Predicting Human Behavior with AI Simulations

Today, many companies are using generative AI to streamline processes, summarize data, and assist with coding, among other things. They are exploring how companies can embed agents in processes (leading to changes in business models) and what it means to have digital colleagues, all based on LLMs—large language models that are trained on massive amounts of human-generated data. One of the interesting things companies can use LLMs for is to simulate how people (customers, competitors, partners) might behave in markets or strategic settings. The challenge, though, is that these simulations are inconsistent; sometimes they look strikingly human, and other times they fail badly. In this video presentation, Ben focuses on methods for making AI simulations more reliable, and on experimental evidence showing when and why those methods work.