The Indian American

Two events, in particular, stage the conversation around AI ethics. The first involves an incident, in March 2018, when one of Uber’s driverless cars landed in a fatal accident. Forbes reports that the emergency brakes in the car were disabled, and as a result, the car did not stop for a pedestrian, and the backup driver was not alerted in time to change direction. The second event is more of a cultural trend—it’s more about how our soci- ety is becoming increas- ingly dependent on apps— Yelp!, Google Maps, Facebook, Tinder—to make decisions for us. Apps, driverless cars, drones all have the same thing in common: they run on algorithms that, in renowned journalist, John Markoff’s words, feed off “vast pools of data [that have] remarkable oppor- tunities to run off the rails.” The need for transparency has now become the sin- gular battle-cry for propo- nents of AI ethics. The quality of data used to train machines and the algorithms used to reach a decision must become more visible. As Markoff puts it, “if you want the machine to behave in an ethical manner, you have to know why it does what it does.” IBM’s leader of AI ethics, Francesca Rossi, advo- cates for a holistic audit process: “My vision for AI is an audit process — a very scrupulous process through which somebody credible has looked at the system and analyzed the model and training data.” If such opaque data-shar- ing policies continue, then an impartial audit must be enforced to mitigate the impact of inevitable biases in datasets. Investing in steps to audit training data would be a major leap towards more ethical builds. A few months ago, Amazon came under heat by the ACLU for its Rekognition technology. The plan was to use this software to identify suspects in legal matters; the ACLU expressed concerns that such facial recognition technology would lead to wrongful accusations or arrests that put people of color at a greater risk. 23 THE INDIAN AMERICAN OCTOBER-DECEMBER 2018