Predictive artificial intelligence is likely to be “more important” than generative AI in the future for defense uses and resiliency, a top official from the Defense Information Systems Agency (DISA) said on Tuesday.
Speaking during the Rubrik Public Sector Virtual Summit powered by MeriTalk, Stephen Wallace, the DISA chief technology officer (CTO), said that despite not garnering as much attention as generative AI, predictive AI is “likely to be more important as we go forward.”
Wallace said that as larger data sets are generated, the predictive models will likely be able to predict attacks and operational trends, as well as find “needle in a haystack” techniques that adversaries use in cyberattacks.
“[By looking for] the low and slow techniques that we know that adversaries are using – and using in some cases quite effectively – predictive AI has a significant opportunity for blossoming even further, and in my mind, possibly outrunning or outstripping generative AI in the long run,” said Wallace.
Generative AI responds to a user’s prompt or request with generated original content and is trained on large volumes of raw data. Predictive AI combines statistical analysis with machine learning to find data patterns and predict future outcomes using historical data.
AI models will play an overall increasingly important role in resiliency and security, Wallace said, saying that despite the “mistakes” made at the dawn of the internet by not building in security measures, AI has helped to “open up a number of doors” in strengthening resilience.
This, he noted, is increasingly more important as the threat of quantum computers nears – which will break most forms of encryption through more powerful computing than systems in use today.
“The investment that we’re all making in trust, risk, and security and AI models is going to be an important factor as we go forward,” said Wallace.
The CTO said generative AI will also play a role by empowering cyber analysts that handle “millions of attacks” by responding and identifying them, and by building general generative models.
