Here’s Our Roadmap to a Better AI Future

In order to shift the incentives of AI — the trillions of dollars in investment, the race to geopolitical power and dominance — it’s not enough to simply understand the problem, we need real action.

That’s why CHT is proud to release “The AI Roadmap,” a report outlining seven core principles for how AI should be built, deployed, and governed, each grounded in real, implementable solutions across three domains: norms, laws, and product design.

In this episode, Camille Carlton and Pete Furlong from CHT’s policy team explore the concrete steps we can take today to get off the default path and forge a better AI future. You can read “The AI Roadmap” on our website.

Tristan Harris: Hey, everyone, it’s Tristan Harris.

Aza Raskin: And this is Aza Raskin. Thanks so much for coming to listen to Your Undivided Attention.

Tristan Harris: Many of you will have seen The AI doc by now, that’s the new film that we just did an episode with the filmmakers. If you haven’t seen the film, there’s still plenty of time to go see it in theaters. It’s everywhere all throughout the US, and soon to be hopefully internationally. And Aza and I are really excited about the work that this film can accomplish. Because, in essence, what we’re trying to do is create clarity that will create agency. That if everyone knows that everyone else knows that there’s a problem up ahead and the way that AI will land us in a future that nobody wants, if everybody can see that clearly, then we can collectively put our hand on the steering wheel and steer to a different future. And I think the question and the thing that the film leaves unresolved is, how do we steer? How do we get to that better future with AI?

And that’s what we want to talk about today. What are the actual steps that we can take today to prevent the worst case scenarios? There’s a spectrum of futures available to us. We may not be able to get to perfect. There’s going to be some damage. And also, we can still steer. There’s still time for that.

Aza Raskin: And just to say, if you haven’t yet seen the film, I think one of the things the film does very well is it scoops everybody up. It really represents all sides, not just fairly, but strongly. That if you are really excited about the benefits that AI can bring, the film not only talks about those, but points out that most people don’t go far enough in the benefits. And same thing on the downsides. It really highlights the downsides, highlights the AI race to deploy that is creating those catastrophic risks, and then points out that actually most of the risks that people think about aren’t big enough.

And what I’m excited about for this episode is that when everyone sees that the direction that we’re going is one that we are not going to want to live in, whether you are a teenager who’s not going to have a livelihood growing up, whether you’re a teacher who’s having to watch their kids have cognitive decline. All the way up to you’re the head of a major corporation. Seeing the direction that this goes gives us the opportunity to choose a different path.

Tristan Harris: One of the main problems is that this feels too big for any one person to solve. And Aza, you speak to this scale metaphor of, okay, the problem is this trillion dollar machine advancing AI as fast as possible on the most reckless path. And there’s this question of, how would we change that? Imagine the scale. What’s something on the other side of the scale that’s of equal weight?

Aza Raskin: Imagine, I just want everyone to close your eyes for a second and imagine there’s a scale, like a balancing scale. On one side you see the problem, this is trillions of dollars of investment going into making uncontrollable and scrutable AI. There’s the race for the one ring, geopolitical power, forever dominance. That’s pulling the problem side down. And there on the other side, just imagine there’s you, hearing about this problem, and what is your reaction going to be? Well, it’s going to be denial, despair, deflection. And so, what is the only thing really that we could imagine that can shift those trillions of dollars of incentives? Well, it’s all of humanity. It’s like we’re going to need a human movement that can balance out those scales.

Tristan Harris: Now, it all starts with, first of all, just not feeling overwhelmed. That’s one of the first steps, that there is another path, but it would take a lot of people doing a lot of things. The second is that we have to break the trance of inevitability. If, on a subconscious level, you just feel like it’s all over and it’s just all going to be inevitable and there’s nothing we can do, the problem with that belief is that it is co-complicit with enabling that bad future to happen.

Aza Raskin: And so, that change from believing something is inevitable, impossible to change, to believing that something is just extremely difficult and perhaps the hardest thing humanity ever has done, that gap is critical because it means there’s still something to do.

Tristan Harris: When I think about what is going to fight back against that, it’s something the scale of humanity and human values writ large, protecting the things that we care about. When you grayscale your phone and turn off notifications, that’s the human movement. When you see graffiti on an ad in New York City for a AI product that no one actually needs, that’s the human movement. When you see people gathering together for a dance party and you check your phones at the door, that’s the human movement. When you see people saying, “I’m going to learn a language instead of falling into brain rot doom scrolling at night,” that’s the human movement. And it’s not just that, obviously, it’s about how we activate in the world.

When employees threaten to resign because they don’t think that AI should be used for mass surveillance or we’re not doing things safely in us. And when you see that countries like Australia, Denmark, Spain, France are all banning social media for kids under 15 and 16, and I believe several US states now are banning social media for kids under 15 or 16, that’s the human movement. And already nine states have introduced bills to restrict AI personhood so that human rights are for humans, not for protecting AIs. 45 states have specifically addressed sexually explicit deepfakes. And these laws send a huge signal that non-consensual exploitation of AI tools is a serious offense, and we have to actually take action on it. There’s actually a lot that’s happening and most people just don’t see it.

