The Rest Is Politics: “What If the AI Revolution Isn’t Real?”

Date: January 25, 2026
Hosts: Alastair Campbell, Rory Stewart
Guest: Arvind Narayanan (Director, Princeton Center for Information Technology Policy)

Episode Overview

This episode centers on a provocative question: Is the AI revolution really as imminent and earth-shattering as headlines claim, or is it a normal technological evolution, which will reshape society much more slowly than the hype suggests? Hosts Alastair Campbell and Rory Stewart are joined by guest Arvind Narayanan, who offers a critical, moderate perspective challenging both the doomsayers and the overzealous techno-optimists. The conversation explores risks, regulatory responses, and the misinformation that permeates current discussions around AI.

Key Discussion Points & Insights

1. Challenging the Language of Probabilities in AI Catastrophe

• Arvind Narayanan casts doubt on attempts to assign probabilities to existential AI risk. He argues that such probabilities are groundless guesses with no empirical foundation:
  • Quote: "We have no empirical basis for predicting what these probabilities might be...these probabilities are all bogus. That's my strongly held view." [04:10]
• He references an extensive Forecasting Research Institute report, arguing it’s no more valid than speculation among laymen.
• Narayanan warns that by arguing about whether the chance of extinction is 1% or 0.1%, we distract from the real decisions about what, if anything, can actually be done.

2. Is AI Too Widespread to Contain?

• Narayanan strongly rejects the idea that shutting down a handful of "frontier" models (run by the richest tech companies) would mitigate most risks:
  • Quote: "You have slightly smaller models...that can run on consumer grade hardware...the cost of running these models is dropping by something like...a factor of 10 to a factor of 100 every year." [07:06]
• He uses the example of GPT-2, considered dangerous just a few years ago but now easily replicated by students, to underscore how thresholds of perceived danger quickly become obsolete.
• Narayanan insists that dangerous capabilities are not strictly correlated with the size or computational heft of a model.

3. Regulatory Possibilities and Practicalities

• There’s agreement on the difficulty of outright bans or pauses, with Narayanan quipping, "The only way [to stop AI] could work is if you have an authoritarian world government that can control every AI developer everywhere." [05:03]
• Matt Clifford probes for regulation that balances both skepticism and legitimate concern.
• Narayanan highlights the importance of "transparency and knowledge" over bans, lauding government bodies like the UK's AI Security Institute for pre-deployment evaluations, even though their influence is currently limited [10:53].

4. Defense, Attacker-Defender Balance, and Learning from Cybersecurity

• Narayanan advocates investing in AI tools and talent to defend against AI-powered threats rather than fixating on offensive “killer robot” hazards:
  • Quote: "One big area is developing AI for defense against these very risks that people are worried about...this is always how it has worked." [11:19]
• He draws parallels to cybersecurity, where superhuman AI tools are already used defensively and offensively, but safeguards prevail chiefly by empowering defenders.

5. The Role and Speed of Regulation

• Matt Clifford voices skepticism about the incentives of current AI developers and executives, citing an "arms race” mentality and hostility towards government oversight [12:29].
• Narayanan acknowledges regulatory lag, especially regarding mainstream harms like psychological impacts and deepfakes, but remains hopeful that more localized, visible harms will force governments to respond:
  • Quote: "I think the arms race narrative is a little bit oversold… we're going to start to see policymakers stop buying this arms race narrative." [14:13]
• He foresees a mindset shift as policymakers realize the most significant AI harms will be felt within their own countries and communities, not just at the hands of big companies [15:53].

6. Security vs. Safety Lenses in Policymaking

• Matt Forde observes a divergence between executive (governmental) security framing and legislative (societal/child protection) concerns:
  • Quote: "Typically legislators tend to see it through a sort of what you might call an online safety type lens..." [15:53]
• Narayanan expects the "online safety" conversation to become more central as AI's effects diffuse through society, with momentum building around both kinds of interventions [17:10].

7. Transatlantic Tensions over Regulation

• Clifford points out European efforts at AI regulation meet with “unmitigated contempt” from Silicon Valley, highlighting accusations that EU efforts are backward or stifling [17:34].
• Narayanan is critical of US resistance, commending the democratic nature of Europe’s approach:
  • Quote: "The hostility of American companies and the government to Europe's approach has been really problematic, and it's not one that I support." [18:27]

Notable Quotes & Memorable Moments

• Arvind Narayanan [04:10]:
  "These probabilities are all bogus...we should not think in terms of probabilities. I do think the risks are potentially real. I'm not advocating for ignoring the risks, but I think the right response cannot be, let's try to stop all this."
• Arvind Narayanan [07:06]:
  "You have...smaller models that are maybe one step below [frontier] that can run on consumer grade hardware...the cost of running these models is dropping...every year."
• Matt Clifford [12:29]:
  “Every time I meet [AI executives], they're in an arms race. They're saying, 'government, back the F off'. None of this gives me any sense that we are in an environment in which these people...are really putting proper safety regulations in place."
• Arvind Narayanan [14:13]:
  "I think the arms race narrative is a little bit oversold… policymakers...are actually taking this very, very seriously. I think we're going to start to see a mindset shift."
• Matt Forde [15:53]:
  "Typically legislators tend to see it through a sort of what you might call an online safety type lens...they were worried about their kids...talking to chatbots."
• Arvind Narayanan [17:10]:
  "As these models get more diffused into society, we are going to see an increase in the salience of that set of concerns."
• Arvind Narayanan [18:27]:
  "The hostility of American companies and the government to Europe's approach has been really problematic..."

Key Timestamps

• 04:10 – Narayanan rejects probabilistic thinking about existential AI risk.
• 07:06 – Argues that dangerous AI cannot be contained to frontier models or banned easily.
• 10:53 – Supports the UK's AI Security Institute and stresses transparency.
• 11:19 – Suggests AI defense investments, learning from cybersecurity.
• 12:29 – Clifford describes the industry's “arms race” mentality.
• 14:13 – Narayanan predicts policymaker mindshift away from arms race framing.
• 15:53 – Forde and Narayanan predict rise of legislative, safety-first approaches.
• 17:34 – Clifford and Narayanan discuss transatlantic rifts on AI regulation.

Conclusion

This episode carefully debunks common AI narratives—whether apocalyptic or triumphalist. Arvind Narayanan urges against probabilistic existential fearmongering, stresses the inevitability and diffusion of AI tools, and contends the best course is practical, transparent regulation and smarter defense, not futile attempts at prohibition. The hosts and guest agree that regulation, though slow, is catching up, and that both executive (security) and legislative (societal safety) concerns will shape the next phase of AI governance. The episode stands out for its balanced, evidence-focused approach to both AI risk and policy—championing moderation, transparency, and collaborative defense over hyperbolic claims or regulatory paralysis.