Half of America Dislikes Him. Zuckerberg Still Wants Presence, Not Screens

Lex Fridman opens with a number. Fifty-four percent of Americans view Mark Zuckerberg unfavorably; Elon Musk sits at 23 percent. Both men run companies that have changed how billions of people live, yet the public scorecards could hardly look more different. Fridman recorded the conversation on February 26, 2022, as war broke out in Ukraine, and he dedicated the episode to people suffering in that war. His grandfather, he says, had been a machine gunner in the Soviet Red Army, holding the line against the Nazi invasion on Ukrainian soil.

The people who run social networks, Fridman argues, hold a responsibility almost unlike any other. When those networks fail they can cause immense suffering; when they succeed they can lessen it. He wants an honest conversation that does not just ride the viral wave of cynicism. Zuckerberg later jokes that five or six hours is still on the short end of the congressional testimonies he has given. Between those poles they talk about presence in virtual reality, Instagram and teenage girls, free speech after a pandemic and an election, and what a four-year-old calls "code art."

Fridman is a stickler for recording in person. He asks when a conversation in the metaverse could have the same chemistry as two people in a room. Zuckerberg treats that as the central product question, not a sci-fi aside. Everything that follows in the interview is downstream of it: identity, security, mental health, and the brand of a man half the country already distrusts.

What makes virtual and augmented reality different from every earlier digital platform, Zuckerberg says, is presence: the feeling that you are in a place with other people, not looking at a screen. Phones and TVs can be high fidelity, but every subtle cue still tells you that you are watching glass. Spatial audio is one of the first cues that breaks that spell. Meta launched Workrooms for meetings, and Zuckerberg made a rule for senior management: standing meetings happen in Workrooms. They have to dogfood it. Even with crude avatars, he says, the experience already beats a typical Zoom call. A voice from across the table sounds like it is coming from across the table. You can lean over and whisper to the person next to you, which video software still cannot do.

Hands were the next surprise. Early headsets let you look around but not interact. Once hands appeared, the team assumed users would need to see a full arm; looking down at floating hands would feel wrong. The opposite happened. The system has to interpolate an elbow angle from the headset and the hand. When that angle is off, people feel a mismatch that breaks presence harder than missing limbs. Showing only the hands turned out to be fine. Zuckerberg, who studied psychology and computer science, treats that as a design lesson rather than a graphics one. Most big tech companies, he says, build technology for you to interact with. He wants technology that helps people interact with each other.

A large part of the human brain is wired to read faces. That is why the next Quest generation is being designed around face tracking and eye tracking. Video conferencing has gone astonishingly far without ever solving eye contact. You look at someone's eyes on a screen and they do not feel looked at; you look at the camera so they feel watched, but it is not a natural glance. VR can fake it from head pose today. With real eye tracking, Zuckerberg says, you can do it for real. He knows the hardware tradeoff. Those sensors add weight and thickness, and another company could race to the lightest, thinnest headset. He wants the most human one. A millimeter of eyebrow movement means something. Losing that signal, he argues, is a loss.

Fridman asks for a singularity moment: when will most people have most of their meaningful experiences in the metaverse rather than in the physical world? Zuckerberg cites a definition he had read. A lot of people think the metaverse is a place. One better frame is that it is a time, when immersive digital worlds become the primary way we live and spend our hours. He thinks that is a reasonable construct, and he adds that he and Fridman already live a large part of life in digital worlds. They are just not 3D and immersive yet. Meetings happen on video. A lot of work happens on email and WhatsApp.

Building a platform, he says, is a duality. You want elegant systems that can scale to billions of people, and you also have to fight a ground game of specific use cases. Meta went after gaming and social first on Quest. Looking back at PCs and phones, gaming was huge but fragmented and hits-based; the companies that focused on communication and social interaction were fewer, and they mattered as much as all the games combined. Productivity is the third case, and historically not Meta's strength. He splits it in two: Workrooms-style meetings, and Word, Excel, coding, knowledge work. He does not think the crossover is more than five years off.

The threshold he offers is almost boring. If VR is clearly better than Zoom for meetings, or if an engineer can code in a virtual workstation that is even 5 percent more productive, Meta would buy the devices for every engineer, and other companies would follow. That is how you rival today's computing platforms. Inside the Lab, Meta showed a Builder Bot that plants trees or sets water bottles on a picnic blanket when you ask. Those first-time wonders matter. Most of life, Zuckerberg says, is not first times. The goal is not to make people compute more. It is to make computing more natural, so everyday work migrates because it is a little better, every day.

When Fridman reads the 54 percent figure, Zuckerberg first asks whether it is a US survey. It is. He says Meta's brand has been uniquely challenged in the United States compared with much of the rest of the world, where sentiment toward the services is often extremely positive. Before 2016, he says, there were probably very few months when coverage of the company was net negative. Since 2016, there have probably been very few months when it was net positive. Polarization is part of it. The structural problem is more specific.

Meta is not a Democratic company or a Republican company. Every day it has to arbitrate content decisions, and those decisions reliably upset about half the country. The people who agree do not think Meta got it right; they think Meta did not mess it up. The people who disagree ask how the company could possibly have done that. From a brand perspective the moves are negative-sum. Each controversy knocks the company down with whoever lost that round. If he wanted people to think positively about him as a person, Zuckerberg says, he is not sure he would have founded a company, and certainly not a social media company. Other internet founders have stepped back into less controversial work. Running something at this scale, he says, is something you only understand by doing it for a long time.

Fridman asks whether anyone in the inner circle will call him on his nonsense. Zuckerberg is proud of a famously open culture that encourages dissent, which is also why so much internal material leaks. The management team, mixing veterans and newer people, is relatively confrontational and high-trust. Friends and family push him too. He says that is not enough. You also need journalists, analysts, peer executives, and certain politicians whose different view of the world is exactly why their view is valuable. He splits critics in two. Good-faith critics are trying to help you do better; those voices should be cherished and heard closely. Tuning out all negativity is about as dangerous as trying to win over people who will never like you. If he judges someone is operating in bad faith, over time it just does not bother him that much.

Fridman puts the Frances Haugen case in its strongest form. The former Facebook employee leaked internal Instagram research on teenagers and wellbeing and argued that Instagram was choosing profit over the mental health of teenage girls, calling the product toxic for them. Zuckerberg's response essay was titled "What Our Research Really Says About Teen Wellbeing." It said that on 11 of 12 wellbeing issues, teenage girls who were already struggling also said Instagram made them better rather than worse. Fridman asks him to steel-man Haugen's characterization, then explain his own.

Zuckerberg does not minimize the subject. As a parent, he says, it is hard to imagine a more important set of questions. The internet and social media are not unique here: magazines have been selling body types to women and children for decades. The company studied the problem because it cared, and it never expected every finding to be flattering. You study in order to improve. Where he disagrees is the coverage. He recalls about 20 metrics; on 18 or 19, using Instagram was neutral or positive for teens' wellbeing. One area showed they needed to improve, and they took steps after the research. An accurate characterization, in his view, would have been that teen use is generally positive for mental health. That is not the narrative that came out.

The intent critique bothers him more. It assumes malevolence: you did the research and then tried to sweep it under the rug. As far as he knows, it is not clear that any of the other tech companies are doing this kind of research at all. Why, then, should the story be that Meta studied an issue in order to understand and improve it, took steps afterward, and was covering something up? He finds that description beyond credibility compared with what the company actually did. These issues will occupy the industry for a long time, he says, and he would encourage everyone else to study them too. Fridman notes that the press obviously gets more clicks for negative stories about social media, which does not help an open conversation about the positives and the negatives.

Fridman presses him as a father. Because so much of life has moved into the digital world, bullying has moved with it, and some of that bullying leads to depression and suicide. Zuckerberg says the company systematically tracks a set of harms, and bullying and self-harm are among the biggest. You will not get rid of all of it. He has two daughters who fight and push each other around; the question is how you handle the situation. You give people tools, you fight it yourself, and you make sure people can get help.

The technical hard part is bullying. It is highly context-specific. Promoting a terrorist group is comparatively easy for a classifier: the content tends to look and feel a certain way. Bullying can be a subtle comment about someone's appearance that is idiosyncratic to them, and it can look like humor. Humor to one person can be destructive to another. So part of the answer is AI, and part is agency. Meta has made it possible to turn off comments or take a break from one person without publicly unfollowing them or announcing that the friendship is over. You can stop hearing someone without turning it into a public stand.

Self-harm and suicide are a different case. Zuckerberg says AI can identify a lot of those signals, along with user reports, in hundreds of languages. The piece he is proudest of is not the model. The systems connect to local first responders. At this point, he says, they have been able to get help to people in thousands of cases because of the plumbing between the AI and local emergency organizations, and they are rolling that out in more places. It is a heavy topic. Some of this, he adds, is just human nature: people hurt each other, and you can still give them tools. Fridman puts the bind in one line. A platform that lets people fall in love is, by nature, a platform that lets people hurt each other. Managing that is psychologically brutal.

Two groups pressure Meta on speech. One is angry that Facebook lets "misinformation" spread. The other worries that Facebook labels ordinary speech as misinformation and censors it. Zuckerberg gave a well-known defense of free expression at Georgetown University in October 2019. Then came COVID and the 2020 election. Fridman asks whether advertisers, politicians, and the public have forced the company to damage the ideal he defended. Zuckerberg says advertiser and politician pressure does not directly change how they think about it. These are just hard topics. You do not build a company like this unless you believe that people expressing themselves is a good thing. The formula he repeats is voice plus connection equals opportunity. Most of what people express is not political. It is identity, closer to the avatar conversation than to a campaign ad. It becomes political when people feel strongly enough.

He walks the history. In the early years, before deep learning, it was Zuckerberg and his roommate Dustin looking at reports and doing their best. For the first seven or eight years the company was not that big; for a lot of that period it was not even profitable, and the AI to scan everything did not exist. Mid-decade that flipped. Image classifiers could find pornography. Terrorist content got easier. Incitement to violence is much harder: you have to tell denunciation from encouragement, in a hundred languages and in dialects, without a pile of false positives. After 2016, he says, a lot of people woke up to how polarized the country was and asked why internet companies had not been on top of it. He put a thousand or more engineers on AI systems that could proactively find harmful content. Every quarter they publish transparency reports on about 20 types of harm, tracking how much is taken down before anyone reports it.

Then the other side showed up. Some of what the AI removed was borderline, bothering no one, never reported. Why take it down? The other camp still says the company does not do enough. As the technical capacity arrived, the interesting questions became philosophical: where exactly do you want the line? Zuckerberg concluded that one person should not make those decisions. Meta built an independent Oversight Board, people from around the world committed to free expression and human rights, to whom users can appeal. The most controversial cases go to them for a final, binding decision. The company then tries to extract the principles behind those rulings and encode them in the algorithms. The first chairs, he says, were chosen for a commitment to free expression and a broad understanding of the tradeoffs around it. They are people who would lean toward speech.

Has his stance changed since Georgetown? "No," Zuckerberg says. Philosophically it is the same. Everyone agrees there is a line: if you are about to do physical harm to people, there should be restrictions. The classic Supreme Court example is that you cannot yell fire in a crowded theater. What everyone disagrees on is the definition of real harm. Bring it back to bullying. Is it harm only when the target is about to hurt themselves, or is the bullying itself the harm, and at what point on that spectrum do you draw the line? There is broader agreement, he thinks, that an acute threat is a reason for a society to tolerate less speech. Nobody expected COVID to last for years. Asked a priori whether a global pandemic in which a lot of people were dying counts as the kind of emergency where yelling fire is a problem, he thinks it probably passes that test. The situation was tricky. The fundamental commitment to free expression, he says, is still there.

Fridman pushes on the institutions. A lot of the criticism is that the CDC, the WHO, and other bodies civilization once trusted to draw health-policy lines failed in small ways and big ones, depending on whom you ask. So where is Meta supposed to get the answer to what counts as misinformation? He mentions his friend Joe Rogan: in Las Vegas someone yelled at Rogan to stop spreading misinformation, while a large audience believes he is not spreading it at all. You cannot pretend that a large number of people do not have a different definition. Who is an expert? Fridman, who has a PhD, says he is not sure he knows, especially in a brand-new pandemic, with politics in the mix and a media business that can make money on outrageous narratives. Everybody turns to Facebook and asks it to name the source of truth.

If he were in Zuckerberg's chair, Fridman says, he would lean toward speech and refuse to be the one who defines misinformation. Let it be a public debate and let ideas stand. He would put more of the burden on scientists' communication: get into the battlefield of ideas, show the data, talk to people spreading anti-vaccine claims, instead of rolling your eyes and declaring expertise. The way to defeat bad ideas is with good ideas, with speech. Some of the best COVID explainers he saw were doctors on social media, moving faster than peer review. The platform should step back from the fight over who defines falsehood, keep intervening on acute, clearly harmful content, and be more transparent about how hard the rest is. Hand any critic the helm for a week, he says, and see how well they do.

Zuckerberg had just announced the AI Research Supercluster, built on NVIDIA DGX A100 systems: 6,000 GPUs now, heading toward 16,000. Fridman, who loves GPUs, wants to talk about translation. He speaks Russian fluently and English somewhat fluently; he imagines a conversation with Vladimir Putin, who has some English and fluent Russian. If an AI could outperform both of their second languages, that would be a clean test of whether the chemistry of a conversation can survive the bridge. Zuckerberg says translation has gone through several technical generations. The state of the art is not to pass through an intermediate symbolic representation. You map concepts directly from one language to another, and you train bigger and bigger models to do it. That is what the supercluster is for. Training eats the compute; running the translator later does not.

They can already go between about a hundred languages seamlessly, with a near-term goal of about 300, any language to any other. The farther you go toward smaller languages, the less data you have, so you have to transfer from models built on richer ones. How far that generalizes is one of the big open questions in AI. For speech, the old pipeline was speech to text, text to text, text to speech. The better path is speech to speech, skipping the transcript. Self-supervised learning, a topic Fridman had discussed with Yann LeCun, plus much larger training runs, has made a lot of previously hard problems newly practical.

He then ties translation to the next device. Future glasses should look like ordinary glasses and overlay holograms so virtual and physical objects share a scene. That would be the first computing device that has as many environmental signals as you do: it can see what you see, from your perspective, and hear what you hear, because the microphones sit by your ears. The AI assistant that lives there is a first-person assistant, and the foundation models it needs will be different from anything on a phone. In his view the metaverse and AI enable each other. You can talk a Builder Bot into placing trees. You can also imagine an assistant that is as intimate as a scheduler or as close as therapy. The hardware and the models have to grow up together.

Identity in the metaverse, Zuckerberg says, will not be one thing. On one axis you go from cartoonish, expressive avatars to photorealistic ones. On another you go from representing yourself to inhabiting a fantasy. The combinations multiply. A more realistic soldier face can help a military simulator. Hanging out with friends, a cartoon version of you is enough if they still know it is you. VRChat already shows the joy of showing up as a meme, a built-in icebreaker. A work meeting wants a photorealistic version of your actual face. Early Facebook, he thought people would have one identity. He is now sure that will not be the case. The interesting systems problem is letting people move between those identities without friction.

He thinks the market for virtual clothing will be large. If you spend your workdays in photorealistic meetings, why would you not spend almost as much on how that avatar looks as you spend on what you wear on a video call? The harder question is whether a shirt you buy for the photorealistic you can be translated onto the cartoon you, or even onto a dragon. He thinks the answer lives in style transfer. A designer might sell not a single garment but a generator, a closet of a style, that can dress any body you inhabit. Digital goods become more valuable when they are not locked to one platform or one use case, which is the core promise he sees underneath the NFT jokes. When goods travel, people spend, and the economy grows.

Fridman asks about the dark side. If identity can be hidden, people can troll, and bots can pretend to be people. How do you know you are talking to a well-intentioned human? Zuckerberg splits the cases. Knowing someone's intentions is unsolved almost everywhere. A dragon is obviously not impersonating a specific person. The scary case is impersonation. In a future Workrooms meeting, someone walks in who looks exactly like you. How do you know it is you? One direction they are thinking about is biometric security: before you use that photorealistic avatar, you prove you are its owner. Fake accounts doing harm is not unique to the metaverse, but presence makes it more painful. His preferred move is to judge accounts by overall behavior rather than policing each utterance. Metaverse interaction, he says, is more like a phone call than a post. Between counter-intelligence teams and years of integrity AI, he thinks Meta is years ahead of the rest of the industry, if still imperfect. Fridman jokes about open-sourcing his avatar so millions of Lexes can walk around like Agent Smith. Zuckerberg tells him that will not help anyone know which one is real.

Fridman had just had his QNAP NAS locked by ransomware, despite doing many of the right security things, and he asks how to think about protecting personal data. Zuckerberg splits the world into fortresses and decentralization. Both have strengths. WhatsApp, he says, is the largest encrypted messaging service in the world, and Meta was among the first to ship multi-platform end-to-end encryption. The promise is that the company cannot see the content of your message; someone who hacks Meta's servers still does not get what you sent a friend. The cost is on the device. If you lose the phone, the content is gone. If you back it up to iCloud, you are at Apple's mercy on whether a government gets it or whether Apple gets hacked. Centralized systems also let you train better personalization. A serious company, he says, runs red teams, pays bounties, and invites the best attackers in the world to break in. Open source got more secure over time because problems surface instead of being hidden. Nation-state attackers keep getting smarter; people posting hate usually do not. After a bathroom break, Fridman turns to *The Social Dilemma*.

