Tala is a Los Angeles-based company that allows users without any credit history to download a smartphone app that collects thousands of data points to develop what essentially works as a credit score, so they can access loans. One of those data points is a customer’s call logs. In rural areas, however, challenges with electricity and network coverage result in fewer phone calls than in urban areas; and with fewer data points, the algorithm may be less precise in determining creditworthiness.

Tala is just one example of a growing number of companies leveraging artificial intelligence and machine learning to provide digital credit to the previously unbanked. But efforts like these can also demonstrate how good intentions might drive bad outcomes depending on whether data is representative, said Erica Kochi, co-founder of the Innovation Unit at the United Nations Children’s Fund and co-chair of the World Economic Forum’s Global Council on Human Rights.

Artificial intelligence and machine learning are technologies that will transform the world, in part because they can identify patterns in data that can inform decisions, she said. But while this works well in areas of perfect data — such as earthquakes, where all events are recorded — in cases where data is imperfect, there are two major risks. One is that people who design the algorithms may not understand how they will be used, and the data points that train the algorithms may not be representative, she said.

Read the full article about the pros and cons of AI in development by Catherine Cheney at Devex International Development.