CITP Seminar: Algorithmic Ecosystems: Understanding Human-AI Interactions from Both Sides of the Algorithm

Click here to join the seminar

---

Machine learning (ML) and artificial intelligence (AI) algorithms constitute a core component of many technology products. Although ML and AI algorithms can be beneficial, they also pose risks to individual users and society as a whole. Most often, the performance of algorithms is assessed using static measures of predictive accuracy. This approach is not only insufficient for reliably and validly estimating model performance, but also provides no information about a system’s ethical risks within a social context. To better understand the positive and negative impacts of AI and ML, we must recontextualize algorithms as embedded within a dynamic social ecosystem in which humans both influence and are influenced by algorithms.

The talk will discuss two ecosystems in which human-algorithm interactions affect broader social outcomes. The first part of the talk will address how ideas from empirical research methods can be leveraged within agent-based simulations to better understand the effects of feedback loops in algorithmic systems. The second part of the talk will address how institutional factors influence technology development in AI startups and how these factors can catalyze or constrain ethical approaches to AI.

Bio: Amy Winecoff is a DataX data scientist at CITP. Her primary interests are in human-algorithm interactions and fairness in machine learning systems. Winecoff received her Ph.D. in psychology and neuroscience from Duke University. After graduate school, she was an assistant professor at Bard College, where she taught neuroscience, abnormal psychology, and research methods. After leaving academia, she conducted research and developed machine learning models for government agencies such as DARPA and the U.S. Air Force to explain and predict human behavior. As a senior data scientist at True Fit and Chewy, she developed product recommendation and search systems. She also conducted quantitative user research to assess how users’ psychology informs their evaluation of algorithmic predictions. Winecoff is passionate about diversity and inclusion in the technology industry.

---

To request accommodations for a disability please contact Jean Butcher, butcher@princeton.edu, at least one week prior to the event.
This seminar will be recorded.

This seminar is co-sponsored by the Center for Statistics and Machine Learning.