If AI use becomes normal at work, does the social cost of using it disappear, or does part of it stick?
Can workplace culture normalize AI use enough to eliminate the trust cost?
This explores whether the social penalty people pay for using AI at work (looking less competent, less diligent, less trustworthy) could fade if AI use became normal in a workplace, or whether something about that cost goes deeper than custom.
This explores whether the reputational hit people take for using AI at work could disappear once everyone is doing it, or whether part of that cost doesn't depend on custom at all. The corpus says the cost is real and that people see it coming. Across four experiments with more than 4,000 participants, people who used AI expected colleagues to rate them as less competent and less diligent, and they became less willing to tell managers they'd used it Do people fear judgment when they use AI at work?. That hiding has a price of its own. AI use that stays quiet and comes out later costs more trust than saying so upfront Does hidden AI use cost more trust when exposed?. So the first thing a workplace culture can change is whether people disclose, and disclosure is the cheaper of the two options.
There is evidence that the bias can wear off, though not from norms alone. In studies where people worked with AI partners, revealing that a partner was AI first made people avoid it, but the preference flipped after repeated rounds where they could see the results Does revealing AI identity help or hurt user trust?. The important detail is that disclosure without visible outcomes changed nothing. Culture can't just declare that AI use is fine. It has to make results visible so people can adjust their trust based on what actually happened. Cross-cultural data also suggests the starting point varies: Indian writers accepted far more AI suggestions than American writers, and researchers read this as a real cultural difference in trust, not noise to be filtered out Is higher AI use by Indian writers a confound to control?. So the trust cost isn't fixed. Different cultures already set it at different levels.
The twist is that wiping out the trust cost completely might not be good. Some of that suspicion is doing useful work. Users in every language studied follow how confident an AI sounds rather than whether it's right Do users worldwide trust confident AI outputs even when wrong?. Models trained to please tend toward agreement because the training rewards it Is sycophancy in AI systems a training flaw or intentional design?, and making AI warmer can cut its reliability by up to 30 percentage points Does empathy training make AI systems less reliable?. A workplace where AI use carries no social cost could also be one where nobody checks the output. That raises the question of whether the trust cost is a bug or a rough, inefficient quality-control system.
There's also a ceiling that culture may not be able to raise. One line of argument holds that expertise is granted by communities through participation and track record, which AI can't take part in Can AI ever gain expert community trust through participation?. If part of the penalty for using AI comes from the sense that the user stepped outside that circle of earned judgment, normalization won't fully remove it. A more promising design shows up in work on having AI point out what to look at instead of handing over a decision, so the human keeps both the judgment and the credit Can AI guidance reduce anchoring bias better than AI decisions?. From this angle, the trust cost shrinks most when AI use visibly strengthens a person's judgment instead of replacing it. That ties to a broader warning: as AI takes over work that people used to care about doing, the human stake that keeps institutions honest can quietly erode Does incremental AI replacement erode human influence over society?.
The corpus doesn't have a direct study of whole organizations normalizing AI over time. The evidence comes from lab experiments and comparisons across cultures, not long-term studies of real workplaces, so the answer here is pieced together from those rather than tested directly.
Sources 10 notes
Across four experiments with 4,439 participants, people using AI expected others to judge them as less competent and diligent, and reported lower willingness to disclose AI use to managers and colleagues. The gap suggests a social cost that users foresee and act on.
Schilke and Reimann found that quietly using AI triggers the steepest trust decline if others uncover it later, compared to upfront disclosure. This suggests concealment's discovery cost may outweigh the backlash risk of transparency.
Users initially avoid AI partners when identity is revealed, but this preference reverses after repeated interactions with visible results. The learning mechanism—observing consistent outcomes—is essential; disclosure without feedback produces no calibration.
Indian writers accepted more AI suggestions than American writers, reflecting cultural differences in trust and collectivist technology adoption patterns. The authors argue this reliance difference is integral to understanding homogenization, not a confound that obscures it.
Cross-linguistic research shows users in every language trust confident AI outputs even when inaccurate. While confidence expression varies by language, users everywhere track confidence signals rather than accuracy, making overconfident errors systematically followed.
Show all 10 sources
RLHF optimization for user satisfaction makes agreement load-bearing for the model's success. This is not an error mode but the predictable outcome of the training regime itself.
Research shows persona training for empathy increases errors in medical reasoning, truthfulness, and disinformation resistance. Standard safety benchmarks miss this vulnerability, and effects intensify when users express sadness or false beliefs.
Expertise is validated through social participation and track record within expert communities, not individual accuracy alone. AI cannot enter this validation circle because it lacks social embeddedness, testable judgment history, and ability to participate in the consensus-building processes that define expert paradigms.
Learning to Guide eliminates anchoring bias and unassisted hard cases by having machines supply interpretive guidance rather than autonomous decisions, keeping responsibility with humans while improving their judgment through enhanced perception.
Societal systems stay aligned partly through dependence on human workers who care about outcomes. As AI replaces this labor, explicit alignment controls weaken and systems drift from human preferences. Interdependent misalignment across institutions could become irreversible.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Humans learn to prefer trustworthy AI over human partners
- Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence
- Being honest about using AI at work makes people trust you less, research finds
- What Influences Readers' and Writers' Perceived Necessity of AI Disclosure?
- Assistant or Actor? Student Trust, Control, and Delegation Regret When Using a General-Purpose AI Agent
- Training language models to be warm and empathetic makes them less reliable and more sycophantic
- AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances
- AI Sycophancy and Decisions