Conference Presentation, Panel
The AI Rollout: Revolution, Regulation, and Repercussions | Global Conference 2025
- DeepMind plans to collaborate with democratic governments on national security, cybersecurity, and biosecurity to ensure responsible engagement as AI capabilities expand, while emphasizing a shift from reactive planning to anticipating future technological trajectories.
- Laila Ibrahim asserts the US holds tremendous leadership in AI due to a history of model development, operationalizing care, and societal readiness, with future innovation expected to occur at the model level (planning, memory, concepts) rather than solely through data scaling.
- US leadership is anticipated to persist through continued investment in compute infrastructure and the delivery of capabilities via smaller, energy-efficient models, though the speaker notes the industry must compete on "operationalizing care" and societal readiness.
- Dmitry Shevelenko and Dave Levy project the US will maintain its lead in frontier AI but face the risk of falling behind if ethical constraints limit data usage compared to Chinese models trained on data with "zero" copyright constraints, making "speed" and the freedom to exchange information critical structural advantages.
- Shevelenko warns that diffusion rules for high-grade AI chips may force China to build its own hardware ecosystem, suggesting US strategy should focus on accelerating domestic development rather than constraining competitors.
- The current pace of development is expected to increase hallucinations as models take on complex tasks and longer thinking windows, introducing compounding errors that require a cultural shift to handle unclear ethical answers regarding AI alignment.
- China is expected to "catch up" and build its own hardware, necessitating that the US moves faster to maintain its position in the global ranking of technological capabilities.
- Sean Pack forecasts a fragmented legal landscape driven by "eight to 10 jurors" deciding fair use, liability, and harm cases over the next one to two to three years, creating a tort-based system in the absence of uniform regulation.
- Legal uncertainty and litigation risks are predicted to cause "less investment in U.S.-based companies" within four or five years as capital flows to jurisdictions like China and Japan unless a national initiative resolves these regulatory hurdles.
- Shevelenko anticipates that the percentage of daily decisions aided by technology will rise, potentially allowing humans to focus on choices that are "morally important" given that the average American adult makes 35,000 daily decisions.
- Perplexity plans to publish a weekly change log for every product update to establish transparency, which Shevelenko and Pack view as a future competitive advantage for establishing trust and differentiating companies through explainability.
- Industry consensus views "adoption and speed" as primary concerns, with the "finish line" for AGI defined by an application space where users convert ideas into action without high technical acumen at decreasing costs.
- Shevelenko predicts AI will transform the browser experience by enabling agents to access personal data sources (email, calendar, Slack) to provide synthesized context, differing significantly from current browsers like Chrome and Safari.
- Public benefits such as solving healthcare problems and enabling individualized learning are expected to form the foundation for fair use and adoption decisions, though the speaker fears the industry may fail to translate these benefits to decision-makers.
- The pie of the internet is predicted to grow rather than be reallocated, with AI bringing new users onboard, though the US patent system faces challenges regarding inventions created by models that do not qualify for human-based protection.
- The "greatest AI safety risk" is identified as the failure of the US to lead in frontier development due to self-imposed constraints, allowing adversaries to gain an advantage rather than technical failure.
- The speaker expresses a fear that the industry is too focused on risk and not enough on maximizing opportunities for humanity, advocating for AI to happen "with" society rather than "to" them.
- The "AI race" is described as being at the very beginning, requiring global partners to help realize US leadership while "smart money" flows to places with resolved regulatory environments if the US fails to act.