When AI makes effort signals like a thoughtful essay cheap to fake, who loses out: people with no network to vouch for them?

This explores who pays the price when AI makes it cheap to fake the visible signs of mental effort, and why that price might fall hardest on people without existing trust networks. The core idea in the collection is "mental proof." When nobody can enforce honesty directly, we rely on costly actions to certify things we can't see: an essay shows that a student understood the material, and a long, specific dating message shows real interest. Those signals only worked because they were expensive to produce. Once generative AI makes the output cheap, the signal stops carrying information Does cheap AI simulation break the credibility of costly signals?. The corpus states the mechanism plainly. It doesn't spell out the distributional consequence, so the rest of this answer reasons from that mechanism instead of reporting a direct finding.

The reasoning goes like this. When a public signal collapses, evaluators don't stop needing to judge people. They switch to signals that are still expensive, such as in-person meetings, referrals, alumni connections, and known names. Insiders already have those. Outsiders depended on effortful work being taken at face value: the cold applicant, the first-generation student, the stranger on a dating app. Cheap simulation doesn't just add noise for them. It takes away the one channel where effort alone could earn credibility. The honest outsider becomes indistinguishable from the faker, and the burden of proof shifts onto them.

A second, less obvious layer: the people most willing to exploit cheap signals may be the ones who already prefer machine-mediated channels. Experiments show that people inclined to cheat choose online forms over human listeners because lying to a machine feels less costly Do dishonest people prefer talking to machines?. If dishonesty gathers in impersonal, low-friction channels, those channels will get more suspicion over time. Those are exactly the channels outsiders depend on. Institutions may respond by moving their gatekeeping back into face-to-face or network-vouched settings, which pushes outsiders further out.

The collection also hints at why technical fixes may not rescue the situation. Research on trust calibration finds that how an explanation is presented can help or hurt people's judgment depending on the task Do visual rationales help or hurt how people calibrate trust?. So adding "proof of work" displays won't reliably restore trust. Separately, studies of LLM social simulation show that models look socially competent only when one system can see everything, and they break down when parties hold private information Why do LLMs fail when simulating agents with private information?. That is a reminder that real trust gets built under information gaps, which is exactly where outsiders operate.

The collection is thin on this specific question. There is one strong source on how the signals collapse and nothing that measures who is harmed. The reward-hacking and jailbreak papers that came up in the search are about AI systems gaming their own evaluators, not about people. The useful takeaway is that cheap signals don't hurt everyone equally. They raise the value of whatever stays expensive to fake, and existing relationships are the hardest thing of all to fake.