DARPA program applies 'trust but verify' to AI
Under the Defense Research Project Agency's (DARPA) Explainable Artificial Intelligence program (XAI), Raytheon BBN Technologies is developing a first of its kind neural network that explains itself.
The XAI program aims to create a suite of machine learning techniques that produce more explainable models while maintaining a high level of performance. It also aims to help human users understand, appropriately trust and effectively manage the emerging generation of artificially intelligent partners.
Raytheon BBN's Explainable Question Answering System will allow AI programs to 'show their work,' increasing the human user's confidence in the machine's suggestions. "Our goal is to give the user enough information about how the machine's answer was derived and show that the system considered relevant information so users feel comfortable acting on the system's recommendation," said Bill Ferguson, lead scientist and EQUAS principal investigator at Raytheon BBN.
EQUAS will show users which data mattered most in the AI decision-making process. Using a graphical interface, users can explore the system's recommendations and see why it chose one answer over another. The technology is still in its early phases of development but could potentially be used for a wide-range of applications.
"A fully developed system like EQUAS could help with decision-making not only in DoD operations, but in a range of other applications like campus security, industrial operations and the medical field," said Ferguson. "Say a doctor has an x-ray image of a lung and her AI system says that its cancer. She asks why and the system highlights what it thinks are suspicious shadows, which she had previously disregarded as artifacts of the X-ray process. Now the doctor can make the call – to diagnose, investigate further, or, if she still thinks the system is in error, to let it go."
As the system is enhanced, EQUAS will be able to monitor itself and share factors that limit its ability to make reliable recommendations. This self-monitoring capability will help developers refine AI systems, allowing them to inject additional data or change how data is processed.