PDF Summary:Nexus, by Yuval Noah Harari

Forget killer robots—according to historian Yuval Noah Harari, the real AI threat is already here, reshaping how we think and interact. In Nexus, Harari reveals how artificial intelligence is poised to take control of humanity's greatest superpower: our ability to tell stories that unite, inspire, and mobilize people.

This guide unpacks Harari's urgent warning about AI's growing influence over our information networks, showing how it could undermine democracy, amplify social division, and erode our grasp on truth itself. But it also offers hope, outlining practical strategies for maintaining human agency in an AI-dominated world. Our commentary draws connections to research in psychology, network theory, and digital ethics to help you understand how information shapes society and why that matters for your daily life. Whether you're a tech enthusiast, a concerned citizen, or simply someone who’s trying to navigate our rapidly changing world, this guide provides context for one of the most critical challenges of our time.

(continued)...

In managing the flow of information among people, social groups have a choice to make: Do they want to prioritize the spread of truth, or control the flow of information to maintain social order? What we should hope for, Harari explains, is information networks that can help us strike a balance between truth and order. Enabling a flow of information that errs too far on the side of one or the other can have disastrous consequences, as we’ll discuss below.

How Do Coral Reef Fish Avoid Spreading Misinformation?

Societies must strike a delicate balance between propagating truthful information and maintaining social order—and this challenge isn't unique to humans. Scientists studying wild reef fish discovered one example of how animals naturally regulate misinformation.

When one fish flees from a perceived threat, like an aggressive neighbor, it creates visual motion cues that other fish can see. If every fish immediately fled whenever they saw another fish fleeing, false alarms would constantly cascade through the group. Instead, the fish have evolved a clever solution: They adjust how sensitive they are to these visual cues based on their recent experiences. If they've seen a lot of sudden movements lately that didn't signal real danger, they become less likely to react to the next sudden movement. This “dynamic sensitivity adjustment” helps prevent false alarms from spreading widely through the group—essentially creating a natural defense against misinformation.

What Happens When We Value Truth More Than Order?

As we saw during the Scientific Revolution, human society can flourish when we seek the truth. It pushes human thought forward when we’re open to questioning long-held beliefs and replacing disproven information with updated observations. But, Harari notes, a tradeoff typically occurs: An emphasis on truth comes at the expense of order. The perception that the facts are changing can be destabilizing. For example, Galileo’s discovery of the heliocentric nature of our solar system upended the religious societies of Renaissance Europe. Likewise, Darwin’s theory of evolution threw the Victorian-era understanding of the natural world into chaos.

(Shortform note: The controversy over the CDC’s changing recommendations about masks early in the Covid-19 pandemic illustrates how an emphasis on truth over order can be destabilizing. Some public health officials suggest much of the confusion and erosion of trust during the pandemic could have been avoided if public health authorities had stated up-front that their recommendations were provisional and subject to change as new information about the novel virus emerged. Instead, they presented information—like the initial recommendation that the general public didn’t need to wear masks—as settled science that they then had to walk back, upending past recommendations and diminishing public trust.)

What Happens When We Value Order More Than Truth?

On the other hand, if a society considers order its highest value, it can take control of the flow of information to achieve that end. (If you manipulate the flow of information, you can manipulate what people think and do.) Unlike in a democracy—where information is shared freely with citizens so they can fact-check it and correct errors and falsehoods, even those put forth by the state—a dictatorship doesn’t want open conversation. Authoritarian regimes selectively promote ideas without regard for whether they’re demonstrably true or untrue. The logic goes that if knowledge becomes too freely available, then the stories the regime is built upon could be thrown into doubt and potentially rejected by the state’s citizens.

(Shortform note: While Harari notes that authoritarian regimes suppress dissent by controlling information flows, some political theorists say their core strategy is the manipulation of language itself: defining the meanings of words to construct an artificial reality that serves the regime's interests, making any divergent perspectives appear insane and false. In George Orwell’s dystopian novel 1984, the Party seeks to control not just information flows, but the very meanings of words to define what is “real” and “true” according to their agenda. By dominating language, the Party can determine people's socio-cultural context and their perception of reality, reducing it to a pseudo-reality manufactured by the regime.)

How Is AI Changing Our Relationship With Information?

Harari explains that we’re living in an “information age,” where knowledge is proliferating, access to information is democratized, and everyone with a smartphone and internet access can share their ideas with the world. As we develop tools like AI, we’re increasing the speed at which stories can be shared. When you consider what makes a human society free, equitable, or democratic, having more information sounds like an inherent good. (An especially American expression of this is the idea, written down by Thomas Jefferson, is that a well-informed electorate plays a vital role in keeping authorities in check and guarding against tyranny.)

