As governments increasingly turn to data science to solve social, economic, and legal problems, citizens should be concerned with how ethical such implementations are. When governments use data science and machine learning models to establish policies, who ensures that such models are accurate?
Journalism is one of the few watchdogs over those in power. Today, journalists have learned how to reverse-engineer algorithms used by governments to examine if they are improperly implemented, and take advantage of the affordability of sensors to gather their own data. This new form of journalism, fueled by data science, has been coined as ‘Algorithmic Accountability’.
Where data science becomes a problem is when models fail to account for human biases. For example, private companies who use inaccurate algorithms to automate the sorting of resumes might result in a bad pick for a job, or a discriminatory hiring process.
Legal consequences can also ensue from algorithmic bias or inaccuracy. As journalist Julia Angwin discovered this May, when governments use algorithms in justice systems, inaccurate models can result in the release of dangerous criminals, or the incarceration of innocent people. Focusing on the predictive algorithm of future criminality, Angwin’s investigation showed that a risk assessment tool used in several state judicial systems was twice as likely to incorrectly flag black defendants as future criminals than white defendants.
Part of the problem is that governments often outsource the creation of such algorithms to private companies, who guard their algorithms fiercely. This means that citizens or defendants have very little ability to challenge the accuracy of their risk scores. To overcome this stumbling block, Angwin’s team obtained 7,000 risk scores of defendants and independently checked to see how many were charged with new crimes in the next two years. Using data science, Angwin was able to uncover discrepancies that private companies might not be willing to release.
In a time when the profession of journalism is undergoing radical changes, the tools of data science give many journalists renewed ability to interrogate the powerful.