MIT EECS — Morais and Rosenblum Undergraduate Research and Innovation Scholar
Inspiring Curiosity in Genesis to Facilitate Its Learning
Patrick H. Winston
Genesis is an artificially intelligent system designed to model human story understanding. Currently, Genesis analyzes English text by decomposing it into relations among events and applying common sense rules to construct an elaboration graph, via which it searches for concept patterns. However, human comprehension is also crucially aided by our curiosity for the hypothetical and unexplained events in a story. Thus we aim to make Genesis inquire about unexplained phenomena in the story to gain deeper knowledge. To this end, I will implement an algorithm enabling Genesis to generate intelligent queries over a story and obtain answers from humans. I will then implement algorithms based on induction heuristics and machine learning methods to facilitate Genesis to learn new and amend original rules from the feedback.
Last year I engaged in a UROP exploring the Genesis system. This past summer I interned at NASA, researching and implementing machine learning algorithms for multiple kernel learning. After taking 6.034 and 6.036, I have been very intrigued to research how we could design algorithms empowering computers with logic and the ability to learn concepts from previous information and apply the knowledge to interpreting new data