The "2026 Caijing Annual Conference: Forecast and Strategy & 2025 Global Wealth Management Forum" was held in Beijing from December 18 to 20, 2025. A dialogue took place between Xue Lan, Dean of Schwarzman College at Tsinghua University and Director of the Institute for AI International Governance, and Professor Terrence Sejnowski, a pioneer in neuroscience and AI and member of the U.S. National Academies.
When discussing the key differences between large language models and human intelligence, Sejnowski noted that intelligence is multifaceted. While human intelligence, AI, and military intelligence share similarities, they differ significantly. Large language models excel in knowledge retention but lack the default mode network of the human brain, preventing autonomous planning and memory. On whether AI could develop self-awareness, he acknowledged the unpredictability but emphasized that consciousness is not unique to humans—species like chimpanzees exhibit self-awareness without complex language, offering insights for AI consciousness research.
Regarding the convergence of bioscience and AI, Sejnowski described it as an emerging field. Current AI architectures, though closer to human brain structures, still rely on rule-based behaviors, whereas humans adapt through experiential learning. Deep integration could enable mutual enhancement, but progress remains in early stages.
Xue Lan likened current AI to a "cognitive steam engine," advocating for new mathematical frameworks to explain its high-dimensional evolution. Sejnowski agreed, highlighting that while humans operate in 3D space, language models function with high-dimensional parameters. Mathematicians are exploring theories to formalize this, akin to thermodynamics.
On AI risks and regulation, Sejnowski stressed balancing risks and benefits, citing parallels in automotive and pharmaceutical industries. Bias and hallucinations in AI are technical challenges solvable through refinement. He favored industry self-regulation, cautioning against premature government intervention that might stifle innovation.
Regarding "red-line" governance and international collaboration, Sejnowski endorsed setting boundaries, drawing from 1960s biotech risk management. He supported Xue’s proposal for global joint labs, emphasizing researcher-policymaker collaboration, possibly in neutral countries to avoid geopolitical hurdles.
On extreme risks like AI失控 (loss of control), Sejnowski acknowledged concerns but deemed timelines speculative. He advocated for international early-warning systems to prevent nuclear-scale disasters.
Addressing AI’s impact on jobs, especially for youth, Sejnowski remained optimistic. While AI may displace roles, it also creates opportunities—e.g., data science hiring surged, with UC San Diego adding 50 faculty in five years. He advised youth to pursue degrees and upskill in AI to enhance productivity amid cycles.
For humanities professionals transitioning to AI, he cited an English literature graduate who became an AI prompt engineer, illustrating how liberal arts skills align with AI ethics and content creation. Over 100,000 AI startups globally offer diverse opportunities.
Looking ahead, Sejnowski projected AI would mimic brain organization, advancing in specialized processing and cooling tech. Fusion with neuroscience might enable machines to intuit and create, while human consciousness research could yield breakthroughs.
Xue Lan concluded that AI governance requires navigating complex stakeholder dynamics. International cooperation, inspired by the Asilomar principles, remains vital to balance innovation and safety while ensuring inclusive adaptation to AI’s transformative potential.
Comments