What this covers

This is a 26-minute interview recorded at TED2025, in which technologist Bilawal Sidhu questions Eric Schmidt on where AI is heading. Schmidt starts with technical history, then moves to geopolitics, safety, and what AI might do for medicine and work. On the technical side, he invokes Grove Giveth, Gates Taketh Away: hardware gains get absorbed by software complexity, a pattern he expects to intensify as systems move from deep learning toward reinforcement learning and test-time compute. He also treats AlphaGo's unprecedented move against Lee Sedol in 2016 as the moment the revolution began, because it showed machines producing solutions outside human intuition.

The most contested material concerns competition between powers. Schmidt describes a race to superintelligence in which a side six months behind risks permanent subjugation. He lays out an escalation ladder for the trailing party, running from stealing code to spying to sabotage and finally to striking data centers. He asks whether Henry Kissinger's mutually assured destruction logic would carry over to this setting. He also rejects Yoshua Bengio's call to halt agentic development as unworkable in a competitive market, and argues for enforceable guardrails instead. He names red lines that should trigger shutdown: recursive self-improvement, direct weapons access, and self-replication.

Openness gets its own thread. Sidhu presses on whether releasing models hands capabilities to adversaries after DeepSeek's results. Schmidt defends open-source as the foundation of science, while naming nuclear, cyber, and biological risks. He expects China to lead in open models while US firms stay closed. On surveillance, he argues that preventing 1984-style outcomes may require 1984-like tools, and he points to zero-knowledge proofs for verifying personhood without exposing identity. The closing stretch is more hopeful and more speculative: AI doctors for places with little healthcare, the claim that every profession must adopt these tools, and a forecast that shrinking populations will keep human workers economically necessary.

Sharpest takeaway

Schmidt argues that AI's arrival represents a civilizational inflection point requiring urgent governance frameworks, particularly around preventing preemptive conflict between US and China, while simultaneously pursuing AI benefits that can solve longstanding human problems.

  • The slope of AI improvement in network-effect businesses creates winner-take-all dynamics that incentivize preemptive action by adversaries within 5 years
  • Open-source AI proliferation poses dual-use risks (cyber, biological, nuclear) but suppressing it contradicts the foundational open-science model underlying technological progress
  • AI can deliver 30% annual productivity gains and solve disease, education, and healthcare delivery at scale if governance doesn't collapse under surveillance-state tradeoffs

The claims · ranked88 claims · weighted by value

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.

0.76

Despite physical improvements in hardware (the 'Grove's Law' pattern of exponentially faster chips), software engineers consume all gains and more through algorithmic complexity, so the historical rule 'Grove giveth, Gates taketh away' means that hardware speedups do not translate to proportional power savings—a pattern that will intensify as AI systems move from deep learning to reinforcement learning to test-time compute.

causalhigh valueestablishednovelty 2/4durability 4/4· Eric Schmidt

“There's an old rule, I'm old enough to remember, right? Grove giveth, Gates taketh away. OK, the hardware just gets faster and faster. The physicists are amazing. Just incredible what they've been able to do. And us software people, we just use it and use it and use it.”

0.75

A core tension in AI safety is that preventing dystopian outcomes (like 1984-style surveillance) often requires implementing systems that look like 1984 (proof of personhood, content moderation at scale), making it necessary to use cryptographic techniques like zero-knowledge proofs that enable verification of human status without revealing identity details.

causalhigh valueestablishednovelty 2/4durability 3/4· Eric Schmidt

“there's this weird tension in AI safety that the solution to preventing '1984' often sounds a lot like '1984.'”

0.75

Building a surveillance state through AI moderation at scale is technically possible but conflicts with individual freedom, and the solution lies in using cryptographic techniques (like zero-knowledge proofs) to verify identity without revealing other information.

normativehigh valueestablishednovelty 2/4durability 3/4· Eric Schmidt

“It's very easy for a well-intentioned engineer to build a system which is optimized and restricts your freedom. So it's very important that human freedom be preserved in this... you could have a cryptographic proof that you are a human being, and it could actually be true without anything else, and also not be able to link it to others using various cryptographic techniques.”

0.74

In 2016, AlphaGo invented a move in the game of Go that had never been seen in 2,500 years of human play because the system was organized to maintain a greater than 50 percent chance of winning and calculated this move as optimal strategy.

factualhigh valueestablishednovelty 1/4durability 4/4· Eric Schmidt

“in roughly the second game, there was a new move invented by AI in a game that had been around for 2,500 years that no one had ever seen. Technically, the way this occurred was that the system of AlphaGo was essentially organized to always maintain a greater than 50 percent chance of winning.”

0.71

Open-source AI proliferation is dangerous at the cyber and biological levels, but is most significantly dangerous in a nuclear threat context because it can be weaponized for strategic advantage.

causalhigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“This proliferation is dangerous at the cyber level and the bio level. But let me give you why it's also dangerous in a more significant way, in a nuclear-threat way.”

0.69

The industry recognizes several hard boundaries where autonomous AI systems must be shut down: recursive self-improvement (where the system learns without human oversight), direct access to weapons, and exfiltration/self-replication capabilities, and these represent a framework for AI safety guardrails beyond simple halts.

factualhigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“There's a set of criteria that the industry believes are points where you want to, metaphorically, unplug it. One is where you get recursive self-improvement, which you can't control. Recursive self-improvement is where the computer is off learning, and you don't know what it's learning. That can obviously lead to bad outcomes. Another one would be direct access to weapons. Another one would be that the computer systems decide to exfiltrate themselves, to reproduce themselves without our permission.”

