Why do we see tremendous human advantage in many natural tasks, when in some domain we have such “superhuman” AI? To gain deeper insight into what makes humans special, this project compares human experts and autonomous software in complex dynamic tasks, where humans have the edge. We combine data from a multi-site experiments with software-in-the-loop simulations. Identical task environments are rendered to the humans (in a high-fidelity simulator) and used to evaluate of autonomous agents. Using the same performance metrics for both allows direct comparison of humans and machines, despite different underlying information processing. Our hypothesis is that the human advantage emerges from general "core skills" that humans exploit in different situations while machines rely on more task-specific capabilities. The results are generalized into a framework for human and machine comparison that helps us to better understand the nature, limits and origins of the Human Advantage.

2024

2028