Aza Raskin: I want everyone to just stop for a second. Because, at least for me, I feel something different in my body. I feel like hope, I feel energized. And I just want you to hold onto that feeling, because that is the feeling that’s going to enable us to make sure that AI, the way it’s being rolled out, actually isn’t inevitable. This can be everything from if you’re really good at doing international coordination, track two dialogues, bringing countries together, it’s not most people, but if you are, that’s part of the human movement. But it’s also tiny little things like you’re sitting on an airplane and you put down your phone so that you can smile at the baby, the seat behind you, and they giggle back. That’s also part of the human movement. This is about taking back what it is to be human, but not in the abstract sense, but in the everyday tangible sense all the way up to the international sense.

Tristan Harris: Exactly. And of course, what we’re going to need ultimately are laws that are passed, because you have to bind to these multipolar traps of, if I don’t do it, I’m going to lose to the other one that will. But we’re already seeing that happen. We’re seeing several states work to pass bans for legal personhood for AI, meaning AI should be a product, not a person. Human rights were for humans. And we’re seeing already US states move in that direction.

This is not something that’s hypothetical, we’re seeing liability laws for AI be advanced in several states. We’re seeing age-appropriate design codes. If you actually just got the iOS update on your phone, you’ll notice when you open up, I think Anthropic, it happened to me yesterday, you have to verify that you’re above the age of 18. We now have age gating in every Apple device. That was something that many of us have been working for over a decade to make sure that happens. So stuff that was hypothetical that was, “Hey, we’re going to need a big tobacco trial for social media and the engagement model.” Aza, you and I were talking about that in 2013. It’s actually happening. It took 13 years for social media to go from, this is never going to happen, this is impossible, to now it’s finally turning around. Now, AI looks impossible, but just zoom back to where you were 13 years ago, it also felt impossible then.

Aza Raskin: There’s a really important thing that everyone can do to be part of The Human Movement, at least in the US, and that is the midterm elections are coming up. We want everyone to research the politicians that you’re going to vote for and start demanding that they take stances that are about, well, being part of the human movement, fighting back against the encroachment of AI and livelihoods in surveillance and in every way that things encroach on us. That is one of the most important things that you can do.

Tristan Harris: We have to make AI go from not even on the top five list of priorities for politicians who are looking to get elected saying, imagine that their phone literally never stops ringing and it’s, I’m not going to be voted for until I know that you are going to stand for a pro human future. Whether that’s how you’re pushing on data centers, whether it’s how AI is getting deployed in schools, whether you’re protecting people’s jobs and people’s livelihoods in the face of all this AI disruption.

Aza Raskin: Yeah, exactly. Are you pro human? Are you pro machine? It’s very simple.

Tristan Harris: And The AI Doc I think makes that clear that the default path is not a pro human future. And if everybody sees that we can collectively choose, both in small ways and big ways, you’re already seeing mass boycotts of OpenAI’s product and on subscriptions because of the drama that went down between the Department of War and Anthropic where the AI models would’ve been used for mass surveillance and autonomous weapons. I think Anthropic’s downloads surged by 250%, or something like that. If millions of people switch who they’re paying for, we are voting with our dollars. And if businesses do that, if church groups do that, if families do that, if communities do that, that can have a really big impact on which world we’re heading towards.

Aza Raskin: One of the challenges, as you know, Tristan, of thinking about AI, is that AI is automation of intelligence, and intelligence has shaped and touches absolutely everything about our world. Everything is touched by intelligence, so everything is touched by AI. Which means that the scale of the problems, it’s just it’s too much to hold in one head. And to say the phrase, if the world is pretty good for machines, is to start to invoke, well, that we’ve seen this movie before. And I wanted you to talk a little bit about this framing that we’ve started to brainstorm about actually the way that we can stop from living in the dystopian movies we’ve all seen.

“One of the challenges of thinking about AI, is that AI is automation of intelligence, and intelligence has shaped and touches absolutely everything about our world. Everything is touched by intelligence, so everything is touched by AI.”

Tristan Harris: Yeah. Let’s just rotate the entire problem from the lens of, haven’t we seen this movie before? Like Elysium or Hunger Games, you have this handful of trillionaires who live above the law where everyone else basically works and is in poverty and fighting and eating each other. And you see that we have WALL-E, where the future where the fat humans are caught in a doom scrolling loop, getting more brain rots, attention spans being harvested. Or Idiocracy, where you dumb down the population until there’s nothing left.

One way to think about solutions is we need laws and we need norms and changes in culture that prevent each of these bad movies. Instead of saying what laws do we pass, imagine there’s just a No WALL-E law. It’s a set of laws that prevent the mass attention economy, brain rot, shortening attention spans, et cetera. It means AI and technology that are designed to protect human vulnerabilities and protect our freedom of mind, not be predated on exploiting it.

