EDGE keynote: Best way to apply AI to supply chain is in constrained spaces like DCs and yards

One of the biggest tech trends of 2023 has been the rise of new applications in artificial intelligence (AI), as generative AI tools like ChatGPT and Bard join more embedded examples in cars and mobile phones. Already, many companies have started questioning whether that new technology can help relieve the pressure of national labor shortages by automating or replacing certain work roles.

So is your own job at risk? Here's the good news: if you’re a logistics industry professional attending the CSCMP Edge conference in Orlando this week, the answer is almost certainly “no,” according to keynote speaker Missy Cummings, a professor at George Mason University who is director of the Mason Autonomy and Robotics Center (MARC). In fact, the answer is probably also “no” for the Uber driver who brought you here from the airport and for the truck driver who hauled the equipment for your company’s booth on the exhibit floor, she said in a session titled “AI reality check: truth vs. hype in the stampede toward AI.” 

That’s because all those activities can be ranked along a spectrum from the most basic jobs—which require simple reactions to obvious inputs—to the most complex, which demand “judgement under uncertainty,” she said. Cummings defines that progression by four mileposts labeled skill, rule, knowledge, and expert (known as the SKRE framework) and says that by that measure, no modern AI platform can progress past a “brick wall” at level two.

The reason is that both neural network-style and large language model (LLM)-style AI platforms produce their outputs through sheer guesswork, not sentient reasoning. “No AI can come up with something it hasn’t seen before. It has to have at least two data points, and then it can interpolate between them, but it needs those data points. It is seeded by human thought,” Cummings said.

In logistics examples, Cummings said that approach works fine for tasks like constrained driving with strict limits, such as: sidewalk robots used for last-mile food delivery, unmanned dumptrucks carrying ore through mines, robots moving shelves of inventory through an Amazon DC, or yard trucks towing containers within a port or warehouse property. 

But it falls far short when used to operate taxis in crowded cities or autonomous trucks on public highways, she said. “I’m not saying self-driving cars and trucks are never going to come, but if you’re willing to bet the farm on self-driving trucks to power your supply chain, you are in serious trouble,” said Cummings. “And if you use large language models at any kind of critical path in your company, let me know so I don’t invest in your company, because you’re going to have serious problems.”