To some people, generative AI can feel like magic, and perhaps it's meant to be that way. Even the experts who designed it don’t fully understand how neural networks are thinking on their own, which is quite magical. 

But if Dr. Balaji Padmanabhan could wave a magical wand over our awe of AI, he’d utter a revealing spell. “The beauty of magic is that even magic is not magic. And that’s a perfect metaphor for AI. Under the hood, there is a very specific way in which it works, and it’s extremely important for all of us to understand the capabilities of AI, rather than just walking around thinking it can do anything under the sun.”

Padmanabhan, a professor of decision, operations and information technologies at the University of Maryland, led a thought leadership spotlight at From Day One’s Washington, D.C. conference last month. He spoke on the promises and risks of AI and the next frontier in the domain of HR and hiring.

In a fundamental shift from earlier approaches, today’s AI represents a dramatic evolution in machine learning, says Padmanabhan. “AI, over the years, has picked problems that demonstrate intelligence.” And while use cases appear similar—chatbots existed 60 years ago and remain prevalent today—the underlying technology has evolved exponentially.

Earlier AI relied on “pre-programmed logic,” while modern systems operate through “learning from continuous, massive data streams, as well as new learning paradigms.” The implications are significant. “Potentially, we are at a time when AI can be not just as good as humans, but potentially much better,” creating both “problems as well as concerns” for society.

Zeroing in on how AI will outperform humans, Padmanabhan pointed to two broad issues. The first revolves around predictive models using AI and the “edge cases” it applies to.

“Predictive models are very good when the future is like the past. So what happens when the future is not like the past? The data that we've trained [AI] on isn't going to be adequate to make those predictions.”

You can think about predictability in terms of what physicians do during a majority of their workweek. While most of a doctor's routine work could be automated—perhaps 39 hours of a 40-hour workweek—their true value emerges in those critical moments of specialized judgment.

“That 20 minutes or one hour in that 60-hour workweek where they make a decision that's different from what something would have automated is what that person is getting paid for.”

This reality shapes how we should approach professional development, and Padmanabhan suggests we "focus on that aspect in our own professions,” specifically, the ability to excel “where the future is not like the past,” he said.

Superpowered AI and Where Humans Remain

Padmanabhan cast a revealing spell on broad assumptions about AI's superiority by highlighting the disconnect between publicized tests and real-world applications. “Where they show AI is better than humans most of the time doesn’t reflect on-the-ground use cases.” 

He bases this reality on his experience helping companies implement AI solutions. “When a company is actually using AI within an HR context, or when a hospital is actually using AI to help a doctor make decisions, those situations are not the same as taking an automated exam.”

“So take these predictions with a big grain of salt that AI can replace humans.”

Another false reality to consider is AI’s reliability. “AI makes mistakes just as humans make mistakes.” And while many organizations have developed robust systems to understand and minimize human error, “most on-the-ground use cases [with AI], it's hard for them to even come up with a number which says what percentage of the time AI is making a mistake.”

Despite some areas of concern, Padmanabhan remains optimistic about AI's potential as a personalized workplace tool, or what he refers to as the “superpower” we’ll all have in our pockets. 

“Regardless of what your job is, whether you're a wealth manager, whether you're an information analyst, whether you're an investment banker, everything that you're doing can be supported by AI in a very significant way, and that’s the superpower in our pockets that we have to try to go towards.”

AI Agents and the Next Frontier

Despite early missteps in automated hiring, Padmanabhan foresees AI continuing to be more useful and practical in HR. For example, HR vendors now use generative AI to automate specific tasks rather than entire workflows. These targeted applications include resume filtering and automatically preparing initial performance review reports—tasks that previously consumed significant time but can now be completed “in a matter of seconds.”

One particularly promising application incorporates nudge theory—a behavioral science concept where subtle, positive interventions guide people toward better choices without removing freedom of choice. AI can deliver these “right pushes at the right time” to enhance employee performance.

“Technology can very easily become our friend. That's what I meant when I was saying it’s a superpower in our pockets to help all of us do our jobs much better.”

Editor's note: From Day One thanks our partner, the University of Maryland, for sponsoring this thought leadership spotlight. 

Matthew Koehler is a freelance journalist and licensed real-estate agent based in Washington, DC. His work has appeared in the Washington Post, Greater Greater Washington, The Southwester, and Walking Cinema, among others.

(Photos by Justin Feltman for From Day One)