But, counter to that notion, Harari worries that recent developments that make information more accessible to us threaten to tip the balance toward the most extreme, least truthful, and most divisive messages.

(Shortform note: Even before AI entered the scene, some experts questioned whether our increasing access to information is an inherent good. Early internet idealists envisioned a digital utopia where open access to knowledge would lead to more informed, rational discourse that disrupted monopolies of information. True to form, the “old internet”—characterized by decentralized blogs and forums—fostered niche communities and a diversity of perspectives. However, the “new internet” is dominated by profit-driven platforms that Kyle Chaka, author of Filterworld, argues “flatten culture” by concentrating the flow of information in particular ways. As a result, new monopolies of information are emerging—and undermining the internet’s democratic potential by steering public discourse toward homogenized content.)

Because humans are wired to seek out a good story rather than to pursue the truth, putting AI in a position to determine what ideas we're exposed to could have potentially disastrous consequences. Harari identifies three main dangers: AI's disregard for truth, its ability to manipulate and polarize us, and its potential to surpass human understanding of the world. For each of these threats, he offers specific recommendations for how we can maintain human agency and control over our information landscape.

Danger #1: Since We Don’t Prioritize the Truth, AI Doesn’t Either

Scientists have made it possible to build AI models that can generate language and tell stories just like humans. Harari contends that AI’s ability to create compelling stories and produce the illusion of emotions (and emotional intimacy) is where its real danger lies. When we talk to an AI-powered chatbot like ChatGPT, it’s easy to lose sight of the fact that these systems aren’t human and don’t have a vested interest in telling the truth. That will only become harder to recognize as AI gets better and better at mimicking human emotions—and creating the illusion that it thinks and feels like we do. So it will only become easier for us to lose sight of the fact that AI isn’t prioritizing the truth when it selects and generates information for us.

(Shortform note: Some experts are skeptical that AI can ever develop human-like emotions. But that hasn’t stopped users from striking up romantic relationships with chatbots—or researchers from trying to teach AI the cognitive empathy to recognize and respond to human emotions. This could have interesting implications: Klara and the Sun author Kazuo Ishiguro sees creating art as one of the most interesting things AI could do if it develops empathy or learns to understand the logic underlying our emotions. Ishiguro says that if AI can create something that makes us laugh or cry—art that moves people and changes how we see the world—then we’ll have “reached an interesting point, if not quite a dangerous point.”)

Harari says that AI already influences what information we consume: An algorithm—a set of mathematical instructions that tell a computer what to do to solve a problem—chooses what you see on a social network or a news app. Facebook’s algorithm, for example, chooses posts to maximize the time you spend in the app. The best way for it to do that is not to show you stories that are true, but content that provokes an emotional reaction. So it selects posts that make you angry, fuel your animosity for people who aren’t like you, and confirm what you already believe about the world. That’s why social media feeds are flooded with fake news, conspiracy theories, and inflammatory ideas. Harari thinks this effect will only become more pronounced as AI is curating and creating more of the content we consume.

(Shortform note: While Harari discusses them both, it’s important to realize that an algorithm isn’t synonymous with AI. But algorithms represent a necessary component of AI systems because each algorithm is a set of instructions that are used to define the process an AI system will use to make a given decision. Once that process is defined, the AI model can use the data it has access to—both the data it’s provided to complete a task and, indirectly, the data it was trained on when it was first learning to complete the task—to follow the algorithm’s instructions, go through the outlined process, and make the decision.)

How to Fix It: Pay Attention to What’s True

Harari argues that we need to take deliberate steps to tilt the balance in favor of truth as AI becomes more powerful. While his proposed solutions are somewhat abstract, he emphasizes two main approaches: being proactive about highlighting truthful information and maintaining decentralized networks where information can flow freely among institutions and individuals who can identify and correct falsehoods.

What Do Other Experts Think of the Solutions Harari Recommends?

Harari's recommendations sometimes raise more questions than they answer. While he suggests that the general public should take an active role in promoting truthful information, he doesn't specify how individuals can meaningfully counter the influence of tech giants that control our primary information platforms. Similarly, his call for decentralized networks doesn’t contend with the reality that a handful of powerful corporations currently dominate how most people access and share information.

While Harari argues that we must hold tech companies accountable and demand transparency in how they develop and deploy AI, he stops short of endorsing specific policies or regulations. Other experts have proposed concrete measures like antitrust legislation, mandatory AI safety standards, or independent oversight boards, but Harari focuses more on raising awareness about AI’s dangers than prescribing particular fixes. His central message is clear—we must be intentional about how we develop and use artificial intelligence—but the exact path to achieving this remains undefined.

