Do workers become dependent on AI when they stop using it for the same task?
This explores whether AI assistance leaves workers genuinely more capable or quietly dependent — what happens to performance when the AI is taken away and they face the same task alone.
This explores whether AI assistance leaves workers genuinely more capable or quietly dependent — what happens when the AI is removed and they face the same task alone. The corpus is unusually direct here: it suggests the gains are mostly borrowed, not earned. The sharpest framing is the "exoskeleton" finding — AI lets workers produce skilled-looking output while it's present, but they revert to baseline the moment access is removed Does AI assistance build lasting skills or temporary abilities?. A controlled study makes the same point empirically: people performed far better on content tasks with AI, but when asked to do similar work independently afterward, their performance showed no improvement at all — the capability never transferred Does AI assistance help workers learn lasting skills?.
Why does the skill fail to stick? One answer is that the measured productivity gains come from applying skills workers already had, not from acquiring new ones — when AI is used to learn rather than to execute, the gains evaporate and learning itself suffers When does AI actually boost worker productivity?. So "dependence" isn't just a bad habit; it's structural. If the AI is doing the part you'd otherwise be building competence in, removing it returns you to where you started.
The more unsettling thread is that workers may not notice this happening. The "LLM Fallacy" describes a self-perception error where people fold AI-generated output into their own sense of competence, believing they hold a skill they don't actually possess — and this happens precisely because fluent, seamless output hides the human-machine boundary Do AI-assisted outputs fool users about their own skills?. It's distinct from hallucination or ordinary automation bias; it's about misattributed credit, and it persists even when the output is accurate How does AI-assisted work reshape how people see their own abilities?. A four-month EEG study puts a physiological floor under the worry: heavy LLM users showed the weakest neural engagement, poorest memory retention, and reduced ability to recall their own recent work — what the authors call accumulating "cognitive debt" Does AI assistance weaken our brain's ability to think independently?.
The interesting wrinkle the corpus offers is that dependence isn't inevitable everywhere — it tracks the *type* of task. AI assistance is reliable on structured, externally checkable work and unreliable on novel judgment, and that boundary stays stable even as specific tasks shift Where does AI assistance become unreliable in research?. There's also a hidden cost during use, not just after: AI suggestions can sever the cognitive immersion needed for hard reasoning, forcing workers to rebuild focus even when the suggestion was correct Does AI assistance always help reasoning or does it carry hidden costs?. And AI doesn't really save task time so much as reallocate it toward prompting and evaluating outputs — which quietly changes what you're practicing Does AI really save time, or just change how we spend it?.
The thing you might not have known you wanted to know: the corpus draws a hard line between *temporary capability* and *durable skill*, and almost everything here lands on the same side — AI reliably boosts the output, rarely the person. The practical implication is that whether you become dependent isn't about willpower; it's about whether the task was one where an external check could carry you, and whether you ever did the part that builds the muscle.
Sources 9 notes
Research shows AI assistance creