Understanding Artificial Intelligence Results to Increase their Value & Avoid Pitfalls

Abstract: Much excitement has been generated over the potential benefits that can be obtained from increasing use of artificial intelligence. This talk will describe points, by reviewing recent cases in a few industries, that should be considered before and while employing an AI system, in order to increase, its utility, as well as to mitigate potential risks. An understanding of the full performance capabilities and risks of an AI system is needed to obtain substantial value from the system. The necessity of use of enough outcome metrics will be demonstrated.
Trade-offs and design decisions associated with different types of performance metrics will be discussed. The import of human training needed to successfully operate an AI system is also illustrated. The quantity, quality, and scope of data used to train and test an AI system, as well as that of the data on which the system is employed, has a substantial effect on the success of the system. Examples will be presented to demonstrate how a comprehensive understanding of these data properties is needed for successful usage of an AI system.

Bio: Dr. Linda M. Zeger leads the design and execution of innovative processes and solutions to derive maximum value from data. Through projects she has led in data delivery protocols, communication and sensor networks, and healthcare analytics, she has developed techniques to substantially improve network efficiency and reliability, guide system operations, and derive key insights, by employing statistical modeling, machine learning, and data analytics.
Dr. Zeger is the founder and principal consultant of Auroral LLC, and she has also held positions at MIT Lincoln Laboratory, Lucent Technologies, Educational Testing Service, and with universities. Dr. Zeger earned a Ph.D. in physics from Harvard University. She is is the author of numerous published papers, and is an inventor on a number of patents.