3:30 pm ERC 161
Brian Nord, Fermilab and UChicago, "AI In the Sky: Implications and Challenges for the use of Artificial Intelligence in Astrophysics and in Society"
Artificial Intelligence (AI) refers to a set of techniques --- like machine learning, deep learning, and data science --- that rely on the data itself to develop models of observed phenomena. AI algorithms have a long history of development, and there has been a recent resurgence in their research and deployment, marked by extraordinary results in many contexts, including scientific ones. However, these algorithms are far from a panacea for our challenging data-modeling tasks.
Three major changes have revolutionized the role of data in our lives: 1) the increased availability of large data sets; 2) advancements in computing hardware; and 3) insights for new mathematical and algorithmic techniques. Because of these elements, AI now permeates society --- from the promise of self-driving vehicles to entertainment choices to cancer-detection and criminal justice. Moreover, in the last few years, it has had substantial impacts on molecular chemistry, particle physics, and more recently astronomy. AI is more than likely here to stay, as well as grow as an important technique in our science toolboxes.
Nevertheless, AI has significant challenges for reaching its full potential for scientific impact --- namely, uncertainty quantification, interpretability, bias, and hybridization with physical models. These challenges also plague implementations in other contexts throughout society. Given these challenges, how do we implement these algorithms in a responsible, careful, and systematic way?
We'll discuss these topics in the context of deep learning and its application to modern astronomical surveys. Finally, we'll discuss the implications for the widespread use of AI in society.