Navier-Stokes and AI generated work: If You Can't Explain It, Did You Really Do It?

The discussion begins with the recent controversy surrounding the AI-assisted solution to a version of the Navier-Stokes equation, involving researchers from NYU’s Courant Institute and Anthropic, with subsequent work by OpenAI. A key concern raised is the potential for large language model (LLM) providers to implicitly use user prompts for model improvement, leading to questions about intellectual property (IP) protection. This incident underscores why some users are shifting towards open-weights models and running them on their own infrastructure for greater control and IP security. The conversation then pivots to the broader impact of AI on knowledge work. The idea of “human reserved” work (e.g., elder care), where tasks are set aside for humans even if AI could perform them, is introduced as a concept for societal benefit. The field of mathematics, akin to coding, is identified as a “ground zero” for AI disruption. Leading mathematicians, like Terrence Tao, express concern that AI solving fundamental problems could deplete the “non-renewable resource” of unsolved mathematical challenges, potentially impacting the profession and the inspiration for future insights. The discussion highlights that AI makes “generation” of solutions cheap, and “verification” (in fields with formal tools like math and coding) is increasingly doable. However, “validation” remains the critical bottleneck. Mathematicians propose that for a solution to be accepted, the human author must be able to present and explain their work to peers, fostering true understanding and accountability. This contrasts with simply generating a paper or code via AI and links to concerns about the loss of the apprenticeship process crucial for developing expertise. From a corporate perspective, the rapid advancement of AI technology outpaces organizational transformation, creating friction between shareholder obligations and employee well-being. There’s a call for corporations to redefine their values beyond mere profit optimization, considering people and the planet. The ability to explain one’s work (even AI-generated) becomes a vital differentiator and criterion for evaluation, especially in team-based knowledge work. The conversation then introduces the concept of “involution” (内卷), originating from China: a state of hyper-competition where individuals must constantly work harder and innovate just to maintain their position, while rewards diminish. Innovation quickly becomes a standard expectation, squeezing margins and making it difficult to raise prices. China’s command economy and provincial competition are cited as factors accelerating involution, particularly visible in sectors like automotive, robotics, and increasingly, AI (among open-weights model providers). The hosts predict that AI will introduce similar “involutionary” pressures to Western knowledge workers. Initial productivity gains from AI tools will be nullified as widespread adoption makes these capabilities standard. This leads to an expectation of more output without commensurate increases in compensation, and a mandate for workers to adopt AI tools lest they be left behind. Evangelos points out that China’s involutionary production, particularly in AI models, leads to exports that cause price depression in international markets, impacting local innovation elsewhere. This fundamentally alters the nature of careers and employment, necessitating continuous innovation merely to stay employed, potentially fostering “gray area” practices by organizations and challenging traditional notions of career progression and meaning derived from work. Regarding AI risk, a notable divergence exists between US and Chinese approaches. US labs, often influenced by “science fiction” narratives, rationalism, and effective altruism, tend to focus on existential risks like superintelligence. In contrast, Chinese risk discussions are more pragmatic and application-specific, centered on tangible concerns like cybersecurity and bioterrorism. The hosts suggest that common ground for international collaboration on AI safety could be found by focusing on these more practical, less speculative risks. Finally, the discussion reflects on the 25th anniversary of 9/11 and its profound impact on data technology. The 9/11 attacks starkly revealed critical information silos within government agencies, leading to a national imperative for better data sharing and analysis. This need spurred initiatives (like the Pentagon’s Information Awareness Project) and catalyzed the development and adoption of “big data” technologies, such as Hadoop, which emerged around 2006. Evangelos highlights the government’s recognition of the importance of “link analysis” using commercial tools to connect disparate data points. This drive for federated database analysis, crucial for national security, subsequently found its way into commercial operations, laying foundational groundwork for the broader big data industry and firms like Palantir. The event underscored the critical value of integrating and analyzing vast, disparate datasets.

• Tags:
• Feature
• Explainable AI
• AI Ethics
• Big Data
• Data Analysis
• Knowledge Work
• Intellectual Property
• AI Risk
• China AI
• Navier-Stokes
• Human-Reserved Work
• Apprenticeship
• Involution
• 9/11
• Hyper-Competition