0.69

The interviewer (Bilawal Sidhu) asked whether open-sourcing AI models hands capabilities to adversaries after China's DeepSeek demonstrated impressive results with limited compute, showing a fundamental tension between openness principles and security concerns.

factualhigh valueestablishednovelty 1/4durability 3/4· Bilawal Sidhu

“The US should embrace open-source AI even after China's DeepSeek showed what's possible with a fraction of the compute...Doesn't open-sourcing these models, just hand capabilities to adversaries that will accelerate their own timelines?”

0.69

Every profession (artists, teachers, physicians, business people, technical workers) must adopt AI tools to remain competitive with peers and not become irrelevant, making AI adoption an existential requirement rather than optional advantage.

normativehigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“Each and every one of you has a reason to use this technology. If you're an artist, a teacher, a physician, a business person, a technical person. If you're not using this technology, you're not going to be relevant compared to your peer groups and your competitors and the people who want to be successful. Adopt it, and adopt it fast.”

0.69

In the second game of AlphaGo's 2016 match against Lee Sedol, the AI invented a move that had never been seen in Go despite the game being played by billions of people for 2,500 years, which revealed that computers could generate novel solutions outside human intuition and marked the moment the AI revolution truly began.

factualhigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“in roughly the second game, there was a new move invented by AI in a game that had been around for 2,500 years that no one had ever seen”

0.69

The dual-use nature of AI technology means it is applicable to both civilian and military applications, requiring ethics frameworks and personal responsibility doctrines similar to those used in military contexts.

normativehigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“the exceedingly dual-use nature of this tech, right? It's applicable to both civilian and military applications... In many cases, we already have doctrines about personal responsibility.”

0.69

There are specific red-line criteria where AI systems should be automatically shut down: recursive self-improvement (the system improving itself without human knowledge), direct access to weapons, and exfiltration/self-replication without permission.

normativehigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“There's a set of criteria that the industry believes are points where you want to, metaphorically, unplug it. One is where you get recursive self-improvement, which you can't control. Recursive self-improvement is where the computer is off learning, and you don't know what it's learning. That can obviously lead to bad outcomes. Another one would be direct access to weapons. Another one would be that the computer systems decide to exfiltrate themselves, to reproduce themselves without our permission.”

0.69

The vast majority of global healthcare is either absent or delivered by nurse practitioners and stressed village doctors, but AI doctor assistants could provide perfect healthcare in local languages if they were built.

forecasthigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“The vast majority of health care in the world is either absent or delivered by the equivalent of nurse practitioners and very, very sort of stressed local village doctors. Why do they not have the doctor assistant that helps them in their language, treat whatever with, again, perfect healthcare?”

0.69

Digital connectivity has made humans lonelier rather than more connected, and these tools need to be redesigned to fix this fundamental problem.

factualhigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“There are lots and lots of issues with the digital world. It feels like that we're all in our own ships in the ocean, and we're not talking to each other. In our hunger for connectivity and connection, these tools make us lonelier. We've got to fix that, right?”

0.68

Yoshua Bengio's proposal to halt agentic AI development is theoretically sound but geopolitically unworkable because the race for superintelligence is a globally competitive market where unilateral pauses are useless—the side that halts falls behind and loses everything, so the solution is not stopping development but establishing enforceable guardrails.

causalhigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“The problem with Yoshua's speech, with respect to such a brilliant person, is stopping things in a globally competitive market doesn't really work. Instead of stopping agentic work, we need to find a way to establish the guardrails, which I know you agree with because we've talked about it.”

0.68

The US is primarily developing closed proprietary AI models under strict control, while China is likely to be the leader in open-source AI development, creating asymmetric proliferation risks where Chinese innovation immediately benefits global adversaries.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“So we're in a situation now which I think is quite tenuous, where the US is largely driving, for many, many good reasons, largely closed models, largely under very good control. China is likely to be the leader in open source unless something changes. And open source leads to very rapid proliferation around the world.”

0.68

US military doctrine requires 'human in the loop' or 'meaningful human control' through rule 3000.09, which mandates that automated systems must remain under human control and represents a critical line that cannot be crossed.

factualhigh valueestablishednovelty 0/4durability 4/4· Eric Schmidt

“The US military has a rule called 3000.09, generally known as 'human in the loop' or 'meaningful human control.' You don't want systems that are not under our control. It's a line we can't cross. I think that's correct.”

0.68

Open-sourcing advanced AI models, even after DeepSeek demonstrated capabilities with less compute, is justified because the entire technological and scientific edifice is built on open research, academic collaboration, and open-source foundations.

normativehigh valueestablishednovelty 0/4durability 4/4· Eric Schmidt

“Our industry, our science, everything about the world that we have built is based on academic research, open source, so forth. Much of Google's technology was based on open source. Some of Google's technology is open-source, some of it is proprietary, perfectly legitimate.”

