When robots meet commercial reality

“The ChatGPT moment for general robotics is just around the corner,” Nvidia’s Jensen Huang said in January last year. This January he said the ChatGPT moment for physical AI was “nearly here”.

ChatGPT was the fastest-growing consumer app in history, according to a report from UBS, reaching 100mn monthly active users within two months of launch. If that sort of explosive adoption is what Huang meant by a “ChatGPT moment” then I think we have much longer to wait.

There have, of course, been advances in the deployment of robotics over the past year, not least the steady and impressive progress of self-driving cars. But there have also been some retreats. US supermarket chain Kroger, for example, announced in November it would close three of its eight robotic warehouses, which pick and dispatch orders to customers. At the same time, it said it was expanding its relationship with gig economy companies like Instacart and DoorDash, which marshal self-employed humans to act as “personal shoppers” for customers.

It’s tempting to ask who needs automation when you can just have a low-paid gig worker nip to the shops for you. Although that’s not entirely fair, the story does offer an insight into what happens when robots meet commercial realities.

First, while “adopting” generative AI as a business can be as quick and simple as paying a monthly subscription to OpenAI or Anthropic, implementing robotic systems generally requires planning, time and money. The Kroger warehouses (powered by technology from UK company Ocado) required huge amounts of expensive kit. In a low-margin business like selling groceries, that means you need to forecast demand very accurately in order to maximise utilisation.

“It’s a bit more like a factory set-up: it needs to have minimum throughput to make money,” Tom Andersson, a warehouse automation expert at research company STIQ, told me. “In the end, you need to have a really good business case for why you do automation, and when you do those business cases — because sometimes these projects can take three years in the planning — if your forecast is wrong at that point it will be tricky.”

That is not to say that automation in the grocery sector is a non-starter. The business case for Ocado’s automated warehouses in the densely populated UK, for instance, seems to stack up better. In the Netherlands, an online grocery company called Picnic has made headway with a similar approach. And plenty of supermarket chains have invested in automation further up the supply chain. Walmart has increased revenues by more than $150bn over the past five years, yet its headcount has come down slightly over the same period. It is still a vast employer, though, with more than 2mn staff on its books.

Robot enthusiasts believe a new breed of humanoid robots will make further inroads into physical work. One of the great advantages of building robots that look like humans is that — in theory — you can slot them seamlessly into workplaces that are already built for humans. That could make the deployment of robots more gradual and flexible, and less of an enormous upfront bet.

But even if humanoids do achieve human-level capabilities this year, as Huang predicts, there will still be practical problems for businesses to consider. First, to deploy AI-powered humanoid robots alongside humans in the workplace, you need to be confident they are safe. In a white-collar setting, a hallucinating chatbot might make up a number or reference and damage your reputation or your bottom line. In a blue-collar setting, a malfunctioning humanoid could be physically dangerous.

Second, humanoid and canine robots just don’t have much stamina for now, because of limitations with battery technology. As James Pikul, associate professor of mechanical engineering at the University of Wisconsin-Madison has written, the Boston Dynamics robot Spot can only operate for about 90 minutes before it needs to recharge. By way of comparison, it is common for human staff in factories and warehouses to work 10-hour shifts with a couple of breaks.

What does all this add up to? There is every reason to believe that automation will continue to increase in a range of physical settings. But technologists and investors should take care not to conflate technical breakthroughs with commensurate transformations of the real economy. As Kroger’s retreat shows, just because a technology exists and works does not mean it will always be commercial to deploy it. In other words, the fact a job can be automated doesn’t necessarily mean it will be automated.

[email protected]