challenges in current models and formalisms, and
pointed to avenues for future research.
On Monday, February 5, Charles Isbell opened the
day with his talk, “How Machines Learn Best from
Humans.” Isbell noted that “we build machine-learn-
ing systems because we want them to behave a cer-
tain way.” In this case, he said, the “we” is usually
human beings. “Whether we want to convey partic-
ular strategies or subtle preferences that define the
objective itself,” he observed, “some form of knowl-
edge transfer from person to algorithm is always
needed.” Interactive machine learning, he said,
“focuses on techniques for facilitating that transfer
in the context of solving artificial intelligence prob-
lems with machine-learning techniques.” Isbell sur-
veyed some of the problems and techniques studied
in interactive machine learning with a special
emphasis on counterintuitive design principles that
have arisen from the results of experiments with
human participants, especially where those counter-
intuitive principles arise from we being wrong about
us. A surprise guest appearance by Michael Littman
made for a lively and engaging talk.
The evening joint AAAI-IAAI talk was presented by
Zoubin Ghahramani, the chief scientist at Uber.
Ghahramani reviewed the foundations of probabilis-
tic AI, highlighted some current areas of research at
the frontiers, and touched on topics such as Bayesian
deep learning, probabilistic programming, Bayesian
optimization, and AI for data science.
Tuesday’s opening invited talk, “Fair Questions,”
was delivered by Cynthia Dwork (Harvard / Radcliffe
Institute for Advanced Study). The unfairness of algo-
rithms, she noted — for tasks ranging from advertis-
AAAI President Subbaro Kambhampati Delivers His Presidential Address at AAAI- 19.