Complex math can help Spotify pick your next favorite song.

“Reasoning is critical to machine learning,” said Nailong Zhang, a software engineer at Meta. Meta is using causal inference in a machine-learning model that manages how many and what kinds of notifications Instagram should send its users.

Romila Pradhan, a data scientist at Purdue University in Indiana, is using conflicting data to make automated decision making more transparent. Organizations now use machine-learning models to choose who gets loans, jobs, forgiveness, housing (and who doesn’t). Regulators have begun asking firms to explain the consequences of many of these decisions to those affected. But it is difficult to reconstruct the steps made by a complex algorithm.

Pradhan thinks counterfactuals can help. Let’s say a bank’s machine learning model rejects your loan application and you want to know why. One way to answer that question is with counterfactuals. Since the application was rejected in the real world, would it have been rejected in a fictional world where your credit history was different? What if you have a different zip code, job, income, etc.? Increasing the ability to answer such questions in future loan approval programs will give banks a way to provide reasons to their customers rather than just saying yes or no, Pradhan said.

Contrasts are important because they’re a way for people to think about different outcomes, says Pradhan: “They’re a good way to capture explanations.

They can also help companies predict people’s behavior. Because counterfactuals allow them to predict what might happen in a given situation, not just on average, but technology platforms can pigeonhole people with greater precision than ever before.

The same logic that debunks the impact of dirty water or creditor decisions can be used to improve the impact of Spotify playlists, Instagram notifications, and ad targeting. If we play this song, will the user listen for a long time? If we show this picture, will that person keep scrolling? “Companies want to understand how to make recommendations to specific users rather than the average user,” says Gilligan-Lee.