Grant Fleming to present AI Alignment Considerations for Data Science Practitioners.
Accounts of AI algorithms making biased, unfair, or otherwise harmful predictions are capturing growing attention from the media, institutional researchers, and the public at large. In the vast majority of cases, these harmful predictions arise as the unanticipated and unintended consequence of using large, highly complex model to solve poorly specified tasks. Research around solving this "AI alignment problem" between the results desired and actually produced is quite nascent. Fortunately, this research has already yielded some best practices and software tools that practitioners can begin utilizing in their own workflows.
This session will discuss a selection of best practices and software tools to mitigate the risks of harmful predictions in certain classification and NLP contexts.
For more information about this event and UMBC's Data Science program.
This lecture will be delivered via UMBC's WebEx platform.