“Our rules of thumb are no longer up to the challenge.” This description by a Chief Supply Chain Officer of the state of the supply chain he found in his new role stuck with me. The tumult of our times has accelerated interest in artificial intelligence (AI) for help more advanced than rules, and ChatGPT’s release has rocketed this already-explosive growth. Its remarkable capabilities seem like magic. But I’d like AI more to a magic trick, so allow me to direct your attention to the technology behind the curtain to highlight both its risks and its rewards and to share the essential leadership skills AI cannot provide.

Generative AI is like a magic trick

Literally millions of people are using generative AI (and generating lots of FOMO) – ChatGPT had already reached 100 million users by January, the fastest consumer adoption of a technology in history. Part of the draw is its seemingly-magic capabilities, from conjuring songs on any topic in anyone’s voice to acing standardized tests to carrying on spookily-human conversations.

But generative AI is less magic and more like a magic trick, since it can do remarkable things but only under carefully controlled circumstances. A magician can pull a rabbit out of a hat on stage but can’t do repeat the trick again on command in a casual, unexpected encounter. Magic works based on a sequence of steps performed in a specific order to (mis)direct the audience’s attention. Unlike a supply chain, these steps rely on no disruption to their execution.

The mathematical models underlying generative AI are statistical sentence completion machines that compose content based on probabilities, not understanding, context, and empathy. A French history professor illustrated this limitation with a lengthy interrogation he called “My dinners with GPT-4,” showing how the right questions forced ChatGPT into contradicting itself and even irritating him by failing to comply with repeated specific requests.

The power of Generative AI and its promise for supply chains

What dazzles about generative AI is its dips into realms we thought were our sole human domain, such as art and creativity. Any human with a browser can produce this result, since it requires no sophisticated coding skills or knowledge. Essays, poetry, and paintings indistinguishable from human creation are accessible with the move of a mouse. More than mere novelty, its applications in business are exploding. Some of the first test cases range from creating novel protein sequences, personalizing marketing messages, and answering complex legal questions. We are only at the most nascent stages of this technology, and the explosion of generative AI startups will only expand the options.

I’ve seen our own R&D team test some very cool applications and have been following the news closely in search of other examples relevant to supply chain. While there is endless speculation, from my reading two supply chain-specific opportunities intrigued me. The first use case is MIT professor Yossi Sheffi’s idea for using generative AI to monitor all possible sources of risk for a given supplier to mitigate any disruptions. The second is Walmart’s use of chatbots to automatically negotiate relatively routine contracts.

Beyond these targeted examples, I see even more promise in broader applications: code generation, knowledge management, and user interfaces. For many years P&G has had an initiative to foster supply chain “citizen developers” to extend their digital transformation by using their own skills to create tools, rather than relying on IT. Generative AI can write code for someone without coding skills (and then even debug it), a capability with huge potential to expand P&G’s idea. For knowledge management, imagine generative AI primed to intelligently search all of a company’s internal documentation to solve all kinds of problems from “how did we do this before” to “how do I do this now.”

Perhaps the most useful way to think of generative AI is as the future of how we interface with computers. Rather than fearing how it will replace us we might consider how it can connect us more effectively to computers and all their cognitive capacity to search the terabytes of knowledge locked in documents. Generative AI may be the gateway to seamless computer collaboration with a user interface far beyond what we experience today.

The perils of Generative AI are many

If the rewards are enticing the risks are equally daunting. These perils are well addressed elsewhere, but I’ll summarize my four biggest concerns. The first is bias and misinformation. Models trained on human data reflect us, both our pride and our prejudice. The output reflects statistical probabilities of occurrence, not accuracy. This orientation leads to what are termed hallucinations, which can range from citing research that doesn’t exist to telling a reporter to leave his wife. Second, generative AI can arm bad actors more easily with even more powerful tools for malfeasance. Both of these risks are significant and must be mitigated, but they are amplifications of existing risk, not new ones.

Third is the concern about job losses, which is both valid and I hope overstated. I share the sentiment of many economists who posit that historically new technologies result in some short-term displacement but net job creation. The fourth risk is what is sometimes called the “alignment problem,” AI that operates on its own interests that don’t match ours, begetting the Terminator. My belief is that we are a long way from the artificial general intelligence that would truly pose this risk.

The only honest answer to any question about the future of AI is “I don’t know.” The pace is moving so quickly that no one can possibly keep up. So instead I want to direct attention to what I do know.

The keys to succeeding in AI

As a trio of economists argue in a pair of books, most recently Power and Prediction: The Disruptive Economics of Artificial Intelligence, we seek predictions where there is uncertainty. Predictions can be powerful, but they also caution against planning in misaligned silos with what they term the AI bullwhip, because a highly accurate AI-generated demand forecast is pointless if there is no capacity to produce it. Supply chains are rife with uncertainty and silos, but instead of producing useless forecasts or buffering with inventory, we can also adopt best practices like concurrent planning.

What I do know best is that guiding a supply chain in the age of AI requires three leadership skills, all of which are uniquely human.

The first is asking the right questions. Even generative AI needs our help in the form of the emerging field of prompt engineers, humans who improve the software by asking it questions. People are often better at describing symptoms than problems, so the ability to ferret out underlying issues for targeted inquiry is critical. Framing the business problem well is key to solving it, something AI cannot do.

Directing attention to the right things is the second critical skill. Magic and leadership both depend on proper direction of our attention, and FOMO about AI may be distracting you. ChatGPT may be the shiny object du jour, but a sure ROI is investing in people and process to enable technology investments. Many problems land on a leader’s desk – the challenge is recognizing which ones really matter and sustaining attention on them. Generative AI has an “attention mechanism” to help it weigh the importance of words in a sentence, but it cannot match human judgement.

And judgement is an inarguably human leadership skill. As the economist trio of Agrawal, Gans, and Goldfarb explain, AI makes predictions, which are inputs into decisions. We create rules for situations when making decisions is difficult, but AI can increase our productivity and the quality of our decisions by automating the obvious with a more precise prediction, such as predicting lead times. As I noted at the start, many of the rules of thumb undergirding our supply chains are no longer up to the challenge.

We should strive to relieve planners groaning under the weight of tedious decisions but not to drive a “lights out” supply chain. We need to keep the lights on, because the most important decisions still require judgement. There are ample situations where the tradeoffs are complex and unclear, the outcomes risky. It is in these situations where people matter the most, and where we most need the ability to ask the right questions, pay attention to the right things, and exercise the right judgement. And these are all capabilities very human in nature.

Polly Mitchell-Guthrie is the VP of Industry Outreach and Thought Leadership at Kinaxis, the leader in empowering people to make confident supply chain decisions. Previously she served in roles as director of Analytical Consulting Services at the University of North Carolina Health Care System, senior manager of the Advanced Analytics Customer Liaison Group in SAS’ Research and Development Division, and Director of the SAS Global Academic Program.

Mitchell-Guthrie has an MBA from the Kenan-Flagler Business School of the University of North Carolina at Chapel Hill, where she also received her BA in political science as a Morehead Scholar. She has been active in many roles within INFORMS (the Institute for Operations Research and Management Sciences), including serving as the chair and vice chair of the Analytics Certification Board and secretary of the Analytics Society.