Mark Zuckerberg on AI's Job Impact, Government Regulation, & How Meta Glasses Changed His Routine

Yeah. Well, this is a big day for us. So, thank you for doing this.

We are very excited for the announcements today. The new line of Meta glasses — it's the first line of glasses that we've designed with our partner EssilorLuxottica from the ground up. So this builds on the work that we've been doing with Ray-Ban and Oakley and iconic brands, and it basically brings all of that functionality to a new line that we hope will be accessible to a lot more people. It brings all of the functionality including the most advanced AI models that we've built right into the glasses. So we're very happy to be doing this.

There's a few things overall that make us optimistic that glasses are the next computing platform. One is that computing gets to be more natural and more integrated in our lives over time, and glasses give you the opportunity to interact with technology but remain present with the people around you, which is a fundamentally important thing that phones don't do. When you interact with your phone, you're interacting with this small rectangle that pulls you away from the world around you. You want technology to be able to effectively fully integrate with the world around you. Another piece is that glasses are the ideal form factor for an AI assistant — to see what you see, hear what you hear, talk to you throughout the day, and eventually display information in your field of view and overlay things on the world around you. You can get parts of that with other form factors — earbuds can hear and talk to you — but the visual part is very important, and glasses are the ideal form factor for that. One more thing is that glasses are already a thing that people have. Almost two billion people in the world wear optical glasses already and billions more wear sunglasses. I think we're at this moment where it's like everyone had flip phones and then smartphones came out and it felt pretty clear that in five years all flip phones were going to be smartphones. That's basically how I feel about glasses today. What is the reason why five years from now any of the glasses that people wear are not going to have this kind of functionality? It just doesn't make sense. I think it's a very natural design that people are used to.

We've been working on this since 2014. The earliest part of our work was on virtual reality and augmented reality. The notion is that virtual reality allows you to do some augmentation of the world in a somewhat larger form factor before you get into the one that people are going to wear every day. But some of the glasses we've been working on — it's more than ten years at this point. The partnership with EssilorLuxottica — we've done multiple generations starting with the original Ray-Bans, and each year we've added more functionality. Now you can do phone calls, listen to music. It has functionality to help people who can't hear well — it can amplify the hearing in their ears. You just add generation over generation. The AI keeps getting better. We're going to be working on this for decades more, but I think the next five years are going to be a really key period for this.

The main thing I notice is that there are a lot of things you could do on your phone, but if you have glasses on, it's just better. Being able to take phone calls with your ear open — you're just walking down the street and you can hear what's going on. The microphone is in the nose pad, so the clarity of the audio is so high that you could literally be in a wind tunnel and it would sound completely clear to the person on the other side. I've taken business calls on a jet ski — not often, but it has happened, and the other person could not tell. It's great. You don't necessarily want to tell the other person you're on a jet ski. But the main way this is going to work is that it's not like all the other computing platforms go away. When we started using phones, our computers didn't go away. I'm often sitting at my desk with my computer right there, but I still reach for my phone to send a message, because the phone has become the primary platform. Similarly, glasses are going to be that. As you can do more with glasses, you'll have your phone but it'll stay in your pocket more of the time, and maybe eventually you'll leave it at home more. But I don't think we're going to stop having phones in the next ten years. I just think glasses are going to each generation become more the primary thing we use and be more integrated into our lives.

Totally. One of the neat things is that if you get the hardware package right, hardware is kind of slow — we've been planning this generation for two years, and we're planning the 2028 line now. Then you get all these releases in between that are basically just the AI getting better and better. So if you buy these glasses today, they're just going to keep getting more useful as the AI becomes more powerful. The biggest advances in the last couple of years — there's multimodal understanding: being able to not just ask it a question but have it see what's going on around you and talk to you about that. Language translation at this point is almost solved — real-time translation built in. But one of the things I'm really excited about is the advance in agentic AI. Moving away from the pattern where you ask one question and get one answer — what I think this evolves towards is either you give it a task and it goes away and thinks and comes back, or you have live sessions where it's just watching. Like, I'm baking something with my daughter — I'm terrible at this — and it's like, 'No, really, don't do that, do this instead.' I didn't even have to ask it. It's watching. It knows. Do not put the flour there. As you tune the system, you can go from half an hour of always-on AI to an hour to two hours to eventually all day. That's a very different mode of using AI. You effectively have the ability for your AI to see what you see, hear what you hear, summarize that, and help you throughout the day. It's amazing how much signal we have throughout our day that we just miss or tune out because we're focused on something else. The ability to have that kind of memory and help in all those ways is going to be incredibly important.