While Harari raises valid concerns about AI's ability to manipulate public discourse, some critics argue his view is too apocalyptic. They point out that current AI systems still heavily depend on human inputs, prompts, and oversight. Some experts believe we need a more nuanced approach that focuses on aligning AI development with human values and ethical frameworks, addressing the conditions and decisions that give AI its power. In other words, Harari's perspective that AI poses an imminent existential threat to humanity may overestimate AI's current capabilities—and underestimate human agency.

Danger #2: We’re Becoming Easier to Manipulate and Polarize

Harari warns that as AI increasingly controls what information we see, algorithms will push us toward more extreme ideas and greater polarization. We can already see this happening with today's algorithmically stoked outrage and clickbait-fueled misinformation. Harari believes the problem will only intensify as AI becomes more sophisticated and commercialized, and he predicts AI systems will create, interpret, and spread stories without human intervention. One system might select pieces of information, another spin that information into a story, and yet another determine which stories to show to which users. This will leave us increasingly vulnerable to manipulation by AI systems and the corporations that control them.

(Shortform note: Farhad Manjoo, author of True Enough, has long argued that algorithms increase polarization by promoting an echo chamber effect and eroding trust in documentary evidence. Algorithms enable the spread of misinformation and manipulative narratives by allowing people to selectively consume information that aligns with their existing beliefs and biases. The abundance of information sources online has not led to more rational, fact-based discourse. Instead, documentary proof seems to have lost its power as people filter evidence through their own biases and conspiracy theorists can cherry-pick information to fit their preferred narratives.)

Harari explains this represents a significant shift in power: The ability to set the cultural agenda and shape public discourse—traditionally the domain of newspaper editors, book authors, and intellectuals—will increasingly belong to AI systems optimized not for truth or social cohesion, but for engagement and profit.

(Shortform note: From the 1950s through the 1980s, newspaper editors and television news anchors like Walter Cronkite wielded enormous influence over public discourse. This era of concentrated media influence—when editors shaped national conversations—stands in contrast to today's fragmented media landscape. Now, with countless online news sources, social media platforms, and AI-curated content feeds, no single editorial voice carries the same weight. Instead, we have a cacophony of voices that often drowns out traditional journalism's authority. Aaron Sorkin's show The Newsroom (2012-2014) nostalgically depicts this loss, following a fictional news anchor's struggle to reclaim journalism's role as a trusted source of truth.)

How to Fix It: Build Institutions to Help People Understand What AI Is Doing

To counter AI's growing influence over public opinion, Harari calls for the creation of new institutions to monitor artificial intelligence and inform the public about its capabilities and risks. He argues that we shouldn't let tech giants regulate themselves. While his vision for these oversight institutions remains abstract, he suggests they should function somewhat like today's free press or academic institutions, serving as independent watchdogs that can help the public understand and evaluate AI's decisions and actions. Harari frames this as primarily a political challenge, arguing that we need the collective will to establish these safeguards.

(Shortform note: Harari’s call for oversight coincides with the 2025 arrival of DeepSeek, a Chinese startup that built two AI models that rival those from the best American labs, but which were trained with innovative techniques to make them more efficient in terms of both cost and computing power. This development renewed concerns about safety and fueled urgent calls for regulation and transparent oversight, given that there aren’t international laws restricting AI development. Many think it’s not possible to stop progress, but that hasn’t stopped them from trying, as when 30,000 people—including Harari—signed a 2023 open letter calling for a moratorium on training powerful AI systems.)

Danger #3: We’re Setting AI Up to Understand the World Better Than We Do

Harari warns that we're creating AI systems that will soon surpass human capabilities in understanding and manipulating the shared stories that organize our societies. This shift represents a real danger: While humans have traditionally maintained power through our unique ability to create and control these shared fictions—like laws, money, and social institutions—AI is poised to eclipse us at our own game.

The root of this problem lies in human nature. We often lack the patience and attention span to dig deep into complex truths, preferring simpler stories that are easier to grasp. AI systems, in contrast, can process vast amounts of information, and work together in ways humans can't—while one AI system analyzes market trends, another can simultaneously study legal documents, and thousands more can coordinate to spot patterns across these different domains. They can comprehend intricate systems—like legal codes and financial markets—far better than most humans can. They can even create entirely new frameworks that go beyond human understanding. This capability gap marks an unprecedented shift in power.

(Shortform note: While Harari warns that AI will soon surpass human capabilities in understanding complex systems and narratives, some AI researchers clarify that a major obstacle is AI's lack of common-sense reasoning and understanding of context, which humans take for granted. For example, AI may fail to differentiate between a white shirt and a white wall, mistakes that seem silly to us but reveal the challenges AI faces in making educated assumptions and responding appropriately to familiar situations. Researchers are now trying to solve this problem by studying how humans process emotions and make decisions, combining insights from AI and cognitive science.)

