Speaker: Kostopoulos, L. (LK Cyber)
Date: 29 January 2019
SMA hosted a speaker session presented by Dr. Lydia Kostopoulos (LK Cyber) as a part of its SMA DHS Speaker Series. To begin, Dr. Kostopoulos spoke about humans’ overconfidence in the validity of technology. She stated that the data that we provide artificial intelligence (AI) devices reflects our personal biases. Therefore, we need to be cautious and skeptical when putting our trust in algorithms. Dr. Kostopoulous also explained that, although accumulating a sufficient amount of data before trusting the validity of AI is important, more data is not always the solution; we will often need to know when and how to correct algorithms to eliminate inherent biases embedded within the data. She proceeded to present the ways in which AI has been and could be used for adversarial purposes. Dr. Kostopoulos also discussed the concerns surrounding the reduction of human agency as decisions are increasingly delegated to algorithms. After discussing her Lethal Autonomous Weapons Systems (LAWS) Human Involvement Table in detail, she emphasized that we should not assume that our algorithmic approaches will be identical to those of our adversaries. Just as our doctrine is different, so will the thinking behind the data organization, algorithmic design, and adversarial examples. She also stated that we must adequately prepare for the unexpected to occur when developing AI technologies. Dr. Kostopoulos concluded her brief by stating that commanders, policy makers, and scientists must not only think about what AI can do to benefit humans; they must also think about the risks involving the use of AI.
Unfortunately, due to a technical error, an audio file for this speaker session is unavailable.