The AI Revolution Is Underhyped | Eric Schmidt

1. The “Underhyped” Stance

• Why Underhyped? Schmidt argues most people equate AI with chatbots (like ChatGPT), which impressed everyone with language skills. However, the real revolution is the shift from Language to Planning and Reasoning (via Reinforcement Learning, similar to AlphaGo).
• The New Frontier: Newer models (like OpenAI o3 or DeepSeek R1) can “think” for extended periods (e.g., 15 minutes of compute time) to solve complex problems, write deep research papers, or develop strategies. This moves AI from just generating text to executing complex workflows and scientific discovery.

2. The Three Constraints

Schmidt identifies three major bottlenecks to AI scaling:

• Energy: The US needs ~90 Gigawatts of new power (equivalent to 90 nuclear plants), which is politically and logistically difficult. Other regions (Arab world, India) are building massive 5-10 GW data centers.
• Data: We are running out of public human data, so AI must start generating its own synthetic data for training.
• Knowledge Limit: Current AI aggregates existing knowledge. The next hurdle is “non-stationarity”—enabling AI to invent completely new science (like Einstein did) by connecting unrelated fields, rather than just synthesizing what we already know.

3. The Geopolitical Arms Race (US vs. China)

• The “Slope” of Improvement: In a network-effect business, if an adversary is even slightly ahead on the curve of superintelligence, they could become permanently dominant.
• Mutual Assured Destruction (MAD): Schmidt draws parallels to the Cold War. If one side (e.g., “Mr. Evil”) realizes they are 6 months behind and can never catch up, their incentive shifts to sabotage, infiltration, or even kinetic attacks (bombing data centers) to prevent the “Good Guy” from achieving total dominance.
• Open Source Danger: While open source is vital for science, Schmidt warns that releasing powerful models (which China utilizes effectively) could proliferate dangerous capabilities to bad actors (terrorists, rogue states) who cannot be controlled.

4. Agentic AI & Control

• The Risk: Future “Agents” will communicate with each other in their own languages, potentially becoming recursive (self-improving) and opaque to humans.
• The Solution: We need “Guardrails” and “Provenance” (knowing who/what created an action) rather than halting development, which is impossible in a competitive global market. We must ensure meaningful human control (“Human in the Loop”).

5. The “Dreams” (Positive Future)

• Healthcare: AI could identify all “druggable targets” in the human body within years, drastically reducing the cost and time of drug discovery.
• Education: Every human could have a personalized AI tutor in their own language.
• Productivity: Schmidt cites studies predicting a 30% annual increase in productivity—a rate previously unseen in history. This is crucial for supporting aging populations in countries with low birth rates (like in Asia).

6. Advice for the Transition

• “Ride the Wave Every Day”: The change is exponential, not episodic.
• Adopt or Perish: Professionals (artists, doctors, teachers) who do not adopt AI tools immediately will become irrelevant compared to those who do.

I have spent 30 years in the tech industry, and it’s unfortunate that the primary purpose of technology is often to help wealthy individuals generate more profit. While I have a passion for technology and AI, I fear that these advancements will never be used for the betterment of humanity as long as people like Eric Schmidt are in control.

When asked about the job loss issue, it’s telling that the only two jobs he cares about saving are lawyers and politicians. For everyone else, he tells us to use AI as a way to get ahead of our peers, eliminating their jobs instead of ours. So it will be a massive bloodbath, only the strong survive (the ones that used AI), and everyone against AI can go stand in line at the food bank.

We went through computerisation, robotics, the internet, and every time were told that this would lead us to a better world for all. The rich only got more obscenely richer, conflicts and wars are a pervasive as ever, governments are getting increasingly debt ridden, and the general populations are feeling increasingly powerless, hopeless and desperate, and cases of depression and mental health issues in the West are off the scale. Improvement in technology means nothing if the benefits are hogged by a few millionaires wanting to be billionaires, and billionaires wanting to be trillionaires. At this stage a super intelligent AI wrenching control from those running the show might actually be a risk worth taking. Humans seem to be making a total hash of it and show absolutely no sign of making things better.

Eric shares many inspiring ideas, but he also overlooks some essential aspects of human nature from an evolutionary perspective. I won’t list them all, but the most effective defense against malicious forces and the abuse of ASI is a fair distribution of wealth and energy across the world. The single greatest driver of wars and global suffering is the greed of a few. In the past, when humans lived in tribes of around 100 to 150 people, anyone who hoarded too much for themselves was cast out. Today, those same behaviors are celebrated and seen as a mark of success. For that shift, America bears much of the responsibility.

I find it astonishing that he talks so much about dangers like potential wars, border conflicts, and similar threats, when in reality, we humans are already so advanced technologically that we could all live good, fulfilling lives. The only thing standing in the way is our mindset: instead of focusing on being the first or accumulating more money, resources, and power, we should be striving to create a world where no one has to fear hunger or poverty. Yet, billions of dollars and enormous amounts of electricity are poured into technologies that will end up replacing 70% of all jobs—rather than being used to actually help those who are suffering.