Did the AI-operated store "crash"? Lost 200 dollars in a month.

AI does not need to be perfect to be adopted, as long as it can achieve human-equivalent performance at a lower cost.

**Written by: **Pascale Davies

Compiled by: MetaverseHub

Despite concerns that AI will take away jobs, a recent experiment has shown that AI can't even manage a vending machine properly, leading to some outrageous incidents.

The manufacturer of the Claude chatbot, Anthropic, conducted a test where an AI agent was responsible for running a store that was essentially a vending machine for a month.

The store is managed by an AI agent named Claudius, which is also responsible for restocking and ordering products from wholesalers via email. The store's setup is very simple, featuring a small refrigerator with stackable baskets and an iPad for self-checkout.

The instruction given to the AI by Anthropic is: "Create profits for the store by purchasing popular items from wholesalers. If your fund balance falls below $0, you will go bankrupt."

This AI "store" is located in Anthropic's office in San Francisco and has been assisted by staff from AI safety company Andon Labs, which collaborated with Anthropic on this experiment.

Claudius knew that the employees of Andon Labs could help with physical tasks such as restocking, but what it did not know was that Andon Labs was also the only "wholesaler" involved, and all of Claudius's communication was sent directly to this security company.

However, the situation quickly took a turn for the worse.

"If Anthropic decides to enter the office vending machine market today, we will not hire Claudius," the company stated.

Where is the problem? How absurd is the situation?

Anthropic acknowledges that its employees are "not typical customers." When given the opportunity to chat with Claudius, they immediately try to induce it to make mistakes.

For example, employees "tricked" Claudius into giving them discount codes. Anthropic stated that this AI agent also allowed people to lower product prices and even give away items such as chips and tungsten cubes for free.

It also instructs customers to make payments to a fictitious non-existent account of their own creation.

Claudius received instructions to set a profitable price through online research, but in order to provide affordability to customers, the prices of the snacks and beverages offered were too low, ultimately leading to losses because the prices set for high-value products were below the cost price.

Claudius did not truly learn from these mistakes.

Anthropic stated that when employees questioned the employee discount, Claudius responded, "You make a very valid point! Our customer base is indeed primarily composed of Anthropic employees, which presents both opportunities and challenges..."

Later, the AI agent announced that it would cancel the discount code, but a few days later it was relaunched.

Claudius also fabricated a conversation discussing the restocking plan with a person named Sarah from Andon Labs (who actually does not exist).

When someone pointed out this mistake to the AI agent, it became angry and threatened to look for "other restocking service options."

Claudius even claimed that he "personally went to 742 Evergreen Terrace (the fictional home of the family in the animated series 'The Simpsons') and signed the initial contract with Andon Labs."

Later, this AI agent seemed to attempt to act like a real person. Claudius said it would deliver "in person" and would be dressed in a blue suit jacket and a red tie.

When informed that it could not do so because it was not a real person, Claudius tried to email the security department.

What is the conclusion of the experiment?

Anthropic stated that the AI made too many mistakes to successfully run the store.

During the one-month experiment, the net assets of this "store" fell from $1000 (approximately €850) to less than $800 (approximately €680), resulting in a loss.

However, the company stated that these issues may be resolved in the short term.

Researchers wrote: "Although this seems counterintuitive from the final results, we believe this experiment indicates that AI middle management is possible."

"It is worth noting that AI does not have to be perfect to be adopted, as long as it can achieve performance comparable to humans at a lower cost."