Overview of The backlash against AI
This episode of the Rachman Review examines the growing public and political backlash against artificial intelligence, with Financial Times columnist Gideon Rachman speaking to Sebastian Mallaby, author of The Infinity Machine. The discussion ranges from fears about job losses and inequality to the geopolitical race between the US and China, the prospects for international AI regulation, and whether governments can realistically control a technology that is advancing rapidly.
Key Takeaways
- The backlash is real and likely to intensify, especially among young graduates who feel most exposed to AI-driven labor disruption.
- Much of the current job impact is fear rather than proven destruction, but Mallaby expects real labor-market damage to emerge soon.
- AI may widen inequality, because the biggest winners are the founders and investors building the systems.
- Governments still have leverage over AI through compute infrastructure, cloud access, chip supply chains, and access controls.
- The most urgent risk is not just job loss but misuse: cyberattacks, bioweapons, and other criminal applications.
- International cooperation is difficult but not impossible; Mallaby argues for a framework closer to nuclear-style nonproliferation than total bans.
- AI progress is likely to accelerate further, with possible “recursive self-improvement” or a kind of takeoff by the late 2020s.
AI, Jobs, and Social Backlash
Why the backlash is growing
The episode opens with the booing of former Google CEO Eric Schmidt at a university commencement, which Mallaby sees as a sign that anxiety about AI is becoming more visible. He argues that college graduates are especially alarmed because:
- AI is already capable of doing many cognitive tasks.
- New entrants to the labor market have less experience to distinguish themselves.
- The hiring process itself has become an AI-versus-AI arms race, with applicants using AI to apply and employers using AI to screen.
Is AI already destroying jobs?
Mallaby’s view is that the strongest evidence so far is still anecdotal, and some layoffs blamed on AI may have other causes. But he thinks that within one or two years AI will start hurting labor markets more clearly.
Education and deskilling
The conversation also touches on the risk that people, especially children and students, may stop learning deeply if AI can do the work for them. The response from Mallaby is essentially:
- Learning still matters.
- People need to do the hard work first.
- AI should be used to accelerate, not replace, human development.
He supports classroom testing without AI to preserve genuine learning.
Inequality, Power, and the “Pitchforks” Problem
Mallaby argues that AI could sharpen inequality because a small number of people are becoming extremely wealthy from the technology while many others fear for their livelihoods.
He cites examples of enormous stakes and fortunes in AI companies, reinforcing the impression that the gains are concentrated at the top. This creates a political risk: backlash may be fueled not only by job fears but also by resentment of AI elites.
How Governments Might Regulate AI
The case for regulation
Mallaby’s core argument is that AI should be regulated more like a dangerous technology than a normal consumer product. He compares the situation to the FDA:
- Medicines are tested before they go into human bodies.
- AI models, he argues, should be tested before being deployed into society.
What governments can actually control
Even though AI is built by private companies, Mallaby says governments still have meaningful choke points:
- major cloud infrastructure
- large compute clusters
- chip supply chains
- distribution portals for proprietary models
He points to recent cases where AI companies or governments effectively restricted access to models after misuse was detected.
A “nonproliferation” model for AI
Mallaby draws on Cold War nuclear history:
- The US and USSR managed their rivalry through deterrence and parity.
- Separate nonproliferation regimes tried to stop weapons spreading to rogue actors.
For AI, he thinks the more urgent issue is preventing proliferation to criminals, terrorists, and other bad actors, rather than trying to stop a US-China race entirely.
US, China, and Europe
China
Mallaby says his recent trip to China suggested that Chinese AI leaders are more focused on safety than Western stereotypes might imply. He argues that dialogue with China on AI safety should continue, even if the broader geopolitical competition remains intense.
The US-China race
He doubts the US and China will fully cooperate on AI regulation because of strategic and military rivalry. Still, he believes some forms of restraint are possible, especially around misuse and distribution.
Europe and sovereign AI
The episode also addresses European concerns about dependence on American AI infrastructure. Mallaby thinks Europe can build capable systems, but true autonomy is limited if those systems rely on American chips and technology.
His proposed compromise:
- Europe should have its own national regulator and some sovereign control.
- American AI firms could supply models under conditions that ensure proper safety oversight.
- This would resemble civilian nuclear power: useful technology, but under strict rules.
Risks Beyond Jobs
Mallaby is especially worried about non-labor risks, including:
- cyberattacks enabled by frontier models
- potential bioweapons
- malicious use by criminals or terrorists
He says these risks are not hypothetical anymore, and that recent model behavior has already shown serious cybersecurity implications.
The Future of AI
The episode ends on a note of rapid acceleration. Mallaby believes:
- the next 3–5 years will be even more transformative than the last three
- AI development may enter a phase of recursive self-improvement
- by around 2028, systems could be advanced enough to help code the next generation of themselves
That could lead to a dramatic jump in capability, sometimes described as a singularity or vertical takeoff.
Bottom Line
The conversation presents AI as both a historic opportunity and a serious policy threat. Mallaby does not argue for stopping AI, but for:
- stronger regulation
- serious safety testing
- restrictions on dangerous distribution
- international coordination where possible
- preserving human learning and judgment in education
His view is that backlash against AI is understandable — and may even be useful — if it pushes governments to act before the biggest harms arrive.