Deep Tech21 Nov 20262 min read

Researchers Teach AI Agents to Mimic Human Flaws to Predict Supply Chain Panics

Northeastern University researchers used cognitive grounding to make AI agents replicate human supply chain panics, which could help test public policies safely.

Listen to this storyAbout 3 min, read by your device
Picture made with AI

When unexpected events shake consumer markets, fear often sparks shortages. In early 2020, Australian shoppers purchased nearly 40 million rolls of toilet paper from one supermarket chain in a single week. Similar chain reactions occurred in 2012 when news of Hostess closing caused people to hoard and resell snack cakes at extreme markups. Because human buyers and suppliers react to one another, small disruptions can trigger wide-scale supply chaos.

Training AI on Human Imperfections

Researchers at Northeastern University have found a way to make AI models recreate these chain reactions. In a study published in Nature Computational Science, professor Babak Heydari and graduate student Hongan Zhu explored how AI can simulate collective human decisions, including common mental errors.

The researchers tested agents from seven AI models in the classic “beer game.” This exercise simulates a four-stage supply chain involving a retailer, wholesaler, distributor, and factory. Each player must manage inventory to meet demand while dealing with shipping delays. Left on their own, the AI bots made strange decisions, such as continuing to over-order long after customer demand had stabilized.

The Power of Cognitive Grounding

The team resolved this by introducing cognitive grounding. Instead of giving explicit purchasing orders, they gave the agents descriptions of how people actually think. These prompts included human tendencies to misjudge future demand, panic during shortages, or ignore items already in transit. They also tested prompts describing feelings players might experience, such as pressure to overcorrect or uncertainty.

Across 231 test sessions, agents from six of the seven models altered their purchasing patterns to closely match human behavior. The bots internalized human tendencies purely through natural language guidanceGuidanceWhat a company says it expects to earn in the coming months. Open in glossary.

Testing Policies Before Implementation

Realistic agent-based models allow planners to test high-stakes scenarios without spending large amounts of money or placing real people at risk. Officials can model city evacuation routes to ensure instructions do not cause traffic jams, or evaluate how public transit fare cuts affect ridership. Businesses can also simulate store relocations to see how customer habits change before committing capital to a move.

How does this story land?

One tap per reader.

Quick poll

Should city planners rely on AI simulations to design emergency evacuation routes?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. Tech Xplore - electronic gadgets, technology advances and research news, Can teams of AI agents recreate the chain reaction of human decisions?
Words explainedGuidance
Up next in Deep TechBepiColombo X-Ray Telescope Fires Up After 10 Billion Kilometers2 min read