AI Now Helps Decide Which Projects Get Your Donations

Manifund is a crowdfunding platform where people post research projects and ask for donations. With hundreds of proposals, it's hard for donors to know which ones are worth funding. So researchers tested whether an AI model (Claude Opus) could read each proposal and predict which ones would reach their funding goals. They fed the AI 530 proposals and asked it to rate each on a 1-5 scale and estimate the chance it would get funded.

The AI turned out to be pretty good at predicting funding success—its ratings matched actual outcomes about 90% of the time, much better than random guessing (which would be 50%). But here's the catch: a much simpler model that just looked at word frequencies did almost as well (88% accuracy). That suggests the AI might not be understanding the proposals deeply; it might just be picking up on surface patterns.

Still, the researchers prefer the AI approach because it gives reasons for its ratings, which humans can evaluate. The simple word model could also be easily gamed—people could just stuff their proposals with certain words to get a high score. The AI is harder to fool, though not impossible.

The experiment is a step toward using AI as a helper, not a replacement, for human judgment in philanthropy. The researchers emphasize that AI is best at flagging proposals for a closer look, not making final funding decisions. As AI models improve, they could become even better at spotting high-potential projects, helping donors make more informed choices.