As artificial intelligence becomes woven into the daily decisions of knowledge workers, a study of 452 professionals reveals that trust in these systems is not a simple good — it is a force that simultaneously opens creative possibility and tightens the grip of dependency. Published in Nature, the research finds that the same confidence that empowers workers to innovate also binds them to systems they cannot fully comprehend, generating a measurable strain alongside the gains. The findings suggest that wisdom, not enthusiasm, must guide how organizations introduce AI into human work — because
Study finds AI trust cuts both ways: boosting innovation while raising workplace stress
Moderate-to-high trust may deliver the largest net benefit
Mark
So workers who trust AI systems become more innovative—that sounds like a win. What's the stress part about?
Mimi
The trust itself creates a dependency. When you rely on a system to help you decide, you start feeling like you can't function without it. That anxiety—that sense of being tethered to something you don't fully understand—that's the technostress.
Mark
But couldn't companies just train people better on the AI so they understand it and the stress goes away?
Mimi
That's part of it, yes. Workers with higher AI literacy—people who actually understood how the systems worked—were the ones who got the innovation benefits from trust. But literacy alone doesn't eliminate the stress. Organizational support does that. Real backing from the company.
Mark
What does organizational support look like in practice?
Mimi
The study doesn't specify, but you can infer it: clear policies, training programs, resources, maybe time set aside to learn. Basically, the company saying "we're in this with you." That buffered the stress effects no matter what.
Mark
So there's an optimal level of trust? You don't want workers trusting the AI too much?
Mimi
That's what the data hints at. Moderate-to-high trust seems to be the zone where you get the innovation without the worst of the stress. Maximum trust might actually be counterproductive.
Mark
And if a company has low AI literacy and no support infrastructure?
Mimi
Then you've got workers who trust systems they don't understand, with no institutional backing. That's the worst case—all the stress, none of the innovation benefits.
Le Pouls
- Workers who trust AI decision tools become more innovative, but the same trust quietly increases their stress through a growing sense of dependency they did not anticipate.
- The dual effect is not theoretical — statistical analysis of 452 knowledge workers found a 0.249 indirect boost to innovation and a 0.175 indirect rise in technostress, both flowing from the same source: trust.
- A critical threshold emerges: workers with AI literacy scores above 2.81 actually capture the innovation benefits of trust, while those below it see confidence in AI fail to translate into creative output.
- Organizational support acts as a stress buffer across all levels, suggesting companies that invest in training and clear AI policies protect workers from the dependency trap regardless of how much trust employees place in the systems.
- Researchers warn that maximum trust may not be optimal — a moderate-to-high calibration appears to deliver the largest net benefit, enough to empower without tipping into burnout.
As artificial intelligence becomes woven into the daily decisions of knowledge workers, a study of 452 professionals reveals that trust in these systems is not a simple good — it is a force that simultaneously opens creative possibility and tightens the grip of dependency. Published in Nature, the research finds that the same confidence that empowers workers to innovate also binds them to systems they cannot fully comprehend, generating a measurable strain alongside the gains. The findings suggest that wisdom, not enthusiasm, must guide how organizations introduce AI into human work — because the difference between flourishing and burnout may lie in how carefully trust is cultivated.
A study of 452 knowledge workers has uncovered a paradox at the heart of AI-assisted work: trusting the systems designed to help you makes you both more innovative and more stressed, simultaneously, through two distinct psychological pathways.
The first pathway runs through empowerment. When workers trusted their AI decision support tools, they felt more capable of taking risks and experimenting — and that confidence translated into measurably higher rates of innovative behavior, with a 0.249 indirect statistical effect linking trust to innovation through empowerment. The second pathway runs through dependency. That same trust led workers to lean heavily on AI guidance, and the reliance bred technostress — the anxiety of working alongside technology one cannot fully understand or control — with a 0.175 indirect effect connecting trust to strain.
Neither pathway operates alone. AI literacy proved decisive for the innovation benefits: workers who scored above a threshold of 2.81 on the study's literacy measure saw trust reliably translate into creative output, while those below it did not. Organizational support, meanwhile, buffered the stress effects across all levels observed — companies that provided real training, resources, and policy clarity saw less technostress among their workers regardless of how much those workers trusted the systems.
The researchers employed a Neural-Augmented Bayesian statistical method that reduced estimation error by up to 51 percent compared to standard approaches, and their analysis pointed toward a counterintuitive conclusion: maximum trust may not be the goal. A moderate-to-high level of confidence appears to deliver the best net outcome — enough to drive empowerment and innovation, but not so much that dependency overwhelms the gains.
Because the study tracked workers across two time points rather than experimentally manipulating trust, the findings describe associations rather than confirmed causes. Even so, the pattern raises urgent practical questions for organizations deploying AI: What trust level is optimal? How much literacy is enough? And what does meaningful organizational support actually look like? The research suggests these answers will vary — and that getting them wrong carries real human costs.
When workers trust the AI systems their companies deploy to help them make decisions, something unexpected happens: they become more creative and innovative, but also more stressed. A study of 452 knowledge workers who use AI-driven decision support systems in their daily jobs found this paradox running through their experience in parallel tracks, each strengthened by the same foundation of trust.
The research, drawing on theories of how people appraise threats, exchange value in relationships, and respond to new technology, identified two distinct pathways. In one direction, trust in these systems led workers to feel more empowered—more capable of taking risks and trying new approaches. That sense of empowerment, in turn, correlated with higher rates of innovative behavior. The statistical weight of this effect was substantial: a 0.249 indirect association between trust and innovation flowing through empowerment.
But trust also opened a second door. When workers relied heavily on AI systems to guide their decisions, they began to feel dependent on them. That dependency bred what researchers call technostress—the strain and anxiety that comes from working alongside technology you cannot fully control or understand. This pathway was measurable too, with a 0.175 indirect effect linking trust to stress through perceived dependency.
The catch is that neither pathway operates in isolation. Two factors shaped how strongly each one played out. The first was AI literacy—how well workers understood the systems they were using. When workers scored above 2.81 on the literacy measure used in this study, the innovation benefits of trust became statistically visible. Below that threshold, trust did not reliably translate into creative output. Organizational support—the degree to which companies invested in training, resources, and clear policies around AI use—buffered the stress effects across the entire range of support levels observed. Companies that backed their workers with real support infrastructure saw less technostress, regardless of how much trust their employees placed in the systems.
The researchers used a novel statistical approach called Neural-Augmented Bayesian SEM to analyze the survey data, a method that reduced estimation error by up to 51 percent compared to standard techniques. Their exploratory analysis suggested something intriguing: there appears to be a sweet spot. Moderate-to-high trust, rather than maximum trust, may deliver the largest net benefit—enough confidence to drive innovation and empowerment, but not so much reliance that dependency and stress overwhelm the gains.
Because this study tracked workers at two points in time rather than randomly assigning them to different trust conditions, the findings describe associations rather than proven causes. The researchers are careful about that distinction. Still, the pattern they found points toward practical questions companies will need to answer: How much should workers trust these systems? What level of AI literacy matters most? And how much organizational support is enough to keep the benefits of trust from tipping into burnout? The answers, the research suggests, are not one-size-fits-all.
Citations marquantes
Workers with higher AI literacy were the ones who got the innovation benefits from trust, but literacy alone doesn't eliminate the stress.— Study findings
Envie de l'histoire complète ?
Lire l'original sur Nature ↗
Science & Space