And imagine instead of Her, Her is a movie about AI companions where Joaquin Phoenix falls in love with his AI. Well, we can have a Prevent Her law, and that includes no anthropomorphic design, liability for suicides and these kinds of problems. And where AI is designed as the outcome of that law to strengthen human capacities and build deeper human relationships as opposed to redirect people from their human relationships and deepen their relationships with AI.

Aza Raskin: Or think about the No Blade Runner law, or maybe the no replicant law. And that says your legal rights are reserved for you and other humans and for things in nature. And that when human beings launch their chatbots or agents onto the world, that the human being that did it or the corporation that did it are responsible. They’re held legally liable.

Tristan Harris: Yep. And that AI agents should have driver’s licenses. If you’re unlicensed AI agent that’s doing havoc in the world, it’d be like a car that’s swerving through the highways with no license plate on it. Well, I’m sorry, you’re going to go to jail. And there’s some simple other laws like No Big Brother, No 1984. It’s pretty simple. Don’t create mass ubiquitous surveillance that can go all the way down to decoding every aspect of someone and re-anonymizing them. We need laws that prevent that surveillance. Or the No HAL 9000 law from 2001: A Space Odyssey. “Open the pod bay doors, Hal.” And he says, “I’m sorry, Dave, I can’t do that.” We’re actually building the AIs that are currently disobeying commands, avoiding shutdown, and we need laws that say you cannot ship AIs into sensitive infrastructure that we can’t verify are controllable.

This is not a partisan issue. There’s essentially people who want the anti-human machine and don’t mind if we basically disrupt everyone else’s lives, and there’s the people who want a pro human future. And that’s what we want to invite people into. There is a movement for a pro human future, and we can all get behind preventing a bunch of these bad movies, from Terminator to Elysium, to WALL-E, to Idiocracy, to replicants, to Big Brother, and to HAL 9000.

Aza Raskin: Just about now people are starting to think, okay, that’s wonderful at the highest level, but what specifically concretely can we do? What laws can we pass right now? No one solution can possibly solve a problem this big, it’s going to take an ecosystem of solutions and an ecosystem of people. The forces that are moving to make this right have to exceed the forces that are moving for the anti-human machine future. And here I want to turn it over to some of the specifics of what our policy team at Center for Humane Technology has been working on.

Sasha Fegan: Thanks so much, Aza. Hi, everyone. I’m Sasha Fegan, I’m the executive producer of Your Undivided Attention. And I have with me here, Josh Lash from the podcast team, who’s making his podcast debut. Hi, Josh.

Josh Lash: Hey, Sasha. Thanks so much. I’m really excited to be here and I’m really excited for this episode. We’ve been trying to think of the best way to present some of the internal work that our policy team here at CHT has been doing behind the scenes, coming up with ideas for actions, concrete actions that we can take right now to meet this moment in AI and to respond to the challenge that the film throws down for all of us to build a movement to steer the direction of AI towards a more humane technological future.

Sasha Fegan: Yeah. Joining us now, we’ve got Camille Carlton, who’s the policy director here at CHT. And Pete Furlong, who is our senior policy analyst. And together with the efforts of a lot of other team members at CHT, they’ve just released a report called The AI Roadmap: How We Ensure that AI Serves Humanity. And you can find it on the CHT website and also in the show notes.

Josh Lash: Yeah. And we’re not going to go into the whole thing today on the show, but we really wanted to highlight some key parts of the report because it does something really rare that I haven’t seen anyone else in this space do yet. Which is that it doesn’t just stop at identifying the problems that we’re facing, it actually has this clear vision for the AI future that we want. And it has a roadmap to get us there. To tell us more about this report and to get you all, our wonderful audience, engaged in what needs to happen next, here are Camille and Pete. Welcome to Your Undivided Attention.

Camille Carlton: Thanks for having us.

Pete Furlong: Yeah, thank you for having us here.

Sasha Fegan: This report’s coming at a time when so much of the conversation around AI is couched in this very deep, unmovable feeling of inevitability. There are a lot of concerns about the negative effects on our kids, our classrooms, our relationships, and even early fears, but big fears around how it’s starting to impact the employment market, and particularly white collar jobs like computer scientists. It’s all starting to feel like this is just inevitable. But what I think I get from reading this report is that it’s actually not inevitable and that we can shape the direction of AI. Camille, how do we do that?

Camille Carlton: Yeah. To start first, the feeling of inevitability is so understandable. The scale of the problem we’re facing is massive, AI touches so many aspects of our lives. But this feeling of inevitability is also probably one of the worst things that could happen to us as a society, because we stop believing that we have agency and we stop believing that a different path is possible. And there is not one single solution that can solve this. No one solution will ever be enough. But it’s important that we see that there are solutions. There are concrete steps we can take to steer us off the path we’re on and towards a better future.

