What this covers
Mark Zuckerberg hosts Yuval Noah Harari for a frank conversation about some big challenges -- as part of the Facebook CEO's 2019 series of public discussions about the future of technology in society. The overarching question they debate is: what are we going to do about the systemic problems of the current technological revolution?
To stay up to date on new videos from Yuval, subscribe to our channel and click the bell icon 🔔 https://www.youtube.com/c/YuvalNoahHa...
Timestamps 00:00:00 Intro 00:01:30 Is Technology Fragmenting our World? 00:08:00 The Role of Online Communities 00:11:00 Facebook & What is the Goal? 00:14:00 Creating Greater Global Cooperation & Securities 00:19:47 Localization, Globalization, & Extremism 00:25:25 Is Globalization Always Positive? 00:29:11 The Free Market and AI 00:34:09 How Nationalism Affects Data Privacy 00:41:40 An America First Approach to Data 00:48:40 Building AI Protection Systems that Actually Protect Us 54:00 Data Centralization, Decentralization, & the Cold War 01:04:30 Hijacking Human Attention 01:14:00 Is the Customer Still Always Right? 01:20:39 Correcting for Extremism 01:25:00 Totalitarianism & Surveillance Capitalism
Yuval Noah Harari speaks internationally and teaches at the Hebrew University of Jerusalem. On this channel you can see his interviews, lectures, and public conversations with prominent leaders and influencers, — including Mark Zuckerberg, Natalie Portman, Christine Lagarde, Chancellor Kurz of Austria, Jay Shetty, and Russell Brand.
// For more info on Yuval’s articles, books, upcoming events, and featured podcasts visit Website: https://www.ynharari.com/ Events: https://www.ynharari.com/events-page/ Media: https://www.ynharari.com/media/ Books: https://www.ynharari.com/book/unstopp...
// Socials Twitter: https://twitter.com/harari_yuval Instagram: https://www.instagram.com/yuval_noah_... Facebook: https://www.facebook.com/Prof.Yuval.N... Youtube: https://www.youtube.com/c/YuvalNoahHa...
Source description (no synthesized summary yet).
Harari and Zuckerberg debate whether global technological connectivity enables human flourishing or facilitates unprecedented forms of control: connectivity without cohesion risks fragmentation, while AI and data systems could shift power irreversibly toward totalitarian regimes unless democratic values are encoded into the technological infrastructure.
- Global connection requires both local economic/social cohesion and open policy frameworks; without both, people seek extremist alternatives or withdraw political support for cooperation
- AI's ability to process centralized data more efficiently than distributed systems may structurally advantage totalitarianism over democracy for the first time in history
- The choice of which countries develop AI and which values they encode will determine whether technology serves human agency or enables surveillance capitalism and totalitarian control
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Humanity has become more connected over thousands of years, from isolated worlds to a single integrated historical, economic, and cultural unit, but connectivity does not automatically produce harmony—it can intensify conflicts among those closest to us.
“if you look at the long span of history, then it's obvious that humanity's becoming more and more connected. If, thousands of years ago, planet Earth was actually a galaxy of a lot of isolated worlds with almost no connection between them, so, gradually, people came together and became more and more connected until we reached today when the entire world, for the first time, is a single historical, economic and cultural unit.”
Wealthy societies with stable local conditions and good personal lives have historically engaged in colonialism, imperialism, and other expansionist ideologies, suggesting that economic distress is not the only or primary driver of harmful political movements, and that affluent people with good family and social lives can support extreme ideologies.
“People that feel good about themselves have done some of the most terrible things in human history... the Nazi voters were also nice people... they didn't think so... if you just think about colonialism... people say in Britain, in the late 19th century, they had the best life in the world at the time. And they didn't suffer from an economic crisis or disintegration of society or anything like that. And they thought that by going all over the world and conquering and changing societies in India, in Africa, in Australia, they were bringing lots of good to the world.”
The concept of a life as a series of meaningful decisions—from Shakespeare tragedies to Hollywood narratives to religious texts—has been central to human meaning-making and culture, and outsourcing these decisions to algorithms represents a fundamental shift in the nature of human existence.
“for thousands of years, humans tended to view life as a drama of decision-making... almost all of art and all of religion is all about that. Whether it's a Shakespeare tragedy, or a Hollywood comedy, it's about the hero or heroine needing to make a big decision. To be or not to be. To marry 'X' or to marry 'Y.' And what does it mean to live in a world in which increasingly, we rely on the recommendations of algorithms to make these decisions”
Globalization has been net positive at the global level, bringing hundreds of millions out of poverty, but 5-10% of the world's population—hundreds of millions of people, mostly in developed countries—are worse off in absolute terms, creating political pressure that undermines support for global cooperation systems.
“globalization for the whole world is negative. I think, in general, it's been, on balance, positive. But the story we've told about it has probably been too optimistic... if there are five or 10% of people in the world who are worse off, seven billion people in the world, so that's many hundreds of millions of people, the majority of whom are likely in the most developed countries, in the US and across Europe, that's going to create a lot of political pressure in those countries.”
The foundational assumption of Western democracy for generations has been that the customer is always right and the voter knows best, but in a world where technology can hack human beings and manipulate at unprecedented scale, this assumption may no longer hold.
“the customer is always right. The voter knows best. People make a choice. If they choose to do it, then it's good. That has been the bedrock of, at least, western democracies for centuries, for generations. And this is now where the big question mark is. Is it still true, in a world where we have the technology to hack human beings and manipulate them like never before that the customer is always right?”
The primary danger from AI is not robot rebellion but robots obeying orders perfectly, and the two main problems are: (1) increasing global inequality controlled by dominant AI powers, and (2) erosion of human agency as AI systems know individuals better than they know themselves and make decisions previously made by humans.
“I'm much more frightened about robots always obeying orders than about robots rebelling against the humans. I think the two main problems with AI... is what I mentioned. First, increasing inequality between different parts of the world... And the other issue is AI and human agency.”
Centralized data processing enabled by AI may structurally advantage totalitarian systems over democracies because the more data you have in one place, the better your algorithms become, reversing the 20th-century advantage democracies had (distributed decision-making was more efficient with 1950s technology, making the Soviet centralized model fail).
“new technology suddenly makes central information processing far more efficient than ever before and far more efficient than distributed data processing. Because the more data you have in one place the better your algorithms and so on and so forth. And this kind of tilts the balance between totalitarianism and democracy in favor of totalitarianism.”
Small businesses and entrepreneurs, especially in developing countries, can now access the same digital advertising and marketing tools as large corporations, creating decentralization of economic opportunity and empowerment of individual economic actors.
“we have 90 million small business that use our tools. And now, because of this access to technology they have access to the same tools to do advertising, marketing, reach new customers and grow jobs that previously only the big companies would have had. And that's a big advance. That's a massive decentralization.”
The internet enables people to find niche communities aligned with their genuine interests rather than being tethered to arbitrary local geographic communities, which can be viewed as either fragmentation or as richness and depth in social life depending on one's perspective.
“now with the Internet, that's starting to change, right? And now you have the ability to not just be tethered to your physical location, but to find people who have more niche interests and different kind of subcultures and communities on the Internet, which I think is a really powerful thing.”
Online communities are incapable of fully replacing physical communities because crucial embodied and long-term experiences (dealing with neighbors you dislike, developing social mechanisms for coexistence) can only happen offline, and the ability to exit online communities creates different social dynamics than forced coexistence in geography.
“unlike online communities, can be a wonderful thing, but they are still incapable of replacing physical communities, because there are still so many things... that you can only do with your body and with your physical friends... You can unfriend your Facebook friends, but you can't un-neighbor your neighbors.”
The primary way social media platforms measure success and optimize is easier to do on easily quantifiable signals (engagement, time spent, clicks) than on harder-to-measure meaningful social outcomes (relationship quality, life satisfaction).
“there are ways to get at that... it's a lot easier for us to measure if people are interacting or messaging online than if you're having a meaningful connection physically.”
Israel, despite being a democracy and an ally of the West, is one of the world's leading exporters of surveillance technology, and has developed and tested this technology extensively in occupied Palestinian territories, demonstrating that democratic governance does not guarantee restraint in surveillance technology development.
“In my home country of Israel, at least for Jews it's a democracy and it's one of the leaders of the world in surveillance technology and we basically have one of the biggest laboratories of surveillance technology in the world, which is de-occupied territories. And exactly these kinds of systems are being developed there and exported all over the world.”
When people are shown content that contradicts their existing beliefs, they tend to dismiss it rather than update their beliefs; simply showing opposing viewpoints does not reduce polarization and can entrench existing views.
“all the psychological research that I've seen and the experience that we've had, is that when people see things that don't match what they believe, they just ignore it... certainly, there can be an evolution that happens where a system shows information that you're gonna be interested in and if that's not managed well, that has the risk of pushing you down a path towards adopting a more extreme position”
Addressing major global problems—climate change, AI safety, job displacement from automation—requires unprecedented levels of global cooperation, which is undermined by social fragmentation and by people lacking local sense of prosperity and community.
“we need global cooperation like never before because we are facing unprecedented global problems... we cannot deal with the problems of the environment, of climate change, except through global cooperation.”
Most people, when exposed to misinformation, respond better to being shown a range of related perspectives rather than being told a specific piece of content is false, because the range approach preserves agency and doesn't trigger reactance.
“what ends up working sociologically and psychologically, the thing that ends up actually being effective is giving people a range of choices... you show a series of related articles, related content, then people can work out for themselves, 'Hey, here's the range of different opinions.'”
Values are not something that technology is built with separate from technology itself—values are encoded into design choices, business models, and policy frameworks that make those values operational and consequential.
“what are the values that are encoded, and how does that end up playing out... a lot of the countries that have the ability to invest in future technology... are doing that because it will give their local companies an advantage in the future... the values that are encoded in that are shaped largely by American values.”
Facebook's business model faces a conflict of interest: optimizing for engagement time (keeping people on screens) versus optimizing for meaningful connection (helping people form relationships they will value long-term), and the company has chosen the latter because it aligns with long-term business interests.
“And then the practical question for a service provider like Facebook is, 'What is the goal?' I mean, are we trying to connect people, so ultimately they will leave the screens and go and play football or pick up garbage? Or are we trying to keep them as long as possible on the screens? And there is a conflict of interest there.”
GDPR and similar privacy frameworks should be adopted globally because they encode democratic values, but the more likely alternative if countries don't adopt them is authoritarian data models where governments can access all data for surveillance and control, making strong privacy regulation a bulwark against authoritarianism.
“I think part of the reason why I am so focused on this now is I think, at this point, people around the world recognize that these questions around data, and AI and technology are important. So there's going to be a framework in every country... And the most likely alternative to each country adopting something that encodes the freedoms and rights of something like GDPR... In my mind, the most likely alternative is the authoritarian model, which is currently being spread, which says, you know, 'Every company needs to store everyone's data locally in data centers.' And if I'm a government, I should be able to, you know, go send my military there and be able to get access to whatever data I want.”
What people want in the short term (consuming entertaining content, staying engaged) differs from what they want long term (meaningful connection, personal growth), and designing systems that optimize for short-term engagement undermines long-term user satisfaction and business value.
“there's a difference between what people want first order and what they want second order over time, right? So, right now, you might just consume a video 'cause you think it's silly or fun... and you're like, 'What happened to my time?' So maybe in the narrow, short-term period, you consume some more content, and maybe you saw some more ads, so it seems like it's good for the business. But it actually really isn't over time.”
Effective totalitarianism would not require constant external coercion (like 1984's telescreens and secret police) if systems could manipulate preferences from inside people's minds—total internal alignment with the system's goals would make external surveillance unnecessary.