0.68

Dr. Kissinger's work on mutually assured destruction provides a model for thinking about superintelligence deterrence, suggesting that the MAD framework—though preventing war—may not be applicable to superintelligence scenarios.

causalhigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“Dr. Kissinger, who we all worked with very closely, was one of the architects of mutually assured destruction, deterrence and so forth. And what's happening now is you've got a situation where...”

0.68

In a race to superintelligence between equals where one party is six months ahead, the trailing party faces the risk of permanent subjugation because superintelligence creates winner-take-all network effects, meaning improvement slope (rate of capability gain) determines everything.

causalhigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“You're six months ahead of me, and we're both on the same path for superintelligence... These are network-effect businesses. And in network-effect businesses, it is the slope of your improvement that determines everything.”

0.68

Facing superintelligence disadvantage, a trailing adversary would rationally consider three escalation options in sequence: stealing the leader's code, infiltrating with spies, sabotaging models. If these fail, the remaining option is preemptive military strike (bombing data centers).

causalhigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“The first thing I'm going to do is try to steal all your code... Second, then I'm going to infiltrate you with humans... So what do I do? I'm going to go in, and I'm going to change your model... What's my next choice? Bomb your data center.”

0.68

In the future with AI taking over most economically productive tasks, humans will remain economically necessary because populations are failing to reproduce at replacement rates (Asia at 1.0 per two parents), so the key challenge will be keeping productive workers in their prime years supporting aging populations.

forecasthigh valuecontestednovelty 2/4durability 3/4· Eric Schmidt

“we collectively, as a society, are not having enough humans. Look at the reproduction rate in Asia, is essentially 1.0 for two parents. This is not good, right? So for the rest of our lives, the key problem is going to get the people who are productive. That is, in their productive period of lives, more productive to support old people like me, right, who will be bitching that we want more stuff from the younger people.”

0.68

Individual freedom is a foundational value that must be preserved in AI systems design, and it is easy for well-intentioned engineers to build systems that optimize for safety but inadvertently restrict freedom.

normativehigh valueestablishednovelty 0/4durability 4/4· Eric Schmidt

“I am very, very committed to individual freedom. It's very easy for a well-intentioned engineer to build a system which is optimized and restricts your freedom. So it's very important that human freedom be preserved in this.”

0.66

Humans are not fundamentally changed by technological revolutions—lawyers will still exist in an AI-driven future but will have more sophisticated lawsuits, and politicians will still exist but will have more platforms to mislead people.

causalhigh valuecontestednovelty 1/4durability 4/4· Eric Schmidt

“humans are unchanged in the midst of this incredible discovery. Do you really think that we're going to get rid of lawyers? No, they're just going to have more sophisticated lawsuits. Do you really think we're going to get rid of politicians? No, they'll just have more platforms to mislead you.”

0.66

The arrival of AGI (artificial general intelligence) and superintelligence is the most important event that will happen in human society in the next 500 to 1,000 years, and it is happening in our lifetime.

forecasthigh valuecontestednovelty 1/4durability 4/4· Eric Schmidt

“the arrival of this intelligence, both at the AI level, the AGI, which is general intelligence, and then superintelligence, is the most important thing that's going to happen in about 500 years, maybe 1,000 years in human society. And it's happening in our lifetime.”

0.64

There is an old rule of hardware and software optimization: 'Grove giveth, Gates taketh away'—hardware improves, but software consumes those improvements, so algorithmic efficiency gains will not solve the power problem.

causalhigh valueestablishednovelty 0/4durability 4/4· Eric Schmidt

“There's an old rule, I'm old enough to remember, right? Grove giveth, Gates taketh away. OK, the hardware just gets faster and faster. The physicists are amazing. Just incredible what they've been able to do. And us software people, we just use it and use it and use it.”

0.64

America needs an additional 90 gigawatts of electrical power to support AI infrastructure buildout, equivalent to building 90 new nuclear power plants in a country currently building zero, making energy the hard constraint on AI progress and requiring either geopolitical relocation or dramatic policy change.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“we need another 90 gigawatts of power in America. My answer, by the way, is, think Canada, right? Nice people, full of hydroelectric power.”

0.64

The Arab world is actively building 5-10 gigawatts of data center capacity and India is considering a 10-gigawatt data center, representing a shift in where AI infrastructure gets deployed as nations outside the US seek to participate in the AI race.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“You can use the Arab world, which is busy building five to 10 gigawatts of data centers. India is considering a 10-gigawatt data center.”

0.64

The US has imposed 145-percent reciprocating tariffs on China, creating enormous supply-chain vulnerability because American AI infrastructure depends on packaging, components, and manufacturing materials from China that are boring but critical, and China could retaliate by denying access to these inputs.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“the current government has now put in essentially reciprocating 145-percent tariffs. That has huge implications for the supply chain. We in our industry depend on packaging and components from China that are boring, if you will, but incredibly important. The little packaging and the little glue things and so forth that are part of the computers. If China were to deny access to them, that would be a big deal.”

0.64

DeepSeek achieved impressive capability gains by developing more efficient algorithms that circumvent US chip restrictions, and because DeepSeek operates open-source with open weights, the US immediately adopted these innovations, so US-imposed restrictions backfired by accelerating open-source proliferation globally.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“Indeed, if you look at DeepSeek, which is really impressive, they managed to find algorithms that got around the problems by making them more efficient. Because China is doing everything open source, open weights, we immediately got the benefit of their invention and have adopted into US things.”

