How AI Helps Find Cause and Effect in Messy Data

Scientists built a new way to spot cause and effect when the data is not neat numbers. They call it GenAI‑Powered Inference, or GPI. The idea is to use smart AI that can read text and look at pictures. This helps researchers figure out if one thing really makes another thing happen.

GPI starts with AI models that are already trained and free to use. These models turn a chunk of text or an image into a short list of numbers that still keep the main pattern. Because the models are not changed, the process is fast and does not need heavy computing. Then regular statistical tools work on those number lists to estimate effects and show how sure we are.

Three examples show how GPI works. First, researchers looked at Chinese social media posts to measure the impact of censorship while removing the influence of other words in the posts. Second, they took pictures and separated the effect of a specific visual detail from other details that usually appear together. Third, they examined political speeches to gauge how persuasive they are, using the same number‑based approach.

Anyone can try the method because the team released an open‑source software package. The package includes the steps to turn raw text or images into the number lists and run the causal analysis. It works on a regular laptop and needs no special AI expertise.