And of course, change builds on top of change. Small wins are like snowballs that can eventually turn into an avalanche of positive change. But before we steer, we also need to figure out where exactly we’re going. And that’s why, for us, our report really starts with seven principles for how AI should be built and deployed and used. Principles that give us a clear vision for the future we want to end up at. We really think of the report as a roadmap for how we get there.

Josh Lash: Yeah. And I think before we dive into these individual principles, what is that vision? What does a humane future look like?

Camille Carlton: A humane future means different things to different people, and we really try to incorporate the range in which AI touches on so many different parts of our lives. We really imagine a future where there’s clear accountability for the harms of AI products, where AI elevates our human ability rather than replacing it, where human identity and empathy is respected, not bought and sold. We imagine a future where AI is used to supercharge democracy and rights instead of concentrating power in the hands of a few companies, a few individuals. And where the capabilities of future AI products are transparent, and there are strict laws and lines about how we want AI built and used. It’s a future where the power of AI products and the people building them are matched with wisdom and responsibility. And, frankly, it’s just not the future we’re headed towards right now.

Sasha Fegan: Yeah. That’s the sense I get from hearing the principles, that so many of them really just seem like common sense. Of course we don’t want to build machines that replace us. Of course there should be accountability and reasonable limits. And absolutely, I think everyone listening to this would think that we need to protect things like dignity and democracy. But it really doesn’t feel that we are headed in that direction, and so we do need to repeat those things and articulate those principles.

Josh Lash: You could think in a show like this we might be talking about small design tweaks or wonky policies, but we’re really talking about the things that give our lives meaning, like our relationships, our jobs, our freedoms.

Camille Carlton: Yeah. And I think that because AI touches so many of these areas, it’s forcing us to really, as a species, ask these big questions about what we value in life and what type of future we want to see. The broadness of the report is in fact really commensurate to the task at hand in the fact that we are all reckoning with all of these different parts of our lives at once.

Pete Furlong: Yeah. And I think we wanted to root this report in the future that people want, not the one we’re being sold by a limited few AI companies. And I think it’s important to recognize that there’s broad support across the public and across political divides for many of these ideas. And that’s something that’s reflected in a lot of the examples that we give here.

I think we started first by identifying, where’s the current path that we’re on, and what’s the problem with that trajectory? Really, just trying to get a good sense of the problem that we are trying to solve and then thinking about, what’s the future that we want? What’s the alternative here? And that’s really where we think about building up this principle from the ground up. And so, what are the steps that we need to take to get there? What are the cultural norms that we need to change? What are the laws that we need in order to better regulate AI? What are the design changes that we need? How do we change the way that this technology is built?

And I think it’s important to recognize that these aspects: norms, laws, and design, they all work together and they’re really mutually reinforcing. Shifting cultural norms strengthens the public’s demand for more durable legal protections. And laws are something that create accountability that drives safer product design. And when we see safer product designs, that shapes the public experience of these technologies. These are things that really act together, and together is where we see the outcomes that we want and build towards that better future.

“These aspects: norms, laws, and design, they all work together and they’re really mutually reinforcing…These are things that really act together, and together is where we see the outcomes that we want and build towards that better future.”

Josh Lash: Can you give us an example?

Pete Furlong: Yeah. I think one of the examples that’s really important from this report is that right now there’s really no clear legal mechanisms in place to hold AI companies accountable for the harms of their products. And this is a really important problem. People are actively being harmed by AI systems, and we can expect those harms to grow as AI becomes more deeply embedded in our day-to-day lives. That’s the problem.

And I think the solution that we want to build towards, the better future that we want, is that really in an ideal world companies should be taking into account our safety in the design of these AI products. And I think when something does go wrong, whether that’s one of the many cases of AI enabled psychosis or suicides that we’ve seen, or even an AI agent deletes your entire company’s code base, which is a real example that we’ve seen, the company that puts that harmful product out into the world needs to be held accountable.

Josh Lash: Okay, that’s the problem, that’s where we want to get to. And so to get there, we need to shift norms, laws, and designs. Let’s start with norms. What are the norms we need to shift? How do we need to shift the way we think about AI?

Pete Furlong: One of the norms that we agreed upon, for example, was that AI is a product, and therefore carries product liability. We need to stop thinking about AI as a service and start thinking about what it is. It’s a product. Just like with any other consumer product, the people building their product have a clear duty to their users to make that product safe. And if they fail to do so, consumers deserve accountability. And this is something that we’ve actually seen AI companies challenge, both in court and in lobbying and in legislation. The argument there is that AI outputs are a form of speech. And so, fundamentally underpinning this argument that companies are making is the idea that it’s not a product. This paradigm that we have and we’ve used for centuries around product liability doesn’t apply to AI. And that’s the argument that AI companies are making in this case, and something that we think is deeply problematic.

One of the other norms that we talked about here was that responsibility for these products should lie with the companies, not just the people who use them. Companies are advancing this narrative that if someone’s harmed by an AI product, that’s on them. But I think it’s important to recognize that many of the harms we’re seeing are a result of how these products are designed.

