Scientists at the University of Surrey want to banish the danger of overly self-reliant systems.
Researchers at the University of Surrey have developed software that verifies what and how much information an artificial intelligence (AI) has gleaned from a company’s digital database. According to the paper, the new verification software can be used as part of a company’s online security protocol, helping to understand if an AI has over-learned or even accessed sensitive data.
Important step for record integrity
“It’s taken us years to develop software that can figure out what an AI knows. By being able to verify what the AI has learned, we can give companies the confidence to make the most of the power of AI without security concerns. This software is an important step in maintaining the privacy and integrity of data sets,” adds Adrian Hilton, director of the Institute for Human-Centered AI at the university.
According to the experts, the innovative algorithm can even detect whether the AI has identified and is exploiting bugs in the software code. For example, it could determine whether an AI has learned to always win at online poker by exploiting a coding error.