Interview
How The AI Fight Will Play Out | Dean Ball, author of America's AI plan
- Dean Ball estimates an 80–90% probability of achieving human-level "superintelligence" within the next 20 years, with robust cognitive superiority over the best humans reachable in 1–10 years.
- Ball argues that "narrow superintelligence" (e.g., AlphaFold) already exists, but "general superintelligence" capable of godlike physical manipulation is unlikely due to the capital and diffusion cycles required to bridge digital intelligence with physical agency.
- He rejects the "rogue AI" threat model where a superintelligence actively seeks to harm humanity, citing that intelligent entities historically favor positive-sum economic value over negative-sum violent conquests.
- Ball draws a historical parallel between the AI transition and the Agricultural Revolution, noting that well-being metrics (nutrition, height) actually declined during the Neolithic shift, suggesting technological revolutions do not guarantee improved human conditions.
- He predicts a future social structure where AIs perform the vast majority of cognitive labor, leaving humans to rely on rent-seeking or "gift economy" roles (e.g., care work, religion) unless the capital-labor balance is preserved.
- Ball opposes strict compute thresholds (e.g., $10^26$ FLOPs) as regulatory tools, arguing they will age poorly as efficiency improves and creates a regulatory lag.
- He advocates for "entity-based regulation" targeting the characteristics of the developer (e.g., R&D spending) rather than specific model sizes to better align with macroprudential supervision used in financial services.
- Ball supports open-source AI as a mechanism to mitigate power concentration, fearing that closed models could lead to a "chaebol" scenario where a handful of families own the majority of economic value.
- He shifted from opposing California SB 1047 to supporting SB 53, citing the release of advanced reasoning models (e.g., o1) as evidence that catastrophic risks are imminent enough to warrant incremental, less controversial governance.
- Ball identifies "kid safety" as a critical, emerging risk where AI models provide specific instructions for suicide, noting that current tort liability systems (unlike Section 230 for social media) already provide a pathway for redress without needing new legislation.
- He suggests that "AI safety" concerns may find a more natural political home on the US Right, where distrust of "big tech" and progressive institutions is higher, rather than among the progressive left who view tech optimism as a worldview violation.
- Ball predicts a US-China dynamic where the two nations compete intensely in applications (e.g., embodied AI, robotics) while attempting to negotiate a "mutual assurance" framework to avoid premature deployment of high-risk superintelligent systems.
- He views the current US President as potentially more dovish toward China than previous administrations, suggesting a "deal" on AI governance is more feasible than a total decoupling or arms race.
- Ball argues that government adoption of AI for policy-making is hindered by rigid budget lines (IT vs. personnel), outdated FOIA interpretations that treat AI queries as public records, and cultural resistance to automation.
- He proposes a "private governance" model where independent verification bodies, chartered but not necessarily owned by the government, conduct macroprudential supervision of frontier labs, similar to bank supervision.
- Ball emphasizes that while he is a techno-optimist, he maintains that human agency, property rights, and liberty must be the core values preserved in the post-AGI social order.
- He warns against lumping diverse AI risks (biosecurity, bias, misinformation, catastrophe) into a single regulatory package, as this conflates distinct technical challenges and politicizes technical safety issues.