The Future of Addictive Design + Going Deep at DeepMind + HatGPT

What It Covers

Two California and New Mexico jury verdicts found Meta and YouTube liable for harmful platform design features, awarding $6M and $375M respectively, marking the first successful use of product-defect legal theory against social media. Sebastian Mallaby discusses his book on DeepMind CEO Demis Hassabis, covering the failed Google spin-out attempt and the race toward superintelligence.

Key Questions Answered

• •Social Media Liability Shift: Plaintiffs successfully argued that platform mechanics — infinite scroll, autoplay video, push notifications, and recommendation algorithms — constitute defective product design, bypassing Section 230 protections. This "design defect" legal theory, modeled after Big Tobacco litigation, is now established as viable in jury trials, opening the door to thousands of similar cases across the country targeting Meta, YouTube, and other major platforms.
• •Section 230 Crack: For 30 years, Section 230 of the Communications Decency Act shielded platforms from liability for user-generated content. These bellwether verdicts found a side door: litigating platform mechanics rather than content. If upheld on appeal, companies face potential billions in damages. Platforms should treat internal communications about addictive design or engagement hacks as legally discoverable and act accordingly.
• •DeepMind's Failed Google Spin-Out: Between 2016 and 2019, DeepMind ran Project Mario, a covert effort to spin out of Google. Demis Hassabis secured Reid Hoffman's pledge of $1 billion and met with Alibaba co-founder Joe Tsai in Hong Kong. Sundar Pichai stalled negotiations through deliberate delays and term-sheet revisions until the effort collapsed from exhaustion by 2019, with internal board documents later leaked to author Sebastian Mallaby.
• •Demis Hassabis's AGI Miscalculation: Hassabis, trained in neuroscience, dismissed large language models around 2018-2019 because language lacks grounding in physical reality — a concept called "action in perception." He underestimated how much real-world knowledge is encoded in internet-scale text. This delayed DeepMind's LLM investment and allowed OpenAI to dominate the consumer AI market with ChatGPT's November 2022 launch, which Hassabis described internally as "parking tanks on my lawn."
• •AI Safety Requires Government Coordination: Hassabis argues that individual lab safety commitments are ineffective because competitors like OpenAI will immediately fill any gap left by a principled refusal — as seen when Anthropic drew Pentagon red lines. The only viable path to safe AGI development is simultaneous government regulation forcing all labs to comply at once. Without a new US administration, Hassabis is keeping this conversation alive with other governments.
• •Teen AI Chatbot Adoption Accelerating: A 2025 Pew Research study found 64% of US teenagers now use AI chatbots, with roughly 30% using them daily. Teen usage of YouTube, TikTok, Instagram, and Snapchat remained stable during the same period. Chatbots are expected to be embedded directly inside social platforms, merging the two categories. Legal and regulatory scrutiny applied to social media addictiveness will likely extend to AI chatbot engagement design within years.