Knowledge and Knowledge Attribution in Neural Networks

Jennifer Nagel (University of Toronto at Mississauga)

Title: Knowledge and Knowledge Attribution in Neural Networks

Abstract: Theoretical discussions of artificial neural networks (ANNs) routinely credit them with knowledge: AlphaGo Zero is said to have “discovered a remarkable level of Go knowledge during its self-play training process” and FaceNet is said to learn to recognize identities from photographs. One might wonder whether these attributions are metaphorical or inappropriately anthropomorphic. If we examine the features of machine learning that are parallel to biological learning, we can defend the attribution of knowledge to ANNs, while also gaining a fresh perspective on the nature of knowledge and the function of knowledge attribution.

The presentation will be 45 minutes, followed by an hour for discussion. The time for the event is: Friday 25 June, 9:00-10:45 am (Melbourne); Friday 25 June, 12:00-1:45 am (London); Thursday 24 June, 7:00-8:45 pm (New York).

This event will be held over Zoom. To receive the link and password, please register at https://acu.zoom.us/meeting/register/tZwucuurqDsoGNNa0R4Va077ZzQan-kFFBin. For any questions, please email [email protected].