Imagine a world in which some essential decision – a judge’s sentencing recommendation, a child’s treatment protocol, which person or company should receive a loan – was strengthened because a well-designed algorithm helped a key decision maker make a better choice. A modern MIT economics course explores these intriguing possibilities.
Class 14.163 (Algorithms and Behavioral Sciences) is a modern, interdisciplinary course focusing on behavioral economics that examines the cognitive abilities and limitations of human beings. The class was taught last spring by assistant professor of economics Ashesh Rambachan and visiting lecturer Sendhil Mullainathan.
Rambachan explores the economic applications of machine learning, focusing on algorithmic tools that influence decision-making in the criminal justice system and consumer credit markets. He also develops methods for establishing causality using cross-sectional and animated data.
Mullainathan will soon join MIT’s departments of electrical engineering, computer science and economics as a professor. His research uses machine learning to understand complicated problems in human behavior, social policy and medicine. Mullainathan co-founded the Abdul Latif Jameel Poverty Action Lab (J-PAL) in 2003.
The goals of the modern course are both scientific (understanding people) and policy-based (improving society by improving decisions). Rambachan believes that machine learning algorithms provide modern tools for both scientific and applied behavioral economics.
“The course explores the application of computer science, artificial intelligence (AI), economics and machine learning to improve outcomes and reduce bias in decision-making,” says Rambachan.
Rambachan believes there are opportunities in ever-evolving digital tools such as artificial intelligence, machine learning and vast language models (LLM) to lend a hand change everything from discriminatory criminal sentencing practices to health care outcomes among populations with insufficient access to services.
Students learn how to employ machine learning tools with three main goals: understand what they do and how they do it, formalize behavioral economics insights so that they fit well with machine learning tools, and understand the areas and topics where behavioral economics integration and algorithmic tools may prove most fruitful.
Students also generate ideas, develop related research, and see the bigger picture. They allow you to understand where an insight fits and see where a broader research program leads. Participants can think critically about what supervised LLMs can (and cannot) do, to understand how to integrate these capabilities with the models and insights of behavioral economics, and to recognize the most fruitful areas of application of what the research discovers.
The dangers of subjectivity and bias
According to Rambachan, behavioral economics recognizes that there are biases and errors in our choices, even in the absence of algorithms. “The data used by our algorithms exists outside of computer science and machine learning and is often created by humans,” he continues. “Understanding behavioral economics is therefore essential to understanding the effects of algorithms and building them better.”
Rambachan tried to make the course accessible regardless of the participants’ educational background. The class included college students from various fields.
By offering students an interdisciplinary, data-driven approach to examining and discovering the ways in which algorithms can improve problem solving and decision-making, Rambachan hopes to build a foundation from which to redesign existing systems of adjudication, healthcare, consumer credit, and industry, to name just a few several areas.
“Understanding how data is generated can help us understand bias,” Rambachan says. “We can ask questions about getting a better result than we currently have.”
Useful tools for rethinking social action
Economics graduate student Jimmy Lin was skeptical of Rambachan and Mullainathan’s claims at the beginning of the class, but changed his mind as the class continued.
“Ashesh and Sendhil started with two provocative claims: the future of behavioral science research will not exist without artificial intelligence, and the future of artificial intelligence research will not exist without behavioral science,” says Lin. “Over the course of the semester, they deepened my understanding of both fields and walked us through numerous examples of how economics influences AI research and vice versa.”
Lin, who previously worked in computational biology research, praised the instructors’ emphasis on the importance of a “producer’s mindset,” thinking about the next decade of research rather than the previous one. “This is especially important in a field as interdisciplinary and rapidly changing as the intersection of artificial intelligence and economics — there is no old, established literature, so you are forced to ask new questions, invent new methods and create new bridges,” he says.
The speed of change that Lin mentions is also an advantage for him. “We see black box AI methods facilitating breakthroughs in mathematics, biology, physics and other scientific disciplines,” Lin says. “Artificial intelligence has the potential to change the way we approach intellectual discovery as researchers.”
The interdisciplinary future of economics and social systems
Studying time-honored economic tools and enhancing their value with artificial intelligence can result in revolutionary changes in the way institutions and organizations educate leaders and enable them to make choices.
“We learn to track changes, adapt frameworks, and better understand how to deploy tools in the service of a common language,” Rambachan says. “We must continually explore the intersection of human judgment, algorithms, artificial intelligence, machine learning and LLM.”
Lin enthusiastically recommended the course, regardless of the students’ backgrounds. “Anyone with a broad interest in algorithms in society, applications of artificial intelligence in academic disciplines, or artificial intelligence as a paradigm for scientific discovery should take this class,” he says. “Each lecture was a goldmine of research perspectives, novel application areas, and inspiration for exciting new ideas.”
According to Rambachan, the course demonstrates that better-built algorithms can improve decision-making across disciplines. “By building connections between economics, computer science and machine learning, perhaps we can automate the best human choices to improve outcomes while minimizing or eliminating the worst,” he says.
Lin is excited about the course’s yet-to-be-discovered possibilities. “These are classes that get you excited about the future of research and your role in it,” he says.