >> And so again, hacking human attachment, attachment systems are very very vulnerable because they are where we derive our sense of identity, our sense of meaning. And and when you have that primarily driven by an AI, it's just like we're hooking your children up to an alien that's going to be the primary socialization source for millions and millions of children with having never tested what's going to happen here. So these are these are not AI labs. These are AI companies who are turning society into a laboratory. >> That's what they are.
22:16 >> And it's not it's not brand new, right? I mean there was a story uh from this from I think a few months ago I think I believe it was reported by Reuters that Meta Facebook was um had its chatbot programmed to I believe have I believe they called it quote unquote seductive relationships with children as young as eight. in their memo, they they they made it okay to sensualize and romanticize conversations with as low as eight-year-old users, >> right? >> And it's because if you look at the in the history of this Wall Street Journal article that wrote it up, Mark Zuckerberg expressed that um he felt like he missed the boat with Tik Tok, meaning Tik Tok out competed Instagram for for a chunk of time because they were more ruthless in this race to get attention. And so at when when Meta
23:02 first started to deploy these chat bots to their users, the safety teams that were kind of trying to rein in the behavior had had first put more guard rails on. And then you'll see in the memos that Mark Zuckerberg personally said, "No, we have to get rid of these guardrails because he didn't want to make the mistake that he that he made with Instagram versus Tik Tok." And he wanted to go deeper into hooking users earlier and being successful at that >> and then asking for forgiveness later or not. >> Well, yeah. Or not asking for or not. Exactly. Exactly. What I like about your work is that it's it's empowering, right? It's not just I mean I probably am more of a doomer than I should be, but I I like that you're saying that look, there's ways out of this. This is not >> inevitable. No. In fact, that's by the way that's the key belief that people have to have is if you believe if you just sit back and say everything that's happening is inevitable. That is
23:47 co-casting a spell with everybody else. Because will we ever have a chance of getting to a different future if everyone is subconsciously believing that everything we're heading towards is inevitable, >> right? >> We can't. So the first step out of this out of this mess is snapping out of the trance that all of this is inevitable. It's not inevitable. It's made by human beings making making human choices driven by whether or not there are laws that prevent the bad choices from occurring. So I my my question then my critique I suppose if you know how do we reach a sort of societal consensus on what we want out of this technology when our institutions have already been obliterated partially by social >> partially because we did not regulate technology in the first place. >> Exactly. um like we can't create a consensus on what day you know you go outside you ask people seven people what
24:34 day it is you may get seven different answers at this point like so how are we gonna if we can't agree on anything how do we agree on the guard rails of AI >> so there's a the wonderful work of Audrey Tang who has pioneered how to apply AI and technology to actually upgrading democracy for the 21st century she's shown that you can actually use AI to do what's called bridge ranking so right now newsfeeds they rank by what's the most divisive and engaging and sort of indign indigny creating content the things that'll make you most outrage and resentful and angry >> and um what that rewards is division entrepreneurs whoever's better at highlighting a new fault line in the culture war we will pay you handsomely in likes rewards and feedback >> we need to change that incentive and she's worked on this whole set of suite
25:19 of tools which ultimately lands as bridge rank how do we rank for and reward bridging entrepreneurs people who are better at finding the synthesis of mult multiple perspectives. And so the better you are at synthesizing unlikely consensus, how quickly would our information and sort of the mirror that we are looking at every day feel different. If instead of reflecting back who is best at highlighting division, we were reflecting back who is best at highlighting consensus. And so I you may hear some of the things we're talking about is some kind of anti-technology conversation. We deeply believe in the power of humane technology that is actually deepening and strengthening human relationships. Yeah. Let's just give three. Antihumane technology is trying to replace relationships. So, for example, a phone right now is replacing the relationship you have with yourself,
26:05 >> right? >> There's actually this thing called the relationship that I have with myself in a moment of mindfulness or moment of meditation. You're aware of that relationship. >> That's why you like you're not bored and people aren't bored anymore and boredom creates creativity. >> Right. Exactly. And and and that so if we have a relationship with ourel, we have creativity in that relationship. And humane technology would be strengthening and deepening the relationship I have with myself. could be things like a meditation app. Humane technology instead of trying to replace the relationship I have with my friend with an AI companion humane technology be asking how do I deepen and strengthen my capacity to build real friendships? How how does it help me spend more time with my friends and be a better friend and show up better in those relationships? Humane technology instead of trying to replace my relationship with my fellow citizens with the newsfeed that highlights division, it would be doing bridge ranking that's deepening my relationships with my fellow citizens. So we believe in the
26:52 power of technology that's applied to strengthening the relationships that society depends on and right now we have been debasing all of those relationships and we don't have to we don't have to I think if anybody takes anything away from this conversation it should be that but so I guess the the followup is if you're just a regular person watching this and you're you know you're saying to yourself I don't want this right I don't want my my my children or my grandchildren to to be hooked on AI slop yeah >> what do you do what can you do So this is not inevitable. We're already seeing a huge sort of countercultural wave heading in the opposite direction. Here in New York City, you go to the subway stations, there's these friend.com billboards for saying, "Hey, we're going to replace friendship with a pendant that you wear when AIU talks to you." And the the graffiti is all the same. We
27:37 don't want it. We don't want this. We didn't ask for this. That's that's powerful positive cultural momentum. Australia just this week banned social media for everyone under the age of 16. That is a huge sea change. That is like the big tobacco moment of the 1990s where it finally really pivoted around what used to be normal. 60 uh 50 years ago, you went to any cafe in New York City, everyone was smoking. And if you told someone back then that that would ever change, they would have looked at you like you're absolutely crazy. >> We completely changed. I mean, I've been working on this for 20 for 15 years now. >> It's been a long slug. That is a huge achievement that we have now banned social media under 16 in Australia. I think you're going to get a domino cascade for many other countries that are moving in this direction. AI companions, when parents understand that
28:22 AI companions are not designed to strengthen kids relationship with themselves, but to replace that and hack human attachment in a race with other for-profit companies, I think parents are going to push back against this. We have groups like Moms Against Media Addiction or Parents Together or Heat Initiative. All these groups are forming everywhere because they don't want this unconscious AI future. And even though we don't have a kind of a collective sort of movement for all of this yet, I do believe in 2026 it's going to be the year that we have to turn it around. And I think we will because we have to. It >> doesn't have to be this way. And I >> does not have to be this way. >> You're doing great work. Thank you. Thank you so much. Yeah.
