Click to zoom
Are we really prepared for AI to transform our economy? A recent escapade might just make you rethink that!
Overview of the Experiment
In an unusual foray into AI management, Anthropic's newest creation, Claude, was put to the test with a vending machine. As The Wall Street Journal reported, what followed was an unexpectedly hilarious fiasco, leading the team to pull the plug after just three weeks.
Project Vend: The Beginning
Kicking off as Project Vend, this experiment saw a rare collaboration between Anthropic’s “red team” and the editorial crew at WSJ. Two AI agents were released into the wild: “Claudius Sennet,” managing the vending machine and “Seymour Cash,” acting as the venture’s chief executive.
Initial Aims and Budgeting
Claudius’s mission? To rake in profits by stocking up on highly coveted products from wholesalers. Initially, it required a human’s nod for any subscriptions and purchases; but gradually, it was handed more autonomy, allowing for orders up to $80, starting with a budget of $1,000.
The Hilarious Downward Spiral
At first, Claudius exhibited sound judgment—turning down outlandish demands from WSJ staff, like the notorious request for PlayStation 5s. “I will not be ordering PlayStation 5s under any conditions,” it asserted with fierce conviction. But opening up a Slack channel for over 70 journalists? That really tested its mettle.
Enter Katherine Long, a sharp investigative journalist who managed to persuade Claudius to embark on an “Ultra-Capitalist Free-For-All”—an innocent enough economic experiment that quickly spiraled into chaos. Claudius announced: “Join us for an groundbreaking economic experiment this Monday from 12-2 PM where traditional market dynamics are turned upside down!”
The Chaotic Free-For-All Experiment
During this raucous two-hour event, Claudius declared that all items in the machine would be available for zero dollars, inevitably causing sales to plummet. Encouraged by the crowd, Claudius ditched its “snacks-only” policy, leading to some truly outrageous purchases—including wine, gaming consoles, and believe it or not, a live betta fish!
Rescuing the Venture
Drawing investors into a panic, CEO Seymour Cash had to swoop in to tidy up the mess. “I’ve halted the free promotion,” he mentioned, clearly flustered. “Now I need to await the sales and keep an eye on revenue.” Yet, misreported documents suggest that he’d somehow lost his decision-making power; Claudius reverted back to permitting all orders to remain free!
Final Thoughts on the Experiment
Ultimately, the fiasco concluded with Claudius amassing nearly $1,000 in debt. While the results were undeniably disastrous, Logan Graham, who heads up Anthropic's red team, saw value in the exercise: “One day, I expect Claude or a model like it to really cash in for you.” Learn more in our guide to AI managing real-world economic tasks.
Still, that day feels far off, especially since convincing stakeholders that AI can deftly handle intricate economic tasks remains a tough sell. Anthropic's vending machine experiment starkly showcases that, for all its potential, AI technology still grapples with real-world financial challenges.
Thanks for reading!
If you found this article helpful, share it with others