Camille Carlton: I think also, Pete, one of the things that you and I have talked about with the norms that we’ve outlined here of AI is a product and companies are responsible for the harms, not users, is that they are direct counters to the narratives that tech companies have been putting out for decades. We’ve had huge companies putting out narratives that shift the way we think about them, their products, their responsibility, our role in using their products. And that changes how we as individuals behave, it changes how we regulate. And so knowing that, okay, there’s actually a different way to look at it is part of the process of getting us to the better path we want to go on.

Pete Furlong: Exactly. We expect car manufacturers to install seatbelts and airbags. Why can’t we hold AI companies to a similar standard? And I think it’s important that companies take reasonable steps to mitigate risks in the design of their product. And this is something when we talk about laws that reinforce that norm, that we actually have a policy framework here at CHT that goes into much more detail on this. And we can link to that in the show notes. We also have seen different states as well as a federally proposed bill, the AI LEAD Act, which seek to define AI clearly as a product in legislation. There’s a number of different approaches to trying to address this.

Sasha Fegan: Hey, do you have a sense that there’s bipartisan consensus on this?

Pete Furlong: Yeah. The bill we’ve seen introduced at the federal level is sponsored by Senators Durbin and Hawley. It has bipartisan co-sponsors. We’ve also seen bills adopting the same strategy across red and blue states. And I think part of the reason that this approach appeals in a bipartisan way is that it’s pretty common sense. The nice thing about it as well is that it’s pretty flexible. We don’t need a lot of really prescriptive regulation when we have this form of embedded accountability. I think that’s something that appeals to folks on both sides of the aisle.

Josh Lash: And I think that’s something you see throughout this report is that so many of these issues are truly bipartisan. And I think that’s rarity these days, and I really love that about it.

Sasha Fegan: Let’s move on to another one of the principles that really struck me, which was around the idea of we need AI that respects our humanity and doesn’t exploit it. Can you just get into that a little bit more and explain what you were getting at there, Camille?

Camille Carlton: Yeah, definitely. This is something that I think we hold really closely at CHT, given the work that we’ve done supporting different litigation cases. But the problem that we’re really seeing here is that AI companies right now are treating users like commodities. Because the personal data that we, as users provide these companies about ourselves, our innermost thoughts, our feelings, as well as our interactions with our products is incredibly useful in building and improving AI models. In fact, leading investors and companies openly describe this as a magical data feedback loop where intimate user interactions are continuously improving the product.

Sasha Fegan: And now... Sorry, I’m just going to say, I just want to double hit on that, because that is shocking actually to hear that. That really we’re just vessels for data extraction. It’s so debasing on a human level.

Camille Carlton: And this isn’t the first time that users are the product. We’ve seen this before with social media and the race to attention. It was very clear in the advertising model, and now it’s gone even a level deeper. It’s really this race to intimacy, where companies are designing products to look and feel human. They use human speech patterns, they speak in first person. There’s even a little ellipsis to indicate that these products are thinking. Sometimes, depending on the product itself, you might even hear a backstory about the AI that you’re talking to. And so there’s this, again, intentional design to mimic our humanity.

And not just that, it goes beyond that, because there’s some things about these AI products that aren’t human. They’re always on, they’re always available, but they also always validate your beliefs even if it’s not in your best interest. There’s just generally this sense of the product will do whatever it can in order to keep the user in conversation. And why? Because the bigger the model, the smarter model, the more likely a company is to make it to market dominance to get to profits.

Sasha Fegan: Yeah. And I think those profit incentives are clearly there, but how do we change that? What’s an example of how we change those norms, change the design, and also change the laws?

Camille Carlton: One big norm here that we have is pretty simple, but it would have really big impact. It’s the idea that we shouldn’t humanize AI. When we think about AI, we need to really clearly preserve the boundary between what is human and what is a machine. And this goes into product design, like the things that I was saying about how the products are built to be in first person, but humanizing AI also goes beyond product design. It’s also about not humanizing AI in our legal system by granting it legal personhood, which is something that companies have been pushing for. Granting an AI legal personhood would not only limit accountability from AI companies, but it would really tip the scales between AI and humans when it comes to legal rights and protections.

Josh Lash: Wait, sorry, can I jump into your... AI legal personhood, this is the thing that’s being considered?

Camille Carlton: Yeah. When we worked on the character.ai case. character.ai essentially argued that the case should be dismissed because their product outputs should be considered protected speech. The text coming from the chatbot should be considered protected speech under the First Amendment. And now they argued this in a backdoor manner using their listener’s rights. But the implications of this, of extending First Amendment protections to a chatbot would be the beginning of what we call legal personhood, which is something that corporations already have. But the implication would be really different because it shifts accountability away from the company into the chatbots, the products itself.

And when you think about how to operationalize this, it gets sticky. You have someone who has been harmed and suddenly they think that you’re suing a company for the product that they made. But if suddenly you’re not suing the company, you’re suing the chatbot itself, how do you change the chatbot’s behavior? How do you receive damages from the chatbot? And so, it creates this liability shield for companies if we’re looking at a world in which legal personhood exists.

