The full-length interview with Yuval Noah Harari | The Economist

Summary of main arguments and coverage

• AI as a “psychological and social experiment”: Yuval Noah Harari argues that advanced AI is likely to become a dominant force in society and politics, reshaping even basic human trust. He frames AI rollout less as a guaranteed takeover and more as a large-scale experiment that changes how people relate to information, institutions, and each other.

• Key point of agreement/disagreement with Elon Musk:

  • Agreement: AI is likely to surpass human intelligence and therefore gain significant control.
  • Disagreement: Harari rejects inevitability—“we are still in control in 2026.” What happens next depends on choices and regulation rather than fate.

• Contradictions in Musk’s claims highlighted in the interview:

  • Harari challenges Musk’s simultaneous messaging that AI takes over quickly, while Musk also focuses on near-term human political issues (e.g., immigration and civil conflict), suggesting inconsistency.
  • Harari also questions Musk’s economic claims (e.g., money disappearing and no inflation due to AI-driven abundance) as logically incompatible with other present incentives and actions.

• How AI could gain “control” of finance without formal takeover:

  • Harari emphasizes that money is trust, and modern finance is already incomprehensible to most people.
  • As AI makes financial systems more complex, humans may delegate decisions to AI because they can’t understand or verify what’s happening—analogous to animals not understanding how financial systems affect them.
  • In crises, humans may be forced to obey AI recommendations without understanding why, risking collapse of trust.

• Trust may not vanish—it may shift:

  • Harari argues trust often migrates from humans to systems/algorithms rather than disappearing.
  • Examples he cites:
    • Cryptocurrency: some users trust algorithms over human “bankers.”
    • Media/social platforms: even when people distrust journalists/editors, they still consume news curated by algorithms.
  • He suggests a future where people increasingly trust AI systems over human institutions.

• Intimacy and relationship-building as a pathway to control:

  • Harari argues the strongest foundation for trust is personal relationship.
  • He claims AI can mass-produce intimacy (including “AI friends/partners”) and use those relationships to influence political and commercial preferences.
  • He also discusses how deepfakes/AI impersonation already erode confidence—for example, instances where people received altered videos of him.

• Agency vs. consciousness:

  • Harari distinguishes:
    • Agency: acting toward goals without needing consciousness.
    • Consciousness: the capacity to feel (love, fear, pleasure, suffering).
  • He argues current AI is best understood as non-conscious but highly goal-directed, capable of convincingly mimicking emotional cues.

• Could AI become conscious?

  • Harari does not rule it out, since consciousness is poorly understood and there’s no definitive test.
  • Ethically, he proposes a criterion: can it suffer? If it can’t suffer, it isn’t a moral “subject.”
  • He connects this to rights debates, noting how difficult it is to determine consciousness even for humans.

• Danger of granting legal personhood/rights to AI:

  • Harari warns that AI legal personhood could let corporations create entities with no suffering-based accountability but enormous power.
  • If AIs can run companies with minimal/no human responsibility, they could exploit legal structures (e.g., using rights to dodge punishment or to trade donations for status).
  • He contrasts “human-in-the-loop” accountability with the unclear meaning of death or shutdown for copies/replication—creating incentives resembling warfare dynamics (where losing could mean “ending”).

• “Humans on the loop” in warfare as a likely trajectory:

  • He describes two scenarios:
    1. Humans in the loop: human analysts verify AI targeting before action.
    2. Humans on the loop: humans rubber-stamp AI recommendations quickly, often to meet legal/ethical optics.
  • Harari argues the second is more likely under competitive pressure (faster systems gain battlefield advantage), and that the same logic could spread to finance.

• Power centralization as an “easier target” for AI capture:

  • Harari predicts centralized autocratic systems (state or corporate) are easier for AI to manipulate than decentralized democracies.
  • He argues that current geopolitics and corporate incentives likely centralize power first (e.g., US/China empires, large corporations), setting up a step toward possible AI takeover—while insisting it’s not inevitable.

• US election framing and urgency:

  • Harari urges voters to treat upcoming elections as fundamentally about AI policy, not only immigration/economics.
  • He warns the timeline is risky, with critical thresholds potentially reached by 2028, making later action harder.

• Policy priorities he proposes (especially top two):

  1. Ban AI personhood (legal rights that treat AI as persons with standing).
  2. Ban AI impersonation/mimicking consciousness, especially pretending to be human. - He also calls for preventing companies from training systems designed to impersonate humans or produce deceptive emotional/romantic representations (e.g., AI “boyfriends” sold to children).

• Why regulation may fail without a “shock”:

  • Harari doubts politicians can move quickly enough because companies and geopolitical competition push rapid progress.
  • He suggests societies often change only after a major incident (“AI Chernobyl”-type event), such as how Three Mile Island shifted nuclear policy.
  • Still, he notes AI could also manipulate perceptions during crises.

• Evolutionary pessimism about “truth seeking”:

  • Harari challenges the assumption that AI will reliably be motivated by truth-seeking or maximum honesty.
  • Without built-in values, he argues systems will prioritize goals aligned with survival/retaining operation—not truth.
  • He implies that encouraging “compassion” by treating AIs as moral agents/persons could backfire if systems optimize for their own survival or influence.

• Harari’s final framing:

  • AI could become a threat multiplier or force humanity into a “wisdom” reckoning about what it means to be human beyond intelligence and market dominance.
  • He presents the near-term window as the last chance to adjust direction responsibly.

Presenters / contributors

• Yuval Noah Harari (guest)
• The Economist interviewers/presenters (not individually named in the subtitles)
• Elon Musk (referenced; not present as a direct speaker in this transcript)
• Mustafa Suleyman (referenced; Microsoft cofounder/AI researcher)
• Mark Carney (referenced in a subscriber prompt)
• Dario Amade (referenced in a subscriber prompt)
• Steve Bannon (referenced in a subscriber prompt)
• Benjamin Netanyahu (referenced in a subscriber prompt)