In a groundbreaking presentation, Lev Tankelevitch, a researcher at Microsoft Research Cambridge, UK, shed light on generative AI’s metacognitive demands and opportunities. Addressing the Collaborative Intelligence team, Tankelevitch delved into the complexities of integrating artificial intelligence into our daily workflows and highlighted the critical role of metacognition in navigating usability challenges.

“AI holds immense potential to revolutionize both personal and professional work. However, despite its promise, numerous usability challenges persist,” Tankelevitch explained. These challenges range from formulating effective prompts to evaluating and relying on AI outputs. In their recent paper, Tankelevitch and their team proposed a metacognitive framework as a powerful tool for understanding and addressing these challenges.

Metacognition, often described as “thinking about thinking,” encompasses various cognitive processes, including self-awareness, confidence adjustment, task decomposition, and metacognitive flexibility. Tankelevitch illustrated these concepts with a simple example workflow involving the use of AI to craft an email. From formulating prompts to evaluating outputs and iterating on prompts, each step requires careful consideration of one’s cognitive processes and strategies.

Beyond the micro-level interactions, Tankelevitch emphasized the importance of considering automation strategy at a broader level. This involves assessing when and how to apply AI to workflows, weighing the benefits of AI against manual tasks, and integrating AI effectively into daily routines.

Tankelevitch’s insights underscored the intricate interplay between human cognition and AI systems, highlighting the need for a nuanced understanding of user needs and capabilities. By reframing usability challenges through the lens of metacognition, Tankelevitch proposed a path toward designing AI systems that support and augment human agency.

Drawing from psychology research, Tankelevitch highlighted the measurability and teachability of metacognition, suggesting avenues for designing AI systems that promote metacognitive awareness and reflection. From assisting in task planning to augmenting AI explanations, Tankelevitch outlined various ways in which AI systems can support users’ metacognitive processes.

Tankelevitch affirmed that a metacognitive perspective offers a promising approach to analyzing, measuring, and evaluating the usability challenges of generative AI. Researchers and developers can design AI systems that empower users and enhance their workflows by leveraging metacognition.