AI That Reads Your Values to Spot Bots and Opinions

The trouble with most social media AI is that it sees only what you post or who you reply to — not the deeper reason why you post it. Two people can argue about the same topic for very different moral reasons. One may care about fairness, another about loyalty. This new system, called ValueGraph, tries to capture that moral backdrop.

Here is how it works. ValueGraph maps conversations as a web of who replies to whom. Then it studies the language in those replies to guess which moral values each person leans toward — things like care, fairness, authority, or purity. It doesn't treat those guesses as perfectly true labels. Instead, it uses them as soft hints to nudge the AI into grouping people in a more human-like way.

In tests, ValueGraph performed better than several strong baseline systems, including standard text models and even large language models, on two practical jobs: detecting Twitter bots and identifying a person's stance on a topic, like support or opposition. That means it can potentially tell a heartfelt opinion from an automated troll, or sort public comments by underlying beliefs rather than just keywords.

Why should you care? Every day, social media companies use AI to moderate content, show you relevant posts, and decide who is a real account. AI that understands moral values could improve those systems, making online spaces safer and less manipulated. But there's a catch: it infers your values from what you say, which can be noisy or misleading — and this technology could also be used for more aggressive micro-targeting of ads or political messages.