The documentary's core claim is that because social media is funded by ads, algorithms maximize attention, and a reliable way to do that is to make people angry at each other. Zuckerberg wrote a point-by-point essay, "What the Social Dilemma Gets Wrong." He steels the worry. Yes, the company generally makes more money when people use the service more, so there is an incentive to be useful enough that people return. The jump he rejects is that agitation is what keeps them. Television has the old line that if it bleeds, it leads. What he has found after 17 years is that what grabs you this week is not what you appreciate having seen, and he is still relatively young, with decades of work ahead. A company playing the long term is incentivized to build what people find valuable over time, not what captures them today. He is also a product designer. Presence in VR is a feeling, not a feature list. He does not want to build products that make people angry. Users can mark a piece of content as something they like, love, or that makes them angry or sad, and the system shows angry content less. That mechanism is a statement of intent.

No business model is clean. Subscription news has its own trap: a partisan outlet only gets one side to pay, so the incentive is not to produce centrist work. Advertising's virtue, if you believe everyone should have a voice, is that the consumer service can be free instead of a luxury. Fridman still sees fights, outrage, and viral pile-ons, and asks how to push back on the story that platforms let anger spread because it works. Zuckerberg wants the conversation grounded in research. Social media exists in almost every country, Facebook nearly everywhere except China and North Korea, yet polarization is falling in some places, flat in others, and rising sharply in the United States. If social media were the driver, that map would not look like that. He cites Stanford economist Matthew Gentzkow and Ezra Klein's *Why We're Polarized*, which traces US partisan forces, including Fox News, that predate the internet. Studies he has seen often find social media use correlated with *lower* polarization. Gentzkow's work after 2016 found the most polarized voters were the ones not on the internet. Some European work finds that people who stop using social media get more polarized. The harm researchers worry about is affective polarization, negative feeling toward another group, often measured by whether you would let your child marry someone from it. In general, he says, social media use has corresponded with decreases in that. The narrative is still compelling to a lot of people. He thinks the mainstream version is just not right.

Asked for advice to young people who want a life they can be proud of, Zuckerberg starts with bedtime. Every night he puts his daughters, Max and August, through a ritual they call the good-night things. First is health: take care of yourself. Second is loving friends and family, and making time for those relationships, which he thinks may be the most important. Third is having something you are excited about in the future. With a four-year-old that can be tomorrow morning with Mom; for him it might be building AR glasses or a tenth wedding anniversary. The world is heavy, and people get through it by looking forward. Last, every day, he asks what they did to help someone: set the table, explain a problem to a classmate. For college, the practical version is who you surround yourself with, because you become like them. His hiring heuristic at Meta is that he will only hire someone if he could see himself working for them in another universe. He likes his job and does not want them to run the company. He wants to know he would be happy to work for them. People, he thinks, are too objective-focused and not focused enough on the people they are becoming like.

What world does he want his daughters to grow up in after he is gone? He quotes a line attributed to Picasso: all children are artists, and the challenge is remaining one when you grow up. Kids have wonderful imaginations. One promise of the creator economy and the metaverse is that more people will make a living from creative work rather than traditional labor or service. He teaches the girls Scratch. For him, coding was building, a utilitarian way to make something exist. His youngest is musical and artistic; she thinks of coding as art and calls it code art, caring about how the thing moves on the screen. For his generation programming was a tool. For the next, he hopes it is an expression of imagination. On mortality, the studio lights happen to go out as they start talking about death. He does extreme sports, including fifteen-foot waves, with safety gear and local knowledge, and he is aware of the risk. Finiteness does not freak him out. "Constraints are helpful." Finiteness makes ice cream taste more delicious. The metaverse may offer nearly infinite expression and worlds. Time will still be finite. Living longer is good. With his wife Priscilla Chan he founded the Chan Zuckerberg Initiative and pledged 99 percent of his Facebook, now Meta, shares. Their philanthropy is not longevity research. The goal he thinks is possible this century is to cure, prevent, or manage all disease. No disease at all in a few decades is probably not possible. Those three verbs might be.

If he died and could ask God one question, he would want to know that Priscilla and the family would be okay. That may depend on the circumstances of the death, but in most cases he can imagine, it is the main thing he would care about. He still struggles with whether running so large an enterprise should outrank the people closest to him. Before he had a family, the company was the only thing that mattered. It still matters deeply; the answer is less obvious. On the meaning of life he offers two ways in. Intellectually, because he built a company around human connection, he would put connection first, and he thinks society systematically undervalues it. Growing up, the rule was do your homework, then play with friends. What if playing with friends is the point? The other way is religious. Studying Genesis with a rabbi, he was struck that the Bible starts with creation: God created people in his image, and created everything. A lot of what religion has been telling people to do, he finds, is to create and build. The two answers are not at odds. He studied psychology and computer science. For him the fusion is to create and to love, love being the highest form of connection. Fridman closes by quoting the end of Kipling's "If."