Because AI isn’t (yet) able to physically restock the machine, the AI model could email company employees who handled such tasks. Beyond that, however, the AI model, dubbed Claudius for the experiment, was tasked with many of the responsibilities of a traditional operator, including selecting and maintaining inventory, setting prices and maximizing profit.
The upshot: “If Anthropic were deciding today to expand into the in-office vending market, we would not hire Claudius,” the company wrote in its blog.
The experiment showed that while the AI model was effective at tasks such as identifying suppliers, adapting to users’ requests and “jailbreak resistance,” as Anthropic employees tried to trick Claudius into stock sensitive items, Claudius failed as a convenience service operator because it ignored profitable opportunities, instructed customers to make payments at a Venmo address it had imagined (instead of the one created), sold products at a loss, offered excessive discounts and mismanaged inventory.
Although version one of Project Vend wasn’t successful at the bottom line, Anthropic predicts that AI middle managers will come to pass. “It’s worth remembering that the AI won’t have to be perfect to be adopted; it will just have to be competitive with human performance at a lower cost in some cases,” the company wrote in its blog.
Read the full story here.
