AI Doesn’t Know What Your Planners Know

In 2025, MIT's NANDA (Natural and Adaptive Decision-Making for Autonomous Systems) research team published a groundbreaking study that revealed a significant challenge facing AI pilots: a staggering 95% failure rate. This revelation highlighted a critical issue rooted in the AI's inability to access the same contextual knowledge as human pilots. The study underscored the limitations of AI systems in understanding the nuanced, real-world information that experienced pilots rely on to make decisions.

Human pilots, through years of training and experience, accumulate a vast repository of contextual knowledge. This includes understanding local weather patterns, airport traffic, and emergency procedures, among other factors. They also possess the ability to interpret non-verbal cues, such as the tone of voice of ground control or the unspoken expectations of other pilots. AI systems, however, operate primarily on structured data and algorithms, which often fail to capture the subtleties of human decision-making.

The NANDA research demonstrated that AI pilots struggled to perform tasks that required contextual awareness. For instance, when presented with a scenario where an unexpected event occurred at an airport, the AI systems were unable to draw on the same level of situational understanding as human pilots. This led to errors in judgment and, ultimately, the high failure rate documented in the study.

The implications of this research are far-reaching. As the aviation industry continues to explore the integration of AI systems, the findings from MIT's NANDA project serve as a cautionary tale. They emphasize the need for a more collaborative approach between AI and human pilots, rather than relying solely on autonomous systems.

One potential solution to this problem is the development of hybrid systems that combine the strengths of both AI and human pilots. By integrating AI with human expertise, such systems could leverage the computational power of AI while still benefiting from the contextual knowledge and intuition of experienced pilots. This approach would require significant advancements in AI technology, as well as a reevaluation of the roles and responsibilities of pilots in the cockpit.

Moreover, the NANDA research highlights the importance of continuous learning and adaptation for AI systems. As the aviation landscape evolves, AI pilots must be equipped with the ability to assimilate new information and adjust their decision-making processes accordingly. This would involve not only updating existing algorithms but also designing systems that can learn from real-world experiences and human input.

The 95% failure rate of AI pilots, as revealed by MIT's NANDA research, is a stark reminder of the limitations of current AI technology in the aviation sector. The study underscores the critical need for a more nuanced understanding of contextual knowledge and the importance of collaboration between AI and human systems. As the industry progresses, addressing these challenges will be essential to ensure the safe and efficient integration of AI into the world of aviation.