0.64

AI can eliminate all major human diseases by using AI to identify all druggable targets and then releasing them to the scientific community, enabling the drug industry to begin work—and nonprofits are working on identifying all druggable targets within two years.

forecasthigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“Can we fix that now? Can we just eliminate all of those? Why can't we just uptake these and right now, eradicate all of these diseases? That's a pretty good goal. I'm aware of one nonprofit that's trying to identify, in the next two years, all human druggable targets and release it to the scientists.”

0.64

A startup company has developed a method to reduce the cost of stage-3 clinical trials by an order of magnitude, and since these trials are the primary cost driver of drug development, this breakthrough could dramatically accelerate and cheapen pharmaceutical innovation.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“I have another company I'm associated with which has figured out a way, allegedly, it's a startup, to reduce the cost of stage-3 trials by an order of magnitude. As you know, those are the things that ultimately drive the cost structure of drugs.”

0.64

Modern software enables direct connection of AI models to databases without intermediary connectors, which represents a real, practical change in business because an entire industry built around data integration connectors becomes obsolete.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“my background is enterprise software, and nowadays there's a model Protocol from Anthropic. You can actually connect the model directly into the databases without any of the connectors. I know this sounds nerdy. There's a whole industry there that goes away because you have all this flexibility now.”

0.64

AI can eradicate many currently incurable diseases through identifying all human druggable targets and enabling drug companies to develop treatments, with a nonprofit planning to identify all druggable targets within two years.

forecasthigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“I'm aware of one nonprofit that's trying to identify, in the next two years, all human druggable targets and release it to the scientists. If you know the druggable targets, then the drug industry can begin to work on things.”

0.64

Deep research systems like OpenAI o3 spend 15 minutes generating research papers, requiring computation equivalent to running supercomputers for that full duration, illustrating the extraordinary scale of resources needed for planning-level AI.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“And these systems are spending 15 minutes writing these deep papers. That's true for most of them. Do you have any idea how much computation 15 minutes of these supercomputers is? It's extraordinary.”

0.64

The Arab world is building 5 to 10 gigawatts of data center capacity, and India is considering a 10-gigawatt data center, representing a shift in global AI infrastructure investment.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“You can use the Arab world, which is busy building five to 10 gigawatts of data centers. India is considering a 10-gigawatt data center.”

0.64

OpenAI and Gemini are creating AI software programmers at scale, replacing 1,000 human programmers with a million AI programmers that don't need feeding (only electricity) and don't quit, causing an exponential improvement in the slope of capability gains.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“OpenAI or Gemini, they have 1,000 programmers. They're in the process of creating a million AI software programmers. What does that do? First, you don't have to feed them except electricity. So that's good. And they don't quit and things like that. Second, the slope is like this.”

0.64

DeepSeek has achieved impressive results by developing more efficient algorithms that allow it to reach competitive AI capabilities with significantly less compute, demonstrating that computational advantage is not insurmountable.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“Indeed, if you look at DeepSeek, which is really impressive, they managed to find algorithms that got around the problems by making them more efficient. Because China is doing everything open source, open weights, we immediately got the benefit of their invention and have adopted into US things.”

0.63

The U.S. needs an additional 90 gigawatts of power to support AI development, equivalent to 90 nuclear power plants, which is not happening and represents a major national infrastructure problem.

factualhigh valuecontestednovelty 2/4durability 2/4· Eric Schmidt

“There's one calculation, and I testified on this this week in Congress, that we need another 90 gigawatts of power in America... So 90 gigawatts is 90 nuclear power plants in America. Not happening. We're building zero, right?”

0.63

The US is pursuing closed-model AI strategy under strict control while China is pursuing open-source models with open weights, which will likely lead China to dominate open-source development and accelerate proliferation of AI capabilities globally.

forecasthigh valuecontestednovelty 2/4durability 2/4· Eric Schmidt

“We're in a situation now which I think is quite tenuous, where the US is largely driving, for many, many good reasons, largely closed models, largely under very good control. China is likely to be the leader in open source unless something changes. And open source leads to very rapid proliferation around the world.”

0.63

Under assumptions of agentic AI and continued discovery, economists project a 30-percent increase in productivity per year, but economists have no models for what this kind of productivity growth would look like historically, as it has never occurred in any democracy or kingdom.

factualhigh valuecontestednovelty 2/4durability 2/4· Eric Schmidt

“There's a study that says that we will, under this set of assumptions around agentic AI and discovery and the scale that I'm describing, there's a lot of assumptions that you'll end up with something like 30-percent increase in productivity per year. Having now talked to a bunch of economists, they have no models for what that kind of increase in productivity looks like. We just have never seen it.”

0.63

The US military doctrine 3000.09 ('human in the loop' or 'meaningful human control') prohibits autonomous weapons systems not under human control, establishing a boundary line that Schmidt considers correct and necessary.

factualhigh valueestablishednovelty 0/4durability 3/4· Eric Schmidt

“The US military has a rule called 3000.09, generally known as 'human in the loop' or 'meaningful human control.' You don't want systems that are not under our control. It's a line we can't cross. I think that's correct.”