Our mantra for designing the glasses is that they have to be great glasses first. The technology is secondary — the fashion and wearing it is actually primary. That was an interesting thing for us to get to as a technology company, because most technology companies think about it as cramming all this technology into a shape that looks kind of like glasses. But the reality is you're going to wear the glasses for many hours and use the technology for some small fraction of that. In order to unlock that, the glasses need to be something you're proud to wear and comfortable so you'll wear them for long periods. They have to be great glasses first. Part of that also means getting out of the way. A lot of technology companies have this vision of, 'We can help you here and here and here — let's bombard you with stuff.' People don't want that. I don't want that. You actually want it to be very sparing in terms of when it will ping you or interrupt you. That's the way we think about it: great glasses first, technology available when you want it with more and more capabilities, but we really don't want it to be intrusive.

Our vision for what we're trying to build on the AI side we call personal super intelligence. The idea is very different from the consensus view of most other AI labs. Most other companies think there's going to be one big AI that everyone uses, or that it's doing stuff for the world. Maybe this is the legacy of building social media — it's all about building tools, putting them in people's hands, believing that people should have agency to use the tools and express the things they want to. There's fundamentally this view that in order to have a good future with strong AI, you want a lot of checks and balances. I don't want to live in a future where there's one big AI — I think that's a bad future, no matter how good the AI is. The way this ends up being good is you direct technology development towards empowering every individual. People have different aspirations that sometimes overlap and sometimes compete, and that competition is important — it's society and markets' way of putting checks and balances on how things work. In a world where everyone has a personal super intelligence with the context they want about their life, helping them achieve their goals 24/7, and as many people around the world as possible have that — to me, that's the best shot we have at a future that is really good. People are empowered, but there's also a fundamental check and balance of power. That's a very fundamental thing for our philosophy and everything we're developing — from the AI software and models to the hardware and glasses.

I'm not as familiar with that book, but the concept of AI safety is obviously a very important one. I'm a little more optimistic than that — I don't think if anyone builds it, everyone will die. But I do think that any powerful new technology gives people the ability to do good things and bad things, and the more powerful the lever, the more good and harm people can potentially do. Taking that seriously and working to mitigate harm while making it as positive as possible is incredibly important. That's where you get to this philosophy: what is the best way to ensure we have a positive future if the technology development continues on its trajectory? Some people would say build one singular AI and make it as aligned as possible. I just don't know that's going to be enough. If you have one thing that is super powerful — there have been a lot of people and regimes that get absolute power over time, and that usually doesn't end well, no matter how enlightened the thing is. I think humans are limited in our ability to build enlightenment into a thing no matter how hard we try. There is a natural mechanism that humanity and civilization has, which are checks and balances. My version of creating a positive future is not having one central thing, but having it widely distributed. It's somewhat controversial — some people would say if you give a model with a lot of capabilities to individuals, won't they do bad things? But this has always been a historical tension between liberty and empowering individuals and how you build society. At every step along the way, we tend to benefit when we give people more freedom and more agency rather than saying, 'This is too much for a person to handle.' I think that's the path to a better future. I'm pretty optimistic if we do that, but different people are trying to build different things, so I hope our vision is the one that gets built.

The ones most people are focused on right now — the huge ability for AI to code, which is helpful because there's a lot of coding you want it to be able to do even if you're not an engineer. I put cameras up in my MMA gym and asked it to watch me and give me feedback between rounds — that's actually a coding problem, dealing with the video pipeline. The better it gets at coding, the more cybersecurity threats there are. That's obviously one of the big topics. One of the things on the horizon I think we're going to start seeing more of is scientific advances. My philanthropic work outside of Meta — Priscilla and I created this thing, Biohub, where our goal is to help advance the pace of scientific research to help scientists cure and prevent all diseases this century. When we started, we thought that was really ambitious. Now it's probably too conservative. I don't know if it'll be a decade or two decades, but I would guess it'll be much sooner than the end of the century. A lot of that is because of the progress of building AI models that can effectively simulate biology and help with drug design and understanding fundamental parts of biology. But the concern is — could someone bad use that to make things that hurt people, or create a bioweapon? That's still an open question, but it's clearly something that needs to be thought through carefully. Those are probably the two things in the next year or two where there's a path to unlock huge value and also a clear roadmap to mitigate concerns, but also pretty big concerns you want to be careful about. Beyond that, you start getting into broader societal questions around the balance of power. People talk about job displacement — companies are always trying to do things more efficiently. But people also improve their own productivity. There's a natural balance. If companies get very good at automating faster than individuals get better at productivity, the number of jobs might decrease. But people assume that's an inevitability — I don't actually think it is. If you focus on empowering people and making them more productive at a faster rate than companies automate things, there should be more jobs in the future, not less. You need this balance. If just a few companies focus on automating all knowledge work, that's probably not a great future. But if there's a balance where some companies focus on efficiency and others focus on this personal super intelligence vision of empowering individuals and making people more productive...