The role and credibility of Artificial Intelligence (AI) in intelligence agencies has recently been voicing skepticism. This has been in response to a number of inaccuracies displayed ranging from ChatGPT’s+ flawed responses to factual questions, to facial recognition algorithms’ consistent inability to correctly identify black individuals, to other bizarre flaws, including Stable Diffusion’s generation of human hands with an unusually high number of fingers.
However, these pitfalls do not shift the reality that AI can still be strategically adopted by government agencies, as long as any AI-generated insights are held at an arm’s length and subjected to high levels of scrutiny. In the right contexts and when viewed with a healthy level of doubt, AI can still provide some form of usefulness, proffered Nand Mulchandani, the CIA’s Chief Technology Officer. He suggests a perspective shift—a more suitable way to regard AI, according to him, is as if it were a “crazy drunk friend.”
AI can essentially force analysts to consider a different perspective and potentially disrupt their usual modes of thinking, or as Mulchandani calls it, their “conceptual blindness.” Nand Mulchandani, who holds the distinction of being the Agency’s first-ever CTO, having previously acted as the chief of the Pentagon’s Joint AI Center, shared this view at the Billington Cybersecurity Summit.
Mulchandani further noted how AI excels in pattern recognition tasks which involve processing large volumes of data. The CIA and other intelligence agencies, which often find themselves literally data-drowned, could potentially benefit from such capabilities. However, areas that demand precision remain challenging for AI.
This is largely because the algorithms in widespread use principally function by identifying statistical correlations, thereby pinpointing what’s probable as opposed to what’s certain. The outputs of these AI models can therefore be profoundly unpredictable, given that they calculate probabilities rather than certainties.
Despite the concerns surrounding AI, it’s this unpredictability that can be particularly beneficial as it can surface new and unexpected insights. Experts, the longer they’ve been specialized in a certain domain, are at risk of “conceptual blindness” and could benefit from the fresh perspectives that AI’s unpredictability can introduce.
In conclusion, Mulchandani advocates for a more discerning integration of AI where logic-based systems provide different benefits for different types of problems. The uniform application of AI to all types of situations—what he refers to as treating it like “peanut butter”—would prove unsuitable.
Access the original article here.