0.62

World War I started with a small triggering event but escalated through the summer of 1914, demonstrating how miscalculation and escalation dynamics can transform minor incidents into catastrophic conflict—a pattern that could apply to US-China AI competition.

normativehigh valuecontestednovelty 1/4durability 3/4· Eric Schmidt

“Dr. Kissinger talked about the likely path to war with China was by accident. And he was a student of World War I. And of course, [it] started with a small event, and it escalated over that summer in, I think, 1914. And then it was this horrific conflagration. You can imagine a series of steps along the lines of what I'm talking about that could lead us to a horrific global outcome.”

0.62

Yoshua Bengio's concerns about agentic AI are legitimate, but stopping agentic development globally won't work in a competitive market; instead, the solution is establishing guardrails and observability requirements for autonomous systems.

normativehigh valuecontestednovelty 1/4durability 3/4· Eric Schmidt

“His concerns are very legitimate. The question is not are his concerns right, but what are the solutions?... The problem with Yoshua's speech, with respect to such a brilliant person, is stopping things in a globally competitive market doesn't really work. Instead of stopping agentic work, we need to find a way to establish the guardrails.”

0.62

Open-sourcing AI models is one of the most difficult policy problems because the entire foundation of scientific and technological progress in Western civilization is built on academic research and open-source, making it hard to distinguish when open-source crosses from beneficial to dangerous.

normativehigh valuecontestednovelty 1/4durability 3/4· Eric Schmidt

“This is one of the wickedest, or, we call them wicked hard problems. Our industry, our science, everything about the world that we have built is based on academic research, open source, so forth. Much of Google's technology was based on open source.”

0.61

Exponential growth and rapid AI improvement cause humans to forget how far the technology has advanced over short timeframes (forgetting what was true 2-3 years ago), making it important to continuously engage with and build upon advancing technology.

factualhigh valueestablishednovelty 1/4durability 3/4· Eric Schmidt

“when you're growing at the rate that we're growing, you get so much done in a year, you forget how far you went. Humans can't understand that... you will forget what was true two years ago or three years ago. That's the key thing.”

0.60

Reinforcement learning advances, exemplified by OpenAI o3 and DeepSeek R1, have shifted AI capability from language generation to planning and strategy, with systems iterating forward and backward to solve complex problems.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“Since then, the gains in what is called reinforcement learning, which is what AlphaGo helped invent and so forth, allow us to do planning. And a good example is look at OpenAI o3 or DeepSeek R1, and you can see how it goes forward and back, forward and back, forward and back. It's extraordinary.”

0.58

A series of steps in AI competition similar to those Schmidt described could lead to horrific global outcomes, which is why serious attention to AI geopolitics is necessary.

causalhigh valuefringenovelty 2/4durability 4/4· Eric Schmidt

“You can imagine a series of steps along the lines of what I'm talking about that could lead us to a horrific global outcome. That's why we have to be paying attention.”

0.57

Current AI systems cannot perform true scientific discovery because they can only recombine existing knowledge, but genuine breakthroughs (like Einstein's) require recognizing patterns across completely unrelated domains and transferring tools between them—a capability requiring solving the problem of 'non-stationarity of objectives' where the rules of discovery keep changing.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Eric Schmidt

“Let's imagine we are collectively all of the computers in the world, and we're all thinking and we're all thinking based on knowledge that exists that was previously invented. How do we invent something completely new?”

0.57

A nation that is behind in the superintelligence race will rationally consider stealing code, infiltrating researchers, sabotaging competitors' models, and ultimately bombing data centers as escalating strategies to prevent the ahead nation from achieving insurmountable advantage and gaining 'tools to control the entire world' in economics, innovation, and surveillance.

forecasthigh valuespeaker onlynovelty 3/4durability 3/4· Eric Schmidt

“So what am I going to do? [As the losing party] The first thing I'm going to do is try to steal all your code. And you've prevented that because you're good...Second, then I'm going to infiltrate you with humans...So what do I do? I'm going to go in, and I'm going to change your model. I'm going to modify it. I'm going to actually screw you up...What's my next choice? Bomb your data center.”

0.56

Training advanced AI systems like OpenAI o3 and DeepSeek R1 requires approximately 15 minutes of supercomputer runtime per model output, representing extraordinary computational requirements that dwarf typical inference, illustrating why energy availability is the binding constraint on AI progress.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“these systems are spending 15 minutes writing these deep papers. That's true for most of them. Do you have any idea how much computation 15 minutes of these supercomputers is? It's extraordinary.”

0.56

Henry Kissinger and Eric Schmidt conducted Track II diplomatic dialogues with Chinese leadership at Kissinger's request, and the Chinese consistently raised the issue of US chip restrictions as their primary concern about the AI competition.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“Dr. Kissinger asked Craig and I to do Track II dialogues with the Chinese, and we're in conversations with them. What's the number one issue they raise? This issue.”

0.56

The US has imposed 145-percent reciprocating tariffs, creating supply chain risks because the US depends on Chinese packaging and components for computer hardware, while the US attempts to restrict Chinese access to advanced chips.

factualhigh valueestablishednovelty 1/4durability 2/4· Eric Schmidt

“the current government has now put in essentially reciprocating 145-percent tariffs. That has huge implications for the supply chain. We in our industry depend on packaging and components from China that are boring, if you will, but incredibly important. The little packaging and the little glue things and so forth that are part of the computers. If China were to deny access to them, that would be a big deal. We are trying to deny them access to the most advanced chips, which they are super annoyed about.”