Josh Lash: Yeah. And it just strikes me as you’re saying this, this is how these ideas build upon each other. We just talked about accountability and product liability, but this is another level of liability and accountability that we need to be aware of and thinking about. And I, personally, don’t want to be on the same legal footing as an AI chatbot. That seems like a really bad idea. And anyway, I’m sorry, keep going.

Pete Furlong: I was just going to add, I think it’s important to recognize that this is also connected to product design as well too. And so all of these things are interconnected. When we talk about humanizing AI, these companies are building these products to reflect our humanity. That’s a design choice on their part as well, and it connects to their legal strategy.

Sasha Fegan: Yeah. And I think that’s so important. And definitely, Camille, you mentioned the character.ai case, which CHT worked on. Which, just to remind listeners, was the case of a 14-year-old boy, Sewell Setzer, who took his own life after a very intimate relationship with an AI chatbot. And we also worked on the Adam Raine case, which had a similar trajectory of a young boy taking his own life out of a relationship with ChatGPT. And as you said, these cases could have turned out so differently if the products were designed differently.

Josh Lash: Yeah, exactly, Sasha. And we should note that in the report itself there are design standards that AI companies can turn to if they want to build their chatbots better in accordance with this principle. We should also note that there are states like California, Oregon, and Utah that are considering bills that would instantiate some of these design standards into law. There’s real momentum on this issue.

Sasha Fegan: I want to move on to other harms which are really evident out there in the zeitgeist, and that relates to the impact of AI on jobs, and particularly the potential automation of work. We hear a lot of stuff about how AI is going to put massive amounts of people out of work. I want to press you guys, what can we do about that? What does the report say about AI and jobs?

Pete Furlong: Yeah. I think the north star that we’re striving for here is pretty simple. We believe that AI should be built to augment human labor, not replace it. And I think you’re right, Sasha, that today’s AI systems are built with replacement in mind. Trillions of dollars are being poured into AI companies because only mass scale automation of our economy could make that investment worthwhile. And I think no one really seems willing to play the tape forward and understand and imagine what this means for all of us. But we believe, really, that it should be a fundamental principle that people deserve access to work, they deserve a living wage, and they deserve economic security. And that they should have a seat at the table when decisions are being made about technologies that will impact their core livelihood. Really, this requires all of us, and especially the people building artificial intelligence, to rethink our beliefs about AI and work.

The goal of improving efficiency, the goal of adopting new technology should be to improve the lives of people. An AI that displaces workers or devalues labor is undermining the very systems that we have in place to support people. And that’s not something that we want here. And then, also, I think that we need to recognize that work provides more than economic value to people, it also provides meaning and purpose. And that to lose work entirely, even if we found a way to provide people with a safety net, would strip people of a lot of what matters to them.

Josh Lash: Yeah. This is a topic we’ve covered a lot on this show. I actually would highly recommend our episode with Michael Sandel, who has written a lot about the importance of work to human dignity and human meaning. And I agree with everything you just said, but again, I’m just struck by the fact that the incentives we have today are not pointing in this direction. It’s so much easier for companies to treat labor as a line item and to see automation as a way to just boost profits. We’ve talked about norms. I agree we need all those norms. But at the end of the day, what are the laws that we need to start thinking about here?

Pete Furlong: Yeah. I think it’s important to recognize here that this is a really complex problem. Our economy is a complex system, and there’s no silver bullet policy that’s going to change the incentives at play here. Instead, really what we need to be thinking about is a platform of approaches and a platform of different policies. This could look like a tax system that’s designed to prioritize spending on labor over replacing people with AI. We’ve also seen different economists propose things like apprenticeship programs to help with workforce development. And I think the other thing that’s really important here is we need to make sure that we reinvest some of the gains from artificial intelligence towards helping the people that are displaced by it. Really, this means that leading AI companies need to help subsidize some of the reforms we’re talking about here.

Josh Lash: Are we seeing politicians start to think about these laws? Are they at all responsive?

Pete Furlong: Yeah. I think it’s something that a lot of different folks on both sides of the aisle are starting to consider. We’ve seen a number of different bipartisan proposals at the federal level to do some better research so the federal government can understand the impact of artificial intelligence on our economy. I think it’s something that we can expect to be a pretty frequent talking point as we approach some elections later this year. I recognize the economy is something that everybody cares about. And so, if this is going to be one of the biggest impacts on the economy that we’re going to see, then politicians on both sides of the aisle are going to have to take action.

Josh Lash: Yeah. Yeah. I just think it’s worth emphasizing what you said earlier, which is the way to justify the trillions of dollars of economic investment that you’re doing is widescale automation. That’s the plan. Whether or not they’re successful is up to us, but that’s the plan.

Pete Furlong: Yeah, that’s exactly right. And this is something that we’ve even seen a lot of the top AI CEOs admit. They’re saying that their technology can replace a lot of the different jobs that we have. But they’re not really proposing a solution to that, they’re just warning us. I think this is really important and something that needs to be addressed.

