Follow us on Google News (click on ☆)
Illustration image Pixabay
When an individual strives to solve a problem or achieve a goal, they do not always opt for the optimal solution. Multiple factors such as time constraints, limited knowledge, or fatigue can influence their choices. Understanding this diversity in human behavior is crucial for improving collaboration between AI systems and users.
The method developed by these researchers is based on the idea that planning time and depth of thought are key indicators of human behavior. They created an algorithm simulating a series of decisions for a given problem, then comparing these decisions to those made by humans. Thus, they identified the moment when individuals stop "planning," giving way to a part that is "irrational" and unpredictable.
This modeling, named "inference budget," assesses an individual's capacity to absorb data before making a choice. With this budget, the model can anticipate the subsequent behavior of the individual facing a problem. This approach allows for a more accurate understanding of human thinking processes in different situations.
The researchers tested their method in three different contexts. First, they observed individuals navigating a maze to understand their approach to progression. Next, they analyzed communications between two people engaged in a color description game, and finally, they studied the performance and strategies of chess players.
The results are promising: the system successfully deduced navigation goals in the maze from previous paths, understood communicative intentions from verbal exchanges, and predicted the next moves in chess games.
This work could pave the way for new applications in the field of AI, allowing systems to better understand and anticipate user needs, thus providing more tailored support and anticipating their future actions.