Like you would say who you think is credible and human. An algorithm would evaluate trust on your behalf and it would look to the people you trust, and then who they trust, and so on and assign scores to people. Distrust, or even other observations, could percolate in a similar way.
Then on social networks, or some sort of small-web, new users would need to find other people to vouch for them to establish trust. When viewing websites or social media posts the trust score of users could be shown alongside content, and used to filter feeds / visibility. A troll or bot could rather rapidly get picked up by a network of distrust so they could be filtered out quickly.
The algorithms and details of such a thing are fuzzy to me, and I think a lot of care and thought would be needed to try to ensure it doesn't collapse under subtle flaws with time.