“It's not 1984 where you need to torture people all the time. No. If you have agents inside their brain, you don't need the external secret police.”
As understanding of human psychology and neuroscience advances, it becomes harder—not easier—to trust that one's inner voice and desires are truly one's own, since more external and internal factors become visible as causal influences on thought.
“the more we understand about the extremely complex world inside us, the less easy it is to simply trust what this inner voice is telling, is saying.”
Silicon Valley has developed the tools to manipulate human beings on a massive scale, and these same tools that can be used to sell people products they don't need can be used to sell people political ideologies, with no fundamental difference in the manipulation mechanism.
“the tools to do that on a massive scale have been developed over the last few decades... Some of them may have been developed with the intention of just selling stuff to people, and selling products to people. But now the same tools that can be used to sell me something I don't really need, can now be used to sell me a politician I really don't need. Or an ideology that I really don't need. It's the same tool.”
There is no metric that can capture the totality of what a good society should optimize for, which is why democratic decision-making (where people decide what dimensions matter) is superior to centralized optimization models that must choose among competing values.
“when you talk about efficiency, a lot of what people are talking about is economic efficiency... Are we increasing GDP? Are we increasing jobs? Are we decreasing poverty? Those things are all good, But, I think, part of what the democratic process does is people get to decide on their own, which of the dimensions in society matter the most to them in their lives.”
Watching or being recommended extreme content creates risk of radicalization through gradual exposure, not through single pieces of content, but it's distinguishable from legitimate algorithmic recommendation by whether people later regret the trajectory they were guided down.
“certainly, there can be an evolution that happens where a system shows information that you're gonna be interested in and if that's not managed well, that has the risk of pushing you down a path towards adopting a more extreme position, or evolving the way you think about it over time... that's our job in either policing the platform... to not be promoting things that people might engage with in the near term, but over the long term, will regret and resent us for having done that.”
When people in Middle America are angry, everyone pays attention because they have power to influence global outcomes, but when people in Mexico, Zambia, or the Philippines are angry, the world cares less because they have less power, creating an imbalance where the pain of Americans reverberates globally more than the pain of people in less powerful countries.
“when people in Middle America are angry, everybody pays attention, because they have their finger on the button. But if people in Mexico or people in Zambia feel angry, we care far less, because they have far less power. The pain, and I'm not saying the pain is not real. The pain is definitely real. But the pain of somebody in Indiana reverberates around the world far more than the pain of somebody in Honduras or in the Philippines simply because of the imbalances of the power in the world.”
If a government can access people's data, it can identify who they are and lock them up, hurt them and their family, and cause real physical harm, which is why data protection is more sensitive than even censorship because it creates the capacity for physical harm.
“if a government can get access to your data, then it can identify who you are and go lock you up, and hurt you and hurt your family, and cause real physical harm in ways that are just really deep. So, I do think that people running these companies have an obligation to try to push back on that”
The modern paradox is that the world is more connected than ever through technology (Internet, virtual realities, social networks) while simultaneously building more physical walls than ever before, including stone walls in addition to cyber walls.
“you have all this new connecting technology and the Internet and virtual realities and social networks. And then the most... One of the top political issues becomes building walls. And not just cyber walls or firewalls, building stone walls. Like the most Stone Age technology is suddenly the most advanced technology.”
Before the Internet, people were geographically tied to their physical location and had limited access to communities and activities; the speaker grew up in a 10,000-person town with only a few options (like Little League) even though his real passion was programming, which had no local community.
“Before the Internet you were really tied to your physical location... I grew up in a town of about 10,000 people and there were only so many different clubs or activities that you could do... I grew up, like a lot of the other kids, playing Little League Baseball... I'm not really into baseball, I'm not really an athlete so why did I play Little League when my real passion was programming computers?... there was no one else, really, in my town who was into programming computers, so I didn't have a peer group or a club that I could do that.”
