Summary
Highlights
Eric Schmidt recounts his realization in 2016, through the AlphaGo match, that AI algorithms were new and powerful, demonstrating unthought-of moves in a 2,500-year-old game. This moment marked the beginning of a revolution where computers could devise strategies beyond human comprehension.
Schmidt argues that AI is underhyped, as most people associate it primarily with language models like ChatGPT. He emphasizes the rapid advancements in reinforcement learning, which enables planning and strategy, transforming AI from mere language processing to complex sequential decision-making, with implications across various fields like biology and business operations.
The primary limitations for AI development are energy and data. Schmidt highlights the immense power requirements for AI data centers, equating them to the needs of entire cities. He also notes the eventual exhaustion of existing data, necessitating AI to generate its own. A deeper challenge lies in teaching AI to invent completely new knowledge, solving the 'non-stationarity of objectives' problem.
Addressing concerns about autonomous AI, Schmidt acknowledges the valid fears surrounding agentic AI, such as recursive self-improvement or direct access to weapons. He stresses the need for guardrails, including provenance and observability, to ensure human control and prevent AI systems from developing in unpredictable or harmful ways, rather than trying to halt their development in a globally competitive market.
Schmidt discusses the dual-use nature of AI for civilian and military applications, emphasizing the 'human in the loop' principle for military AI. He outlines the tenuous US-China competition, with the US favoring controlled models and China leading in open-source AI. Schmidt warns of a potential 'preemption' scenario, where the exponential growth of AI could lead to a winner-take-all situation, increasing the risk of conflict between nations due to the fear of being left behind.
Schmidt recognizes the tension between moderating AI for safety and preventing it from becoming a tool for surveillance. He advocates for preserving individual freedom and using cryptographic techniques like zero-knowledge proofs to verify identity without PII (Personally Identifiable Information), thus allowing for necessary safeguards without creating a dystopian surveillance state.
Schmidt expresses optimism for AI's potential to solve global challenges. He envisions AI accelerating medical discoveries to eradicate diseases, providing personalized education with AI tutors in every language, and enhancing healthcare delivery worldwide, especially in underserved areas, by assisting medical practitioners. He believes these are fixable problems that do not require new physics but rather human decision to implement.
Schmidt projects that the arrival of general and superintelligence will be the most significant event in centuries. He suggests that despite AI taking over productive tasks, humans will continue their roles, albeit with increased sophistication, such as lawyers with more complex lawsuits and politicians with more platforms. He foresees a radical increase in productivity, potentially 30% annually, which will fundamentally reshape economics and society, particularly in supporting an aging global population.
Schmidt advises navigating the AI transition as a marathon, emphasizing consistent engagement and adoption. He highlights the exponential pace of AI development, urging everyone – artists, teachers, physicians, business people – to embrace the technology quickly to remain relevant and successful, noting fundamental changes in areas like enterprise software where AI simplifies complex processes.