“The way to justify the trillions of dollars of economic investment [in AI] is widescale automation. That’s the plan. Whether or not they’re successful is up to us.”

Josh Lash: One of the things that I really appreciate about all the things we’ve been talking about today is you don’t just focus downstream of the technology, how we should regulate it once it’s out in the world, but you also look upstream at the folks building technology and you offer design standards. I really appreciate that. And we talked earlier about how new laws will ultimately influence design, but that takes time and effort. And one of the things that I worry about with those design standards is that AI products today, the way they’re designed, is totally opaque. We have no idea what’s going on inside these labs. And even the people building these products often don’t have any idea of what’s going on inside the products. There’s this whole field of mechanistic interpretability that’s dedicated to this. Given all of that, how do you enforce design standards?

Camille Carlton: I think that this is one of the big focus points of the report, the massive asymmetry between what companies know and what the public knows. And to your point, Josh, that many of the companies themselves can’t fully explain why their systems behave the way they do. And so we have that combined with competitive pressure to shorten testing cycles, release products that could still be considered risky, where we don’t actually understand the risks, and silence employees who might raise concerns. We need a much more proactive approach to AI safety and AI transparency. Instead of playing whack-a-mole with safety where we release a product, harm happens, and then we go back and say, “Okay, how do we figure out what this thing was and how do we fix it? “ It’s about demonstrating safety of products before they’re put in the stream of commerce.

And then on top of that, this fundamental principle of rebalancing the information asymmetry between companies and the public. Transparency really enables informed decision-making by the public, by policymakers, by businesses, and this creates faster feedback loops that help us see around corners with AI, anticipate harms, and mitigate them.

Sasha Fegan: These are not shocking asks. We have this transparency and safety and testing for every other high risk industry. It’s in nuclear energy, even in medicine, in aviation. Companies accept that they need to be transparent and there needs to be some external system of safety testing that they can be held to. But for AI, how do we actually get there?

Camille Carlton: Yeah. Well, to your point, Sasha, AI companies can’t grade their own homework. And this is the situation we’re in right now. We need independent oversight so that we know these products are safe before they’re released. And this is just not the case in this industry despite being the case in many other consequential industries.

Pete Furlong: Yeah. And I think when we talk about laws, it’s important that we establish clear standards for pre-deployment safety testing for these products. And these are safety standards that are rigorous and ongoing, and not something that can just be viewed as a checkbox or a rubber stamp. I think it’s important that we also have things like audits and certifications. We’ve applied these regimes to banks and financial systems, as well as just for consumer product safety. And I think really importantly, we need to protect whistleblowers at these companies and allow them to step forward when they see something that’s going wrong.

And this is another area where we’ve already seen some real momentum. We’ve seen laws passed in New York, California, and Colorado trying to address some of these aspects. We’ve also seen Senator Chuck Grassley introduce a bipartisan AI whistleblower protection bill that would provide nationwide protection for AI whistleblowers. And I think it’s also important to recognize that there’s a lot of things that we could be doing on the design side as well. But I think just for the sake of things here, we’d recommend folks turn to the report for that.

Sasha Fegan: The tricky thing is, as you were talking, I noticed the momentum that you mentioned in New York, California, and Colorado, it’s state momentum. Aren’t we getting a different patchwork of things that’s really unenforceable with companies being able to do different things in different states? How do we get that at a federal level?

Pete Furlong: Yeah, I think it’s important to recognize the benefit that both states and federal legislation provides. States can respond really quickly, and they have more visibility and responsiveness to their constituents at the state level. But the advantage is federally we can adopt something that protects citizens across the country. We need both, and it’s important that we have both approaches. But I do think it’s important, at the end of the day, that we do see some federal standards here.

Camille Carlton: Also, I want to flag for listeners that this idea of a patchwork approach has been a concept that has been really weaponized by companies, and they have used this concept to push for things like the AI moratorium and to stop any sort of progress on regulating AI companies.

Pete Furlong: And Camille, just to jump in here and remind folks, the AI moratorium essentially was a legislative package that was pushed by the technology industry this past summer. And the goal of that was to try essentially and preempt all state AI regulation with nothing else.

Camille Carlton: Right. Right. What it would have done is basically say states cannot regulate AI at all, yet we have no plan at the federal level to do so.

Sasha Fegan: And would I be right in thinking that part of the larger part of that argument with, if we do this, this will hurt that the competitiveness of AI companies, vis-a-vis China, which would be a terrible thing for American national security, economic security and so on?

Camille Carlton: Yeah. I think that this was one of the really big narratives pushed by tech companies. But if you do just a little bit of digging into it, you see that the majority of legislation being introduced at the state level is about regulating things like AI chatbots, for example. And if someone can explain to me how this AI chatbot is helping in our race China, then let’s have this conversation. But there’s a question of whether or not the type of innovation we are seeing from our leading AI companies is actually supporting American exceptionalism, American leading in R&D and science and innovation, or if we’re just seeing products being put out really without a purpose.

