In the long run, we want machine learning to help us answer open-ended questions like “Should I get this medical procedure?” or “What are the risks of deploying this AI system?“ Currently, we only know how to train ML systems if we have clear metrics or can easily provide feedback on the outputs. Andreas Stuhmueller, president and founder of Ought, wants to solve this problem. In this talk, he explains the design challenges behind ML’s current limitations, and how we can make progress by studying the way humans tackle open-ended questions.
To learn more about effective altruism, visit effectivealtruism.org
This talk was filmed at EA Global 2019: San Francisco. You can learn more about these conferences at eaglobal.org.