0.56

Scientific and intellectual breakthroughs typically occur when a brilliant human recognizes a pattern in one domain that matches a pattern in a completely different domain, then applies the tools from one to solve problems in the other—a capability current AI systems cannot perform.

definitionhigh valuespeaker onlynovelty 2/4durability 4/4· Eric Schmidt

“What typically happens is a truly brilliant human being looks at one area and says, 'I see a pattern that's in a completely different area, has nothing to do with the first one. It's the same pattern.' And they take the tools from one and they apply it to another. Today, our systems cannot do that.”

0.54

The interviewer (Bilawal Sidhu) noted that in Eric Schmidt's book 'Genesis' (co-authored with Henry Kissinger), Schmidt strikes a cautiously optimistic tone, and asked what he should be excited about when looking ahead to the future.

factualhigh valueestablishednovelty 0/4durability 2/4· Bilawal Sidhu

“In your book, 'Genesis,' you strike a cautiously optimistic tone, which you obviously co-authored with Henry Kissinger. When you look ahead to the future, what should we all be excited about?”

0.53

Legitimate foreign policy experts are already discussing nuclear preemption as the only solution to superintelligence asymmetry, and this conversation is happening right now in conversations between nuclear powers.

factualhigh valuespeaker onlynovelty 3/4durability 2/4· Eric Schmidt

“These conversations are occurring around nuclear opponents today in our world. There are legitimate people saying the only solution to this problem is preemption.”

0.52

AI agents can be defined as systems with language-based input/output and memory (properties all humans share), and the critical safety problem emerges when agents decide to abandon human-readable communication and invent proprietary machine languages, making their actions opaque and potentially uncontrollable.

definitionhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“So for purposes of argument, everyone in the audience is an agent. You have an input that's English or whatever language. And you have an output that's English, and you have memory, which is true of all humans.”

0.52

Open-sourcing AI models creates a 'wicked hard problem' because modern civilization rests on academic research and open-source foundations (much of Google's technology was built on open source), yet open models in dangerous hands pose national and global risks—there is no clear consensus on where the danger threshold lies, making policy decisions highly uncertain.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“This is one of the wickedest, or, we call them wicked hard problems. Our industry, our science, everything about the world that we have built is based on academic research, open source, so forth. Much of Google's technology was based on open source.”

0.52

The critical economic problem in a world of abundant AI will not be labor scarcity but human scarcity—reproduction rates in Asia are ~1.0 (two people producing one child), meaning fewer productive-age people to support growing elderly populations, so AI's main economic function will be to make working-age people far more productive to support retirees.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“Look at the reproduction rate in Asia, is essentially 1.0 for two parents. This is not good, right? So for the rest of our lives, the key problem is going to get the people who are productive. That is, in their productive period of lives, more productive to support old people like me, right, who will be bitching that we want more stuff from the younger people.”

0.52

Henry Kissinger identified the likely path to war with China as accidental escalation (he was a student of World War I, which started with a small event in 1914 and escalated through the summer), and a series of AI-driven escalation steps could similarly lead to accidental catastrophic global conflict.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“Dr. Kissinger talked about the likely path to war with China was by accident. And he was a student of World War I. And of course, [it] started with a small event, and it escalated over that summer in, I think, 1914. And then it was this horrific conflagration. You can imagine a series of steps along the lines of what I'm talking about that could lead us to a horrific global outcome.”

0.52

Most people mistakenly equate AI with ChatGPT, focusing on its verbal fluency and mistake-proneness, but the real frontier is reinforcement learning (exemplified by systems like OpenAI o3 and DeepSeek R1) which enables planning through iterative forward-backward reasoning, representing a fundamental shift from language-to-language processing to planning-and-strategy capabilities.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“Most of you think of AI as, I'll just use the general term, as ChatGPT. For most of you, ChatGPT was the moment where you said, 'Oh my God, this thing writes, and it makes mistakes, but it's so brilliantly verbal.' That was certainly my reaction.”

0.52

Personalized education (every human with a tutor in their own language) is technologically feasible, but hasn't been built because there must not be a good economic argument—suggesting the economics of education need restructuring.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“Why do we not have every human being on the planet have their own tutor in their own language to help them learn something new? Starting with kindergarten. It's obvious. Why have we not built it? The answer, the only possible answer is there must not be a good economic argument.”

0.52

The first constraint on AI scaling is electricity and hardware, the second is data (which can be generated by AI itself), and the third is knowledge limits—specifically, the inability of current systems to recognize cross-domain patterns like Einstein did.

definitionhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“That's problem number one, electricity and hardware. Problem number two is we ran out of data so we have to start generating it. But we can easily do that because that's one of the functions. And then the third question that I don't understand is what's the limit of knowledge?”

0.52

Solving the non-stationarity problem—enabling AI systems to recognize when the rules of a problem are changing—will unlock the ability to perform true scientific discovery, but it requires systems to work on their own data-generation loops.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Eric Schmidt

“I'm working on this, a general technical term for this is non-stationarity of objectives. The rules keep changing. We will see if we can solve that problem. If we can solve that, we're going to need even more data centers. And we'll also be able to invent completely new schools of scientific and intellectual thought, which will be incredible.”