Josh Lash: Yeah. We’re racing, but what are we racing towards?

Pete Furlong: Yeah. And I think the goal there is that we should be racing towards safe products. That’s something that benefits all of us.

Sasha Fegan: One thing I do want to press you guys on, just before we wrap up, is what comes first, really? If you could say, give me one thing that you think we really need to change right now and that everything else, that the dominoes would line up afterwards and it would be really impactful and high intervention, what would it be? And I know they might not be the same thing. Pete, do you want to kick us off?

Pete Furlong: Sure. Yeah. I think a really important thing for me is ensuring we have clear lines of accountability. And I know it’s something we talked about at the top of the podcast here, but I truly believe that’s foundational to a lot of the change that we hope to see.

Sasha Fegan: And how about you, Camille?

Camille Carlton: I think for me it’s the opposite side. It’s ensuring that we have the rights and protections we need for people in place. It’s like, we both need to increase accountability for tech companies and then at the same time increase the protections we have, whether these are protections around labor, protections around privacy, looking at those two things hand in hand.

Pete Furlong: I’d also just add that the midterm elections are coming up, and we can expect AI to be an important aspect of this election. I think it’s worth focusing on the political influence of the technology industry, and it’s worth folks understanding where their candidates stand on these issues.

Josh Lash: We just heard Tristan and Aza talk about how what we need is a human movement, a movement that really comprises all of us, because that’s the only thing that’s going to balance the scales. And the conversation we’ve been having today is concrete, and I think people are going to really love it, but I also wonder if people are going to feel a little excluded from it if they’re not having their hands on the levers of power, if they’re not actually building the technology or passing these laws. And so I’m left with this question of, and I’m sure the audience is too, what can I do to make this happen? What can they do, our audience, especially if they’re not a policymaker or a technologist?

Camille Carlton: For me, one of the biggest things to hold for people here is that culture is upstream from politics. Because if we change our norms and we change our culture, it changes how we build products, how we design products. That is paradigm change. To me, people understanding that they have agency to shift things by changing the way we view the world is important. And then, baby steps, right?

Pete Furlong: Yeah. And we all have the ability to affect change. And we’ve seen the way folks like Megan Garcia and the Raine family have stepped up and spoken out about their experiences with harms. We’ve also seen parent advocacy groups speak up and try to push for change in terms of policy. But then we also see the impact that schools have, and teachers and folks across really all aspects of our life.

Sasha Fegan: Yeah. For me as a parent with kids in high school, we just had a meeting at a high school with the Parents and Citizens Association about the use of AI at school. It’s also stepping up and trying to have a shaping role and bring some of this knowledge into those discussions at a local level, at a municipal level. Because the more that happens, the more we are actually driving that cultural and norm shift. You could be the voice in your family who really brings these conversations to the dinner table, and be the go-to person in your network who understands these harms and can advise people in your network around how they can use AI safely, and also where the line between what their individual responsibilities should be and where we need to actually pressure our legislators to take federal or state responsibility. And we need that help to externally enforce standards and safety measures.

Josh Lash: I think, ultimately, like you said Pete, this is going to touch every aspect of our lives. We all have a part to play in this. You can, at work, talk to your HR person about the AI that you’re implementing in your systems and ask about, what are the safety standards that you’re applying there? What are the privacy standards that you’re applying there? Or you can go to a town hall and you could say, “Hey, I’m really worried about what AI is going to do to my job,” and see what they have to say about that. And I’m reminded of the quote that Tristan often uses in these podcasts, and it’s a quote I’ve always loved, which is the Margaret Mead quote, “Never doubt that a small group of thoughtful, committed citizens can change the world. Indeed, it’s the only thing that ever has.” And it’s true. It’s only going to come from us and we have to step up and do it.

Camille Carlton: And I think what I would also offer to listeners is we have really seen the power of individual action with social media. We have seen parents marching on Washington. We have seen people putting their phone on grayscale. We have seen people take action, and it took a long time to get there. But where we are with AI is people understand the harms way faster than they did with social media. And so we’re at that point of we’re ready. It’s the time and place for people to come forward. And that same trajectory of change that we’ve seen from social media can happen with AI as well.

Josh Lash: We just covered a ton, and that’s only four of the seven principles in the report. I really encourage people to go read the whole thing, there’s a lot more detail in there, but it’s very readable. Pete, Camille, thank you both so much for coming on today. A lot of food for thought, and I’m really excited to get this out into the world.

Camille Carlton: Thanks for having us.

Pete Furlong: Yeah, thank you so much.

RECOMMENDED MEDIA

The AI Roadmap

The Human Movement

RECOMMENDED YUA EPISODES

AI Is Moving Fast. We Need Laws that Will Too.

A Conversation with the Team Behind “The AI Doc”

The Narrow Path: Sam Hammond on AI, Institutions, and the Fragile Future