For tens of thousands of years, humans have been the sole architects of our information networks, generating and sharing the ideas that shape our societies. But as AI systems become more sophisticated, we'll increasingly rely on them to process information and make decisions. When we delegate decisions, we also surrender our understanding of the information that drives them—potentially giving up our position as the primary shapers of human society. (Shortform note: Experts say AI can augment our decision-making by sifting through mountains of data to spot patterns invisible to humans. From financial markets to climate science, this enables AI to provide insights that could solve our most complex challenges.)

(Shortform note: Many experts worry that as AI learns to understand our world, it’s adeptly picking up our biases and prejudices—for example, it misidentifies people of color with facial recognition AI—and might threaten social justice by reproducing the social inequalities it observes. AI’s “algorithmic bias” could pose a disproportionate risk to women, undermine LGBTQ identity, and put transgender people at particular risk. These injustices might be even worse if AI leaves us feeling disconnected from each other, as happened in the 2008 film WALL-E, where humans have left Earth and live on a spaceship piloted by robots. By putting AI in charge, the humans in the movie lose focus on what’s happening in our world.)

How to Fix It: Focus on Maintaining Human Agency

Harari believes to deal with this transition, we must develop new frameworks to maintain human agency and ethical guardrails. He explains we should consider training AI systems to express self-doubt, seek human feedback, and acknowledge their own fallibility—essentially equipping them with a self-awareness of the limits of their knowledge. He also recommends that we use AI to augment human decision-making instead of replacing it, which would help retain human values and oversight.

(Shortform note: Can AI really express self-doubt? Experts say that dealing with uncertainty through probabilistic techniques can make AI smarter by allowing it to express confidence levels in predictions and avoid overconfident mistakes, which is crucial for high-stakes applications like autonomous vehicles.)

(Shortform note: Experts say the extent to which we can maintain human agency as AI becomes more powerful depends heavily on whether we establish appropriate governance and ethical frameworks—and whether AI developers prioritize giving users control and transparency in AI decision-making processes. Experts contend that AI developers must strive to give users control over AI systems through adjustable settings, explainable decisions, and override options to make sure that humans stay informed and in control. But they worry some industries and companies may opt for more autonomous AI decision-making without human intervention, which may reduce individual control.)

The Real Risk: How Humans Choose to Use AI

The existential threat of artificial intelligence, Harari argues, doesn't come from malevolent computers but from human decision-making. While we often hear that technology itself poses the danger—that we repeatedly create tools with the potential to destroy us—Harari sees the core problem differently. The real risk lies in how humans choose to use these powerful new tools, especially when we make those choices based on bad information.

This insight shifts the focus from AI itself to the human systems that control it. Harari warns that if paranoid dictators or terrorists gain unlimited power over AI systems, catastrophic consequences could follow. But these outcomes aren't inevitable; they depend entirely on human decisions about how to develop and deploy the technology.

Harari's conclusion is ultimately hopeful: If we can understand the true impact of our choices about AI—and ensure those choices are based on reliable information rather than manipulation or misinformation—we can harness this powerful technology to benefit humanity rather than harm it. The key is not to fear AI itself, but to be thoughtful and intentional about how we choose to use it. Like any tool, we can use AI to achieve positive or negative ends, and we have to prioritize making choices that will benefit humanity, not destroy it.

How Can We Use AI Wisely?

Other experts join Harari in seeing AI as a powerful tool that can be used or misused. Geoffrey Hinton, whose groundbreaking work on neural networks and deep learning algorithms earned him the 2024 Nobel Prize in Physics, believes advanced AI systems could pose ethical dilemmas, regulatory challenges, and inherent risks. He lists a number of possible dangers: AI could surpass human intelligence, develop its own goals, learn unexpected behaviors from the vast datasets it analyzes, or be misused for malicious purposes, like manipulating elections or fighting wars.

Hinton advocates international regulation to address these risks. He suggests curbing the development of potentially dangerous AI systems in the same way we ban chemical weapons.

Hinton's departure from Google and his warnings about AI echo the misgivings of other scientists who have expressed regret or apprehension about the potential misuse of their creations—like Albert Einstein and J. Robert Oppenheimer, whose work contributed to the creation of the atomic bomb.

But fortunately, unlike nuclear weapons (which atomic scientists agree have a singular, destructive purpose), generative AI can be beneficial if used wisely. Hinton believes the timeline for artificial general intelligence (AGI) may be much shorter than previously thought; it might emerge in decades rather than centuries. This means governments, tech companies, and the public must act now to create regulations, ethical guidelines, and oversight mechanisms that can help us harness AI's benefits while protecting against its risks.