The next 50 years: humanity, AI, power | Yuval Noah Harari

In this conversation with Tom Parker, Yuval Noah Harari argues that AI's political significance lies in its growing ability to operate the language-based institutions through which humans govern. The interview moves from democracy and legal personhood to technological dependence, intimate relationships, and the difference between solving problems and choosing worthwhile goals.

Key points

• Democracy depends on the means of conversation. Responding to Parker's question about democratic disruption, Harari argues that large populations need communication infrastructure to deliberate together. Newspapers, broadcasting, and the internet made such conversations possible at scale; changes to those systems therefore alter democracy itself, rather than merely changing how politicians communicate. He treats social media as an example of that disruption. 05:30–09:05(opens in a new tab)

• He distinguishes agents from tools by decision-making and invention. A printing press reproduces a book but does not choose what to publish or write the next book. Harari contrasts this with his conception of AI agents that make decisions and generate new ideas. His projected route to greater AI power is bureaucratic: operating banks, companies, and legal systems, not necessarily a visible robot rebellion. These are his proposed distinctions and forecasts, not a test establishing that every AI system has those abilities. 09:05–14:27(opens in a new tab)

• Legal personhood could turn capability into institutional power. Parker asks which legal decision would be hardest to reverse. Harari points to allowing AI to act as an independent legal participant: holding accounts, investing, employing people, or bringing lawsuits. He contrasts corporations whose decisions humans make with his scenario of corporations run by AI without human executives or shareholders. His concern is that formal authority would let AI operate the institutions that already allocate resources and constrain people. 14:27–19:22(opens in a new tab)

• Dependence on remotely controlled infrastructure can limit sovereignty. Harari contrasts land and physical goods, which cannot all be moved to an imperial center or disabled after sale, with information and networked services that can remain centrally controlled. He warns that countries relying on foreign AI infrastructure could lose access if its controllers withdraw service. This is a conditional argument about particular architectures, not a claim that every AI deployment has a remote shutdown mechanism. 24:02–29:32(opens in a new tab)

• More knowledge need not make everyday systems easier to understand. Harari argues that truth requires costly investigation, while fiction can be cheap, simple, and flattering; journalism and science help sustain the effort to establish facts. He then distinguishes discovering facts from creating systems: knowing more about biology does not mean understanding the financial institutions that shape one's life. His forecast is that AI bankers and investors operating continuously could make finance too fast and mathematically complex for humans to follow. That is a proposed risk of greater capability, not evidence that AI has already made human oversight impossible. 30:32–37:50(opens in a new tab)

• Persuasive emotional language does not settle whether AI feels. Harari separates the ability to describe love convincingly from the experience of love. He treats AI consciousness as unresolved and worries about relationships built around convincing expressions of feeling. Parker's closing question assumes an absence of emotion; Harari's preceding answer is more cautious: he says we do not know what, if anything, lies behind the words. 40:24–44:38(opens in a new tab)

• Cheaper intelligence increases the importance of choosing goals. Harari defines intelligence as solving a problem and wisdom as knowing which problem deserves solving. His genie analogy illustrates the distinction: getting a wish fulfilled can go badly if the wish was poorly chosen. His hoped-for outcome is both safer AI and greater human attention to which desires to pursue, not problem-solving capability alone. 44:38–48:03(opens in a new tab)

Full conversation: Watch on YouTube(opens in a new tab).