AI Safety Emphasizes Human-in-the-Loop and Interdependent Systems

A professor leading Pennsylvania State University's aerospace department outlined a design approach for commercial aircraft where artificial intelligence, the pilot‑in‑command and the second‑in‑command operate as interdependent participants. She argued that safety requires all three to share real‑time data, explicitly announce actions, and request assistance when limits are reached, noting the added complexity in design, testing and certification for such integrated systems.

A separate analysis of AI product development highlighted the need for a human‑in‑the‑loop framework that balances autonomy with oversight. It recommends placing humans at decision points where the cost of error is high, actions are irreversible, or model confidence is insufficient, using calibrated confidence thresholds tailored to each task—ranging from low thresholds for support tickets to near‑certain confidence for financial or medical decisions.

Both discussions stress that fully autonomous AI can lead to critical mistakes, while judicious human involvement can reduce errors significantly without compromising product viability.