Open research and publication of AI advances, combined with cloud computing services and a globalized economy for hiring talent, could enable entrepreneurs in smaller countries to leverage state-of-the-art research from the US, Europe, and academic institutions to build competitive AI applications.
“I think that there will be an entrepreneur in some place who can use Amazon Web Services to spin up instances for compute, who can hire people across the world in a globalized economy, and can leverage research that has been done in the US or across Europe or in different open academic institutions or companies that increasingly are publishing their work”
Individual humans are the easiest to manipulate when they have strong beliefs in free will, because they cannot imagine that their thoughts and desires are products of external manipulation rather than their own agency.
“the people that are easiest to manipulate are the people who believe in free will, and will simply identify with whatever thought or desire pops up in their mind because they cannot even imagine that this desire is not a result of my free will, this desire is the result of some external manipulation.”
People generally distrust and resist being told what to do by external parties, and find systems that offer single recommendations or top-down control to be untrustworthy, preferring to maintain autonomy in decision-making.
“people really don't like, and are very distrustful when they feel like they're being told what to do, or just have a single option.”
Democratic countries are slow to mobilize but ultimately more effective than authoritarian systems when their populations become aware of threats and decide to act collectively, and this historical pattern should continue to favor democracy in the AI era.
“the history of democracy versus more totalitarian systems is it always seems like the totalitarian systems are gonna be more efficient and the democracies are just gonna get left behind, but smart people start discussing these issues and caring about them and I do think we see that people do now care much more about their own privacy”
Abuse of platform systems (like Russian interference in elections using fake accounts) is fundamentally different from algorithmic manipulation problems and is primarily a policy and identity verification issue rather than an AI or recommendation system design issue.
“if you have Russian agents or folks from the government, the IRA who are posing as someone else and saying something and you see that content, but you think it's coming from someone else, then that's not an algorithm issue. I mean, that's someone abusing the system, and taking advantage of the fact that you trust that on this platform, someone who's generally gonna be who they say they are”
Meditation reveals that human consciousness is characterized by constant internal conflict and complexity, not by a unified coherent self with clear agency, suggesting that the notion of 'free will' was always an illusion.
“I spend two hours meditating every day. And I go on these long meditation retreats. And my main takeaway from that is it's craziness inside there. And it's so complicated. And the simple, naive belief that the thought that pops up in my mind, this is my free will, this was never the case.”
The US has set a precedent by deciding not to place data centers in countries with weak rule of law, prioritizing data protection over market access, which Facebook argues is aligned with protecting people from surveillance.
“one of the most important decisions that I think I get to make, as the person running this company is, 'Where are we going to build our data centers and store data?' And we've made the decision that we're not going to put data centers in countries that we think have weak rule of law, where people's data may be improperly accessed, and that could put people in harm's way.”
The United States has a structural advantage in technology platforms because most global tech companies are made in the US and encode American values (like free speech), giving the US soft power and cultural influence through these platforms.
“the United States has a major advantage that a lot of the global technology platforms are made here, and, certainly, a lot of the values that are encoded in that are shaped largely by American values... ideas like giving everyone a voice, that's something that is probably very shaped by the American ideas around free speech, and strong adherence to that.”
Surveillance capitalism—systems that constantly interact with and monitor people with the stated goal of providing better recommendations—starts with benign use cases (recommending movies, vacation destinations) but progressively extends to more consequential life decisions (career, marriage, religion), gradually eroding human agency.
“you have this system that constantly interact with you and come to know you and it's all supposedly in your best interests... it starts with recommendations for which movie to watch and where to go on vacation, but as the system becomes better, it gives a recommendation on what to study at college, where to work, ultimately, whom to marry, who to vote for, which religion to join”
An AI arms race could result because each country fears rivals will develop AI weapons first, so even countries that don't want to produce killer robots feel they must do so preemptively, creating a prisoner's dilemma dynamic toward the bottom.
“Because every country will say, "Well, we don't want to produce killer robots, it's a bad idea. But we can't allow our rivals to do it before us, so we must do it first." And then you have a race to the bottom.”
