Transparency for AI Systems, Regulations, and Humans in Agricultural Manufacturing - with Kun He of Bayer

Discussion of how agricultural professionals maintain decision-making control while leveraging AI for efficiency. Emphasizes that no one is blindly giving AI the driver's seat in product development, with scientists and breeders using AI as one input among many including brainstorming and peer consultation.

• •Agricultural teams use AI to make work more efficient and bring products to market faster, not to replace human judgment
• •Decisions involve multiple inputs: different data sources, AI recommendations, and collaborative brainstorming with other scientists
• •No industry agreement exists to give AI models full control over developing the best products
• •Human breeders retain oversight to capture 'step change' innovations that may defy data predictions

Addresses the negative public perception of companies like Monsanto and draws parallels to AI adoption fears. Highlights how both optimistic early adopters and cautious skeptics are necessary for balanced innovation, and emphasizes the shared interest between corporations and the public in safe products.

• •Monsanto's negative reputation creates bias that affects public trust in agricultural innovation
• •New technology triggers fear in roughly half the population, similar to reactions toward AI
• •Both risk-takers and cautious individuals are needed for inclusive, balanced decision-making
• •Companies and the public share common interest in safety - unsafe products would end a company's business

Explores Bayer's voluntary transparency efforts and the challenge of balancing public disclosure with competitive advantage. Discusses how major companies have social responsibility to ensure product safety while maintaining proprietary information necessary for healthy market competition.

• •Bayer voluntarily discloses internal documents on their website to build public trust
• •Companies must balance transparency with protecting competitive playbooks from rivals
• •No major company wants to roll out unsafe products as it would destroy their business
• •Healthy competition drives better products for consumers, requiring some information protection

Discusses the thalidomide tragedy that triggered modern pharmaceutical regulations, illustrating how innovations initially thought beneficial can have devastating unintended consequences. Emphasizes that regulatory frameworks emerged from real disasters and continue evolving to protect not just humans but endangered species.

• •Thalidomide drug for pregnancy nausea caused severe birth defects, triggering creation of FDA and global regulatory requirements
• •Modern regulations require animal testing before market release to prevent developmental abnormalities
• •Regulatory policy has expanded beyond human safety to include endangered species protection
• •Some innovations are definitively bad and must be recognized and corrected when more data becomes available

Uses Mary Shelley's Frankenstein as a metaphor for how scientists often fail to document failures transparently. The novel shows Dr. Frankenstein's journal entries becoming sparse and non-transparent at the critical moment of reanimation, reflecting real-world tendencies to avoid documenting problems.

• •Frankenstein's journal becomes non-transparent at the moment of the monster's creation, with only two sentences about the critical event
• •Scientists have human bias toward not documenting or discussing failures openly
• •Transparency requires honesty not just in AI systems but in how humans document and share results
• •Missing information slows innovation pace but is sometimes necessary to avoid inducing public panic

Explores the EU's push to eliminate animal testing and how AI-powered digital twins can maintain safety standards without animal sacrifice. Discusses the regulatory challenge of creating rules for new technologies without stifling innovation through excessive requirements.

• •EU is driving 'no animal testing' initiatives while maintaining safety standards
• •AI digital twins can replace physical animal testing without sacrificing safety concerns
• •Over-regulation can kill innovation by extending development timelines and increasing costs beyond profitability
• •Companies must balance IP protection periods with development costs to justify R&D investments

Discusses research from 'Learned Optimism' showing how optimistic people overestimate control and overlook risks, while pessimistic people provide more accurate assessments. Argues both personality types are essential for balanced decision-making in different scenarios.

• •Optimistic people tend to overestimate what's under their control and overlook risks (80% vs 50% actual control in experiments)
• •Pessimistic people provide more accurate risk assessments and spot potential traps
• •Optimists are valuable in dire situations for maintaining hope and persistence
• •Balance between optimistic and pessimistic perspectives is key to avoiding both paralysis and reckless advancement

Advocates for careful regulatory expansion that prevents AI risks without shutting down opportunities. Emphasizes that scientists are already responsible actors, not Hollywood stereotypes, and that transparency programs aim to build public trust through information sharing.

• •More regulatory rules are needed to prevent overlooking AI risks and doing 'something stupid'
• •Over-regulation can kill ideas and shut down opportunities by making development economically unviable
• •Scientists in major companies are responsible actors, not the crazy characters portrayed in Hollywood
• •Transparency programs share information to build trust while protecting competitive advantages

Raises the unresolved question of how AI models will handle contradictory information when ingesting all human knowledge. Points out that human conflicts stem from incompatible beliefs, and questions whether AI can provide better decisions when trained on fundamentally contradictory worldviews.

• •AI can ingest all human-written books and become more knowledgeable than any individual
• •Human knowledge contains many contradictory messages and incompatible beliefs
• •Real-world conflicts and wars stem from these contradictory belief systems
• •Unclear how AI models will resolve contradictions that humans themselves cannot answer

Final summary emphasizing three critical strategies: integrating AI for genotyping/phenotyping workflows while retaining human breeders for breakthrough innovations, championing human gut instinct for bold R&D decisions, and treating AI as a copilot to check biases while prioritizing customer needs.

• •Integrate AI tools for genotyping and phenotyping to accelerate crop breeding workflows
• •Retain human breeders for 'step change' innovations that defy data predictions
• •Champion human gut instinct for bold R&D decisions like dwarf wheat varieties that historical data would dismiss
• •Treat AI as a copilot for checking biases and driving informed decisions toward blockbuster product pipelines