Public transcript

Kissinger and the Future of AI ft. Eric Schmidt

By: Institute of Politics at Harvard Kennedy School
Length: 1 h 20 min
Language: English
Text from: the video's captions
Transcribed: 14 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

Eric Schmidt, former Google CEO, joins Graham Allison at the Harvard Kennedy School's JFK Jr. Forum to discuss Henry Kissinger's legacy and the future of AI. Schmidt describes his friendship with Kissinger, the

Key points

Eric Schmidt tutored Henry Kissinger on AI late in Kissinger's life, and their book Genesis argues AI is an epochal change because humans have never faced a non-human competitor of similar intelligence.
Schmidt describes the San Francisco consensus: language models, agents, and reasoning will combine into recursive self-improvement, estimated at 2 years in San Francisco and 4 years by Schmidt.
Schmidt reframes Kissinger's 1969 nuclear challenge for the AI era: there is no higher duty than to preserve human agency and human freedom.
China is focused on embedding AI in everything and will outcompete the US on adoption, while the US pursues superintelligence; China adds about a gigawatt of electricity to its grid every day.
China's open-weights strategy means most of the world will likely use free Chinese models, which come with Chinese values and biases embedded.
Schmidt predicts large companies will ultimately not release their largest models because it will be too dangerous to do so.
Schmidt's primary security concern is cyber attacks, since AI that can write code can write attacks; the eventual solution is good AI fighting bad AI.
Schmidt distinguishes superhuman behavior from true superintelligence like Einstein's, and argues the latter remains a very hard boundary.

Questions it answers

12:00Opening and introductions

Who is speaking and what is the event about?

Students introduce a JFK Jr. Forum discussion on Kissinger, AI, and the future, featuring Eric Schmidt and Graham Allison. The framing is that AI is reshaping economies, warfare, and society at a pivotal moment in history.

The forum is framed around AI's impact on global stability and human society.
Schmidt is presented as a 'statesman' voice in AI rather than someone talking his own book.

17:55Remembering Henry Kissinger

What did Henry Kissinger mean to Eric Schmidt and Graham Allison?

Schmidt recounts tutoring Kissinger on AI late in his life and describes Kissinger's brilliance, his escape from Nazi Germany, and his lifelong drive to avoid another world war. Allison adds that Kissinger's realism was always about constructing a viable order to prevent catastrophic war.

Kissinger's core goal, per Schmidt, was preventing World War III.
Kissinger's Harvard record includes the longest undergraduate thesis, prompting the 350-page limit.
Allison describes Kissinger's strategic acumen and pursuit of a viable world order.

27:00Kissinger and the AI question

How did Kissinger think about artificial intelligence?

After hearing Demis Hassabis speak, Kissinger grasped that AI raises the question of what it means to be human. Schmidt and Kissinger's book Genesis argues AI is an epochal change because humans have never faced a non-human competitor of similar or greater intelligence.

Kissinger framed AI as an epochal change in human reasoning.
The Genesis book warns of an existential security dilemma over superintelligence.
Schmidt sees AI's effect on child development as a mass experiment with no data.

28:45The San Francisco consensus and superintelligence

What is the San Francisco consensus on superintelligence?

Schmidt describes a belief that language models, agents, and reasoning will combine into recursive self-improvement, with San Francisco estimates of 2 years and his own estimate of 4 years. He stresses the need to keep humans in charge and to preserve human agency and freedom as the modern Kissinger challenge.

Schmidt estimates recursive self-improvement in about 4 years.
He argues some point requires humans to say 'too far' and retain agency.
The Kissinger challenge is reframed: no higher duty than preserving human agency and freedom.

35:55US-China AI rivalry

How do the US and China's AI strategies differ?

Schmidt says China is focused on embedding AI in everything and will outcompete on adoption, while the US pursues superintelligence. China has vast electricity and open-source models; America has chips and deep capital markets, but both rely on the private sector.

China adds roughly a gigawatt of electricity to its grid every day.
China's open-weights strategy means most of the world may use Chinese models for free.
America leads in chips and capital markets; China leads in energy and adoption.

38:45Upside, risks, and open vs. closed

What excites and worries Schmidt about AI's near future?

Schmidt says AI is not a bubble but underhyped because it automates business workflows, and describes watching AI write programs as the end of his own 55-year craft. He worries about malicious actors using these tools, and predicts companies will not release their largest models because it is too dangerous.

Schmidt says AI is underhyped, not a bubble, because it automates businesses.
He predicts large companies will not release their largest models for safety reasons.
Free Chinese models come with Chinese values and biases embedded.

51:30Audience Q&A

What did Schmidt tell students in the audience Q&A?

Schmidt fields questions on cyber governance, open vs. closed models, moral responsibility for AI-shaped decisions, US-India partnership, enterprise adoption, consciousness in AI, an IAEA for AI, jobs and purpose, algorithmic diplomacy, and whether superintelligence will be copied. He emphasizes cyber attack risk, human dignity and purpose in work, and that superhuman behavior is not the same as true superintelligence.

Schmidt expects many more cyber attacks as AI writes code.
He sees misinformation and anchor bias as core threats to democracy.
He distinguishes superhuman behavior from true superintelligence like Einstein's.

Notes

Kissinger's Legacy and AI

Eric Schmidt tutored Henry Kissinger on AI late in Kissinger's life after Kissinger heard Demis Hassabis speak and immediately grasped the question: "What does it mean to be human in the age of AI?"
Schmidt argues Kissinger's driving goal, after fleeing Nazi Germany and serving in World War II, was to prevent a World War III; historians agree on the goal even if tactics are debated.
Kissinger's undergraduate thesis on Kant was the longest in Harvard's history, prompting the 350-page limit that still applies.
Graham Allison emphasizes Kissinger's strategic acumen: lifting a problem to its full 360-degree strategic challenge and finding points where policy could be impacted.

The San Francisco Consensus

Schmidt describes a progression: language models, then agents (which can be concatenated into workflows), then reasoning.
The consensus holds these will combine into recursive self-improvement — AI learning on its own. San Francisco estimates 2 years; Schmidt doubles it to 4 years.
Scaling laws have not slowed: Gemini 3 beat OpenAI 5, which beat Claude 4.5, which beat DeepSeek.
Schmidt insists a point will come where someone must say "we went too far" and keep humans in charge — for example, if a computer seeks access to weapons on its own.

US-China Rivalry

China focuses on embedding AI in every product and will outcompete on adoption; the US pursues superintelligence.
China adds roughly a gigawatt of electricity to its grid every day; the US has built zero new nuclear-scale plants and struggles to provision power.
China pursues open weights and open source, so most of the world may use free Chinese models — which come with Chinese values and biases.
American closed-source models are driven by economics: companies borrowing $50 billion cannot give models away.

Key Concerns

Cyber attacks: if AI can write code, it can write attacks; the solution to bad AI is good AI fighting bad AI.
Children: society is running a mass experiment with addictive systems; we have no data on children whose best friend is non-human.
Democracy: algorithmic misinformation exploits anchor bias; Schmidt favors free speech but not boosted speech.
Jobs: human dignity involves purpose, and losing jobs is an emotional crisis, not just financial.