Imagine entering college for the first time. In addition to starting day one in a completely new environment, you have countless decisions to make within just a couple of weeks that will determine your path for the next two to four years: “What do I want to study?” “Which courses do I need to take?” “Should I work? Full-time, part-time?” “How can I balance my credits so I graduate before my financial aid expires?” Maybe your school has a well-staffed advising office and you stop in for some advice – or, perhaps it is under-resourced, and you wait 45 minutes (or days) for an impersonal 15-minute appointment before being rushed out with a stack of papers and few answers.
Not surprisingly, you would not be the first person to ask many of these questions, yet there often aren’t clear (or easy to find) answers. But what if there was a way to see, at a glance, what other students like you (with many of these same questions in their minds, and similar situations in their lives) had chosen, and how it ultimately worked out for them?
Based on a combination of existing data like your interests and past academic performance, you could get an automated recommendation tailored to outline the best potential pathways for you—providing simpler options for the choices you can make, and leaving you with greater confidence that you will graduate on time. Following your personalized recommendations, you would also be enrolled in an automated support program. Designed specifically for a student like you, the program would help keep you on track for your first year of school, using precise levels of contact to provide no more and no less support than you’re likely to need.
By combining behavioral science and machine learning, this is not a hypothetical scenario, it’s entirely possible. At ideas42, we’ve been working at the intersection of these disciplines for many years to design effective and beneficial programs, and we can now fully leverage the power of data science and machine learning to magnify our social impact at scale.
We believe machine learning is poised to revolutionize the application of behavioral science. Very soon, policy-makers will be able to solve problems previously considered intractable, deliver and evaluate behaviorally informed programs in more intelligent, efficient and effective ways, and, perhaps most profoundly, improve human decision-making itself.
While the possibilities are vast, our work will initially focus on two applications of machine learning methods that can strengthen the potential impact of behavioral science programs.
The first focus will be optimal policy assignment, meaning applying machine learning methods to personalize programs to individuals’ predicted needs. For example, we are currently working with the City University of New York (CUNY) to personalize successful messaging campaigns aimed at nudging students to complete (and renew) their Free Application for Federal Student Aid (FAFSA). Using machine learning to pick up on signals from students, like course selection and major, we will be able to better match students to specific messages in the campaign that will resonate with them (and drive behavior change) most. At scale, such an application could lead to tens of millions more dollars in much-needed financial aid disbursed to students nationwide.
Our second focus will expand on our existing work by using machine learning to better understand and ultimately improve human decision-making. Some policy problems rely heavily on the use of prediction. For example, in coastal areas local governments decide which neighborhoods are allocated rescue boats during a hurricane, which relies heavily on predicting the likelihood and severity of flooding in different areas. Doctors often decide whether or not to order risky procedures by relying on predicted outcomes for a patient’s health. Predictive models driven by machine learning can improve decision making in these and other scenarios even when humans ultimately have the final say.
In our work at ideas42, we develop models to make these sorts of predictions and provide the resources, support, and behavioral lens necessary to meaningfully operationalize them. After all, machine learning predictions present new tools and challenges for human decision-making, and our expertise in behavioral science provides us with the ability to help humans thoughtfully and effectively work harmoniously with machines.
Over the course of the next year, we will be collaborating with leading academic researchers operating at the intersection of behavioral science and machine learning, including ideas42 co-founder Sendhil Mullainathan and our academic affiliates Susan Athey and Jens Ludwig, to test, refine, and evaluate the impact of machine learning-driven methods in field settings. Throughout this work, we are placing a premium on ensuring that our data science practices are fair, ethical, transparent, safe, and in line with best practices. Along the way, we will share details both about how to effectively address these concerns, and what we learn while exploring cutting-edge approaches to support our mission of improving the lives of millions of people.
In the meantime, if you have a program, issue, or opportunity that you think could benefit from the combination of behavioral science and machine learning, please don’t hesitate to reach out to one of the members of the ideas42 team, Rachel Rosenberg, at firstname.lastname@example.org.