#GeneralsThoughtofTheDay: I watched/listened a Congressional testimony last week while catching up on my work email. During the line of questioning, a member of the House of Representatives said something that caught my attention, "AI codes better than humans." I pivoted my full attention to the subsequent conversation between him and the government witnesses, both of whom didn't challenge the declaration. Had I been in the hot seat, I would have been tempted to say, "Sometimes AI codes better than humans. Sometimes it doesn't. When you assume the AI is always right, you are always wrong." 

Those of us in the cyber research arena participating in AI-related R&D are highly encouraged by the accelerating capabilities of AI-enabled systems but also are careful to identify for resolution vulnerabilities and flaws that potentially increase risks for consumers of these exquisite systems. Our researchers here at Carnegie Mellon University and the Software Engineering Institute are at the forefront of AI engineering, AI security, and Adversarial AI research. We've seen the accuracy of AI-enabled systems increase dramatically, but they aren't close to perfect, nor do they always have to be. 

There are many types of AI systems and numerous use cases that require various levels of precision, accuracy, veracity, and safety. So I'm not considered too biased for the home team (which I am), here's a study from a team at Stanford that illustrates that AI coding isn't quite perfect. (see: https://lnkd.in/ePRkHfKV.) Admittedly, that article is a bit older, but it is still noteworthy as it showed that the humans in the study thought the AI was correct and had a sense of overconfidence in the security and efficacy of the codebase. 

I like to think of AI as "Augmented Intelligence", i.e. assisting humans perform tasks that the human cannot do as fast due to scope and size (or interest level.) I also believe that when it comes to decision-making, that's a human function; an AI system can help process information, but the responsibility for making a decision on that information rests with a human. Does AI code better than humans? I'm in the camp that says sometimes, but not always. How about you?

hashtag#CISO hashtag#CIO hashtag#CERT hashtag#SEI hashtag#leadership hashtag#AI hashtag#software hashtag#softwareengineering hashtag#softwaredevelopment hashtag#AugmentedIntelligence hashtag#DoD hashtag#SoftwarePathways hashtag#SoftwareModernization