Innovative AI Techniques Reveal How We Make Decisions

Recent advancements in artificial intelligence are shedding new light on the intricacies of decision-making processes in humans and animals. Traditionally, psychologists and neuroscientists have studied decision-making by examining trial-and-error behavior influenced by recent information, often assuming that decision-makers aim for optimal choices based on past experiences. However, this perspective may overlook the complexity and suboptimal strategies individuals sometimes employ.

A groundbreaking study published in Nature introduces a novel approach by utilizing tiny artificial neural networks to explore the actual cognitive strategies behind decision-making. Unlike complex AI systems, these small neural networks are simple enough to interpret but still capable of capturing complex and realistic behavior. They allow researchers to identify the underlying reasons for specific choices, even when those choices are not optimal.

The research was led by scientists from New York University and the University of California, San Diego. As explained by Marcelo Mattar of NYU, this method acts like a detective, uncovering how decisions are truly made in brains—not how they should be made in theory. Ji-An Li, a doctoral student involved in the study, emphasizes that the small neural networks serve as effective tools for predicting individual choices better than traditional models that assume optimal decision-making.

One significant advantage of using minimal neural networks is their interpretability. They provide insights into the mechanisms driving decisions, which are difficult to extract from larger, more complex AI models employed in commercial applications. Marcus Benna from UC San Diego notes that while large neural networks are excellent at making predictions, understanding their decision processes remains challenging. Smaller networks enable scientists to analyze behaviors using physics-inspired methods, thereby revealing the inner workings of decision strategies.

The findings highlight that human, primate, and rodent decisions often involve suboptimal strategies, reflecting real-world decision-making more accurately than traditional models. Furthermore, the models can predict individual strategies, showing that each subject employs different approaches based on their experiences.

This research has profound implications beyond neuroscience. By understanding individual differences in decision-making, this approach could transform mental health treatments, cognitive assessments, and even inform strategies in business and policy. As Mattar concludes, recognizing the diversity in decision strategies can pave the way for more personalized and effective interventions in mental health and cognitive function.

For more details, see the full study: Discovering cognitive strategies with tiny recurrent neural networks.