Every day, we take in information and use that knowledge to make decisions, ranging from quick choices to thoughtful deliberations, with consequences for health, outcomes of elections, and many other areas of our lives. Dr. Sudeep Bhatia, an Associate Professor of Psychology at the University of Pennsylvania, has proposed that training a computer to decide like a human could allow researchers to predict and improve human decision-making. His Computational Behavioral Science Lab is at the forefront of creating models that respond to everyday choices in a human-like manner. For his many contributions to decision-making science in over 70 scientific articles, Dr. Bhatia was nominated by the Society for Judgment and Decision Making to receive the 2023 Federation of Associations in Behavioral and Brain Sciences (FABBS) Early Career Impact Award.
Key Findings
- Computers can be trained to make decisions in human-like ways.
- Models of nutrition knowledge and food choices can be used to simulate responses to different public health interventions.
- Human knowledge can be quantified by mapping the association of different concepts in large samples of human language, such as in social media or the news.
Pasta or salad? When asked to make a choice, the brain draws on its knowledge about each option. Dr. Bhatia’s work reveals that computers exposed to the same information as us can represent this knowledge mathematically and use it to make predictions about human behavior. After scanning the internet, the computer’s algorithm may find that “salad” is closely associated with the word “weight loss” across many dieting websites, while “pasta” is closely associated with “delicious” in online recipes. Perhaps both are equally associated with “dinner.” Dr. Bhatia’s models map the associations between a given food and thousands of other words to mathematically represent the knowledge humans use to decide.
To do this, Dr. Bhatia frequently uses embeddings, the stored memories of a form of artificial intelligence called a neural network that is trained to process language in a human-like manner. Neural network embeddings allow him to quantify human knowledge for millions of different topics across vast and varied sources of information. Rapid and efficient processing of large amounts of news, advertising, or social media data could identify ways to correct misinformation and strengthen the public’s decision-making capacity. Researchers before him have tracked the spread of misinformation within social networks, but Dr. Bhatia’s computational models could extract the meaning from misinformation and predict how individuals in that network might respond to different attempts to correct misperception. Artificial intelligence will allow researchers to keep pace with ever-changing environments and develop data-driven solutions to vexing social or health challenges.
Rather than building a computer that makes perfectly optimal choices, Dr. Bhatia’s team builds human-like models that mimic the mental shortcuts, misinformation, or biases incorporated into human choices. He believes that if we can engineer an accurate model of the mental processes underlying behavior, we can reverse-engineer solutions to improve decision-making. This could radically shift how different interventions are designed and studied in the behavioral sciences. For example, Dr. Bhatia’s algorithms could be used to select the most promising among multiple public health messages, saving significant resources relative to surveying hundreds or thousands of participants about each version. In one study, Dr. Bhatia and his team built a model that mimicked how humans rated the health of different foods and simulated the impact of a common nutrition-labeling strategy in the United Kingdom. In the future, his model could rapidly and accurately compare similar strategies used in France or the United States, without ever having to leave the lab at the University of Pennsylvania. By processing vast amounts of data and forecasting responses to different interventions, Dr. Bhatia’s models could speed up the rate of progress across many fields of study.
Artificial intelligence has numerous applications in the behavioral sciences, yet Dr. Bhatia points out that many researchers are not trained to implement these powerful tools in their research. To capitalize on this untapped potential, he encourages funders to incentivize collaboration with computer scientists and promote uptake of new technologies in the behavioral sciences. Throughout his own career, numerous National Science Foundation grants have allowed him to pursue interdisciplinary collaborations and to train his students in cutting edge methods. Beyond his own lab, Dr. Bhatia hopes that widespread uptake of new technologies will maintain U.S. scientists’ competitive advantage worldwide and lead to scientific breakthroughs that support a healthy, informed society.
Future Impact:
- Training behavioral scientists to use cutting-edge computational methods.
- Analyzing vast amounts of social media and news data to understand people’s beliefs and predict their behavior.
- Using computational models to improve decision-making and algorithmically develop interventions for pressing issues.
Dr. Sudeep Bhatia is a recipient of the 2023 Federation of Associations in Behavioral & Brain Sciences (FABBS) Early Career Impact Award and was nominated by the Society for Judgment and Decision Making.
The SJDM Annual Meeting takes place in San Francisco, CA, from November 17-20, 2023.
Read more about Dr. Bhatia’s work at the links below:
FABBS Comments on the NIH Minority Health and Health Disparities Strategic Plan for 2026-2030