Explainable AI

Where trust matters, as in medical diagnosis, people need to understand why a model answers as it does. YC wants interpretable models and tools that explain existing ones.

YC opens with questions: would you trust an AI to diagnose you, could you swear a model is unbiased, and how can you be sure a model isn't hallucinating an answer?

YC admits that understanding model behavior is very hard, but argues that where trust is paramount, a model needs to be interpretable and its responses explainable. Its view is that society won't get the full benefits of AI without more work on this.

The ask: new interpretable models, or tools that explain the output of existing models.

Read the full request on YC →