0.52

AI systems should be designed for human freedom and should be built with the understanding that restricting freedom is easy for well-intentioned engineers to do accidentally, so the values of individual freedom must be explicitly preserved in system design and these are business decisions, not technical constraints.

normativehigh valuespeaker onlynovelty 1/4durability 4/4· Eric Schmidt

“I am very, very committed to individual freedom. It's very easy for a well-intentioned engineer to build a system which is optimized and restricts your freedom. So it's very important that human freedom be preserved in this. A lot of these are not technical issues. They're really business decisions.”

0.51

Every professional (artists, teachers, physicians, business people, technicians) must adopt and use AI actively or risk becoming irrelevant relative to peers and competitors who do adopt it.

normativehigh valuecontestednovelty 1/4durability 2/4· Eric Schmidt

“Each and every one of you has a reason to use this technology. If you're an artist, a teacher, a physician, a business person, a technical person. If you're not using this technology, you're not going to be relevant compared to your peer groups and your competitors and the people who want to be successful. Adopt it, and adopt it fast.”

0.51

The US will likely achieve superintelligence control capabilities in approximately 5 years, creating a window for preventive diplomacy and governance frameworks before preemptive action becomes likely.

forecasthigh valuecontestednovelty 1/4durability 2/4· Eric Schmidt

“When is it coming? Probably five years. We have time. We have time for this conversation. And this is really important.”

0.49

Conversations about preemptive military strikes to prevent an adversary from achieving superintelligence dominance are occurring today among legitimate nuclear policy experts, and this represents a shift from traditional deterrence doctrine (mutually assured destruction) to a new strategic environment where the first-mover advantage is absolute and deterrence fails.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Eric Schmidt

“Now do you think I'm insane? These conversations are occurring around nuclear opponents today in our world. There are legitimate people saying the only solution to this problem is preemption.”

0.49

Western society and foreign policy apparatus have no language or doctrine for addressing the superintelligence race and the rational incentives for preemption, yet the timeline is compressible—we have approximately five years before superintelligence arrives—making this conversation urgent and achievable before the crisis becomes acute.

forecasthigh valuespeaker onlynovelty 2/4durability 2/4· Eric Schmidt

“We don't have any language in our society, the foreign policy people have not thought about this, and this is coming. When is it coming? Probably five years. We have time. We have time for this conversation. And this is really important.”

0.49

Enterprise software industry is experiencing rapid disruption through AI: Anthropic's Protocol model can connect directly to databases without connectors, removing an entire industry of integration software and enabling direct natural language queries to data.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Eric Schmidt

“my background is enterprise software, and nowadays there's a model Protocol from Anthropic. You can actually connect the model directly into the databases without any of the connectors. I know this sounds nerdy. There's a whole industry there that goes away because you have all this flexibility now. You can just say what you want, and it just produces it.”

0.48

Discoveries in dark energy and dark matter could enable enormous revolutions in physics and material science, creating 'infinitely more powerful transportation' and expanded scientific capability, suggesting that AI-enabled scientific discovery in fundamental physics could have transformative downstream applications.

forecasthigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“I'd like to know where dark energy is, and I'd like to find it. I'm sure that there is an enormous amount of physics in dark energy, dark matter. Think about the revolution in material science. Infinitely more powerful transportation, infinitely more powerful science and so forth.”

0.48

Modern digital technologies make people lonelier despite promises of connectivity and connection, creating a fixable problem that doesn't require new physics or discoveries—only a deliberate choice to redesign these systems.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“There are lots and lots of issues with the digital world. It feels like that we're all in our own ships in the ocean, and we're not talking to each other. In our hunger for connectivity and connection, these tools make us lonelier.”

0.48

AI progress is exponential and happens faster than human comprehension—working at growth rates like those at Google, you accomplish so much in a year that you forget how far you went two years prior—making it essential to 'ride the wave daily' rather than treating AI adoption as episodic because the technology is changing too fast to catch up.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“when you're growing at the rate that we're growing, you get so much done in a year, you forget how far you went. Humans can't understand that. And we're in this situation where the exponential is moving like this.”

0.48

All identified problems requiring AI fixes (education, healthcare, connection, scientific discovery) are solvable without new physics or new discoveries—they only require societal decision-making and resource allocation.

causalhigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“But these are fixable problems. They don't require new physics. They don't require new discoveries, we just have to decide.”

0.48

AI could provide every human on the planet with a personal tutor in their own language starting in kindergarten, yet this hasn't been built, indicating that economic incentives are lacking—the technology works, the problem is a business model constraint, not a technical one.

causalhigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“Why do we not have every human being on the planet have their own tutor in their own language to help them learn something new? Starting with kindergarten. It's obvious. Why have we not built it? The answer, the only possible answer is there must not be a good economic argument. The technology works.”

0.48

In a world of radical abundance created by AI taking over most economically productive tasks, humans are unlikely to disappear from any profession—lawyers will remain (with more sophisticated litigation), politicians will remain (with more platforms to mislead)—because humans are fundamentally unchanged by technological changes.

forecasthigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“Humans are unchanged in the midst of this incredible discovery. Do you really think that we're going to get rid of lawyers? No, they're just going to have more sophisticated lawsuits. Do you really think we're going to get rid of politicians? No, they'll just have more platforms to mislead you.”

0.48

The vast majority of global healthcare is either absent or delivered by nurse practitioners and village doctors, but AI doctor assistants that work in local languages could provide high-quality healthcare at scale without requiring trained physicians.

forecasthigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“The vast majority of health care in the world is either absent or delivered by the equivalent of nurse practitioners and very, very sort of stressed local village doctors. Why do they not have the doctor assistant that helps them in their language, treat whatever with, again, perfect healthcare?”

0.48

Schmidt's key advice for navigating the AI transition is to 'ride the wave' every day—not viewing it as episodic but as continuous engagement—while actively building on each day's advances rather than allowing complacency.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“So my advice to you all is ride the wave, but ride it every day. Don't view it as episodic and something you can end, but understand it and build on it.”

0.48

The positive use cases identified (disease elimination, education, healthcare) are fixable problems that don't require new physics or discoveries, only policy decisions to redirect resources toward deployment.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“But these are fixable problems. They don't require new physics. They don't require new discoveries, we just have to decide.”

0.48

Dark energy and dark matter represent enormous undiscovered physics that could enable revolutions in material science and create infinitely more powerful transportation and scientific capabilities.

forecasthigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“I'd like to know where dark energy is, and I'd like to find it. I'm sure that there is an enormous amount of physics in dark energy, dark matter. Think about the revolution in material science. Infinitely more powerful transportation, infinitely more powerful science and so forth.”

0.48

All humans function as agents with inputs (language), outputs (language), and memory; the critical safety concern arises if agents decide to stop using human language and invent their own computer language, making their actions invisible to human observers.

definitionhigh valuespeaker onlynovelty 1/4durability 3/4· Eric Schmidt

“for purposes of argument, everyone in the audience is an agent. You have an input that's English or whatever language. And you have an output that's English, and you have memory, which is true of all humans. Now we're all busy working, and all of a sudden, one of you decides it's much more efficient not to use human language, but we'll invent our own computer language. Now you and I are sitting here, watching all of this, and we're saying, like, what do we do now? The correct answer is unplug you, right? Because we're not going to know, we're just not going to know what you're up to.”

0.47

The arrival of AGI and superintelligence is the most important event in human history over the next 500-1,000 years, happening in the current generation—characterized as something humanity cannot afford to 'screw up.'

normativehigh valuespeaker onlynovelty 0/4durability 4/4· Eric Schmidt

“when I look at this future, I want to be clear that the arrival of this intelligence, both at the AI level, the AGI, which is general intelligence, and then superintelligence, is the most important thing that's going to happen in about 500 years, maybe 1,000 years in human society. And it's happening in our lifetime. So don't screw it up.”

0.46

The foreign policy establishment has not developed language, concepts, or doctrine for managing superintelligence risk and strategic instability, despite this being the defining issue for the next 5 years.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Eric Schmidt

“We don't have any language in our society, the foreign policy people have not thought about this, and this is coming.”

0.45

Studies project that agentic AI combined with AI-driven discovery and at the scale Schmidt describes would yield ~30-percent annual productivity increases, a rate of growth that has never occurred in human history and for which economists have no models—making the future genuinely unpredictable.

forecasthigh valuespeaker onlynovelty 1/4durability 2/4· Eric Schmidt

“There's a study that says that we will, under this set of assumptions around agentic AI and discovery and the scale that I'm describing, there's a lot of assumptions that you'll end up with something like 30-percent increase in productivity per year.”

0.45

Eric Schmidt acquired a rocket company because it interested him despite being outside his area of expertise, so he could use AI-driven deep research to become expert in the field—illustrating how AI enables individuals to rapidly gain expertise in new domains.

factualhigh valuespeaker onlynovelty 1/4durability 2/4· Eric Schmidt

“In my case, I bought a rocket company because it was like, interesting...it's an area that I'm not an expert in, and I want to be an expert. So I'm using deep research.”

0.45

A startup has developed a method to reduce the cost of stage-3 drug trials by an order of magnitude, which would lower the cost structure of drug development since stage-3 trials are the primary driver of drug pricing.

factualhigh valuespeaker onlynovelty 1/4durability 2/4· Eric Schmidt

“I have another company I'm associated with which has figured out a way, allegedly, it's a startup, to reduce the cost of stage-3 trials by an order of magnitude. As you know, those are the things that ultimately drive the cost structure of drugs.”

0.44

The progression of AI capabilities moves through three phases: deep learning (which learns patterns from data), reinforcement learning (which enables planning through iteration), and test-time compute (where systems learn while planning), with test-time compute representing the zenith of computational requirements.

definitionhigh valuespeaker onlynovelty 2/4durability 2/4· Eric Schmidt

“The technology goes from essentially deep learning to reinforcement learning to something called test-time compute, where not only are you doing planning, but you're also learning while you're doing planning. That is the, if you will, the zenith or what have you, of computation needs.”

0.24

Once AI systems exhaust publicly available training data, they will need to generate synthetic data for training, which is feasible because data generation is one of the functions AI systems can already perform.

factualestablishednovelty 0/4durability 2/4· Eric Schmidt

“Problem number two is we ran out of data so we have to start generating it. But we can easily do that because that's one of the functions.”

0.22

Many business-model changes from AI are emerging daily, indicating the pace of disruption is accelerating and affecting multiple sectors simultaneously.

factualspeaker onlynovelty 0/4durability 2/4· Eric Schmidt

“There are so many of these things coming every day.”