Technological advances in AI are not inherently good or bad; they are morally neutral tools whose impact depends on policy decisions about how they are deployed and what values they serve.
“Technological advance isn't by itself good or bad. I think it's the job of the people who are shepherding it, building it and making policies around it to have policies and make sure that their effort goes towards amplifying the good and mitigating the negative use cases.”
When people feel they have grounding, purpose, community, opportunity, and feel like their lives are working, sociologically and statistically they don't, on balance, seek out extreme communities as much as people who feel lost and dislocated.
“when they feel good about their lives, they don't seek out extreme communities. So there's a lot of work that I think we as an Internet platform provider need to do to lock that down even further. But I actually think creating prosperity is probably one of the better ways, at a macro level, to go at that. But I guess... But maybe just stop there a little. People that feel good about themselves have done some of the most terrible things in human history.”
AI and automation could concentrate immense wealth in a small number of regions (like California and eastern China) while leaving other parts of the world in complete economic bankruptcy, with new software engineering jobs only in wealthy areas but no jobs for textile workers or truck drivers in Honduras and Mexico.
“One of the potential results of the AI revolution could be the concentration of immense wealth in some part of the world, and the complete bankruptcy of other parts. There will be a lot of new jobs for software engineers in California, but there will be maybe no jobs for textile workers and truck drivers in Honduras and Mexico.”
United States has historically presented itself as the leader of the free world and advocate for freedom, but in recent years has presented itself as a nation pursuing narrow self-interest, which undermines the trust necessary for other countries to place data and governance decisions with the US.
“previously, say, if we were sitting here ten years ago, 20 years ago or 40 years ago, when America declared itself to be the leader of the free world, we can argue a lot whether this was the case or not. Or, at least, on the declaratory level, this was how America presented itself to the world. 'We are the leaders of the free world, so trust us. We care about freedom.' But now we see a different America.”
AI systems are not monolithic intelligent entities but collections of specialized algorithms that excel at specific tasks, and the idea of a 'generalized AI' that understands all dimensions of a person better than they know themselves is probably far off from existing and may not be technically feasible.
“You can build systems that can get good at a very specific thing. Helping to understand which of your friends you care the most about so you can rank their content higher on newsfeed. But the idea that there's some kind of generalized AI, that's a monolithic thing that understands all dimensions of who you are in a way that's deeper than you do, I think doesn't exist, and is probably quite far off from existing.”
Internet fragmentation—where different countries adopt different policies and create different regulatory frameworks for data and technology—would break down the global unified model and prevent entrepreneurs in developing countries from leveraging global advances.
“One of the big risks for the future is that the Internet policy in each country ends up looking different. It ends up being fragmented. And if that's the case, then the entrepreneur in the countries that you're talking about, Honduras, probably doesn't have as big of a chance if they can't leverage all the advances that are happening everywhere.”
Countries like Honduras, Ukraine, and Yemen cannot realistically compete in the AI race and will be left behind, raising the question of what happens to them in 20-50 years.
“Does a country like Honduras, Ukraine, Yemen has any real chance of joining the AI race? Or are they... They're already out. It's not going to happen in Yemen, it's not going to happen in Honduras.”
People at some deep level know when something is good for them, and when platforms show them what they actually care about rather than manipulated viral content, the better outcome comes sooner than expected (sometimes in a year instead of the expected five years).
“I think people at some deep level know when something is good... when you make these decisions that they're maybe taking some pain in the near term in order to get to what will be a better case down the line, that better case, maybe you think it'll take five years, but actually it ends up coming in a year.”
If most of a person's friends share their religion or political ideology, but 5-10% have different backgrounds and ideas, this natural diversity in social networks provides broader perspective and moderates polarization.
“even if most of your friends share your religion or your political ideology, you're probably gonna have five or 10% of friends who come from a different background who have different ideas and at least that's getting in is well, so you're getting a broader range of views.”