The Al question no one wants to ask

Overview: Viral Backlash to Eric Schmidt’s AI Graduation Speech

The subtitles discuss a viral backlash to former Google CEO Eric Schmidt’s graduation speech about AI. Critics argue that Schmidt’s “rocket ship / don’t ask questions” rhetoric smuggles in corporate interests while using “science” as a shield from criticism.

The speaker and guests claim the public isn’t booing scientific progress itself, but rather the greed, wealth concentration, and power behind AI deployment—especially fears that the future is being run by a small group of Silicon Valley decision-makers.

Schmidt and the “science” defense

• Eric Schmidt is portrayed as using a familiar Silicon Valley persuasion tactic:
  • make a provocative claim (e.g., “get on the AI rocket ship / you’ll be left behind”),
  • then retreat into a safer, generic argument (“science and medicine” will benefit).
• Critics argue Schmidt conflates “AI” with the broader scientific method to dodge accountability, while minimizing ethical conflicts tied to:
  • autonomous weapons,
  • military AI startups,
  • and influence over public funding and contracts.
• The core critique: Schmidt’s framing discourages public scrutiny (“don’t ask which seat”), even though the real question is whether AI will be used to preserve democratic control and human welfare.

Why people supposedly “boo AI”

The video argues that public hostility is mainly about:

• corporate greed and extreme wealth concentration,
• privatization of social safety nets and worsening labor conditions,
• job insecurity narratives meant to keep workers compliant,
• fear of democratic erosion as AI becomes a tool of power.

The speaker insists many people do not reject pro-humanist AI science (such as medical advances). Instead, the anger targets the incentives and governance surrounding AI.

What “AI” actually means (and why the label is misused)

A major analytical point is that “AI” is too broad. Systems range from:

• narrow, research-oriented tools (e.g., protein folding prediction),
• general-purpose, commercial chatbots,
• weapon-adjacent autonomous systems.

The video argues that collapsing everything under “AI” benefits actors who profit from vagueness—making it easier to sell whatever outcomes they want.

A proposed way to look for “what could go right”

The speaker introduces a framework dividing people by knowledge and optimism, then focuses on talking with “PhD doomers” (high knowledge, low optimism) to extract realistic concern—and sometimes grounded hope.

A key contributor is Peter Lebedev, described as an AI safety and science-communication figure who studies how to “put the handbrakes on AI.”

Core risk analysis: AI is powerful, and “doom” has tradeoffs

The video emphasizes a common AI-safety warning: current models can be dangerous because they are trained rather than “hand-coded” for behavior, so harmful outputs can’t be removed by deleting a single line of code.

It lists multiple danger domains already present or accelerating:

• bio risks (AI-assisted drug discovery can also accelerate toxins and viruses),
• cognitive manipulation and propaganda,
• cyber/security vulnerabilities and dependence on infrastructure,
• surveillance and military applications,
• environmental costs (data centers),
• IP theft,
• gradual disempowerment (outsourcing decision-making and control until nothing is truly governed by people).

The video also critiques “AI doom” as a rhetorical strategy that can benefit companies by making apocalypse feel inevitable and reducing public agency (demoralizing people).

What could go right: examples and “good problems”

Lebedev argues the best use-cases are where AI helps solve “good problems” without harmful downstream incentives—especially in research and medicine.

The strongest example is AlphaFold (DeepMind):

• predicts protein structures,
• speeds drug and vaccine discovery,
• produces real-world positive impacts (including informing vaccine development and supporting malaria vaccine research).

The video contrasts these outcomes with systems optimized for profit, control, or military use. It argues that both “good AI” and “bad AI” often rely on similar underlying neural-network technology—so governance and purpose matter more than the label “AI.”

“Political first”: Tony Benn and democracy as the decisive question

The video returns repeatedly to a political-first argument (citing Tony Benn): new technologies should be evaluated in terms of power—who uses them, for what purposes, who is accountable, and whether democratic control can prevent abuse.

The danger is framed as technocratic domination: AI can amplify existing power imbalances unless democratic oversight keeps pace.

Audience action and conclusion

The speaker ends by framing AI backlash as a galvanizing effect: public pressure, education, protest, and accountability efforts can change how AI is built and deployed.

The implied “what could go right” is not naïve optimism about inevitability. Instead, it’s optimism that public resistance can shift incentives toward human flourishing—more medical and environmental benefits, less weaponization, and less unchecked corporate control.

Presenters or contributors

• Eric Schmidt (former Google CEO; discussed via his graduation speech)
• Peter Lebedev (guest/contributor; AI safety/science communicator)
• Demis Hassabis (mentioned in connection with AlphaFold’s Nobel Prize)
• John Jumper (mentioned in connection with AlphaFold’s Nobel Prize)
• Geoffrey Hinton (referenced in the risk discussion)
• Mustafa Suleyman (referenced in bio-weapons/AI risk framing)
• Sam Altman (referenced/quoted indirectly in company-centric discussion)
• Tony Benn (quoted/cited on political questions of technology and power)
• Adam Savage (mentioned in a Veritasium anecdote)
• Paulo Freire (mentioned as an influence related to public education)
• Max Tegmark (referenced for a definition of “intelligence”)