- The article argues that the crucial question is not whether AI is trustworthy, but how systems are designed to control errors from both AI and humans.
- The author notes that while large language models can fabricate information and fill gaps with assumptions, human memory also frequently reconstructs events rather than preserving them accurately.
- A personal example regarding a high school musical—where the author remembered performing but actually only worked on the set—illustrates the inaccuracy of human memory.
- The aviation industry achieves very high safety levels through checklists, redundancy mechanisms, black boxes, and investigating “near-miss” incidents.
- Medicine has also improved outcomes through standardized diagnostic processes and independent reviews rather than relying solely on personal experience.
- In business, many decisions still depend on training, experience, and intuition, making it difficult to trace the source of errors.
- A 2024 systematic review in Science Advances analyzing 77 studies shows that anti-bias training has limited effectiveness.
- More effective factors include clear processes, standardized criteria, reduced discretionary power, and real-time reminders during decision-making—which are also the foundations of effective AI governance.
📌 Both AI and humans are imperfect, so absolute trust should not be placed in either side. Instead of just focusing on AI accuracy, organizations need to build transparent processes, consistent standards, monitoring mechanisms, and error controls similar to those in the aviation and medical industries. The 2024 study of 77 works also reinforces that effective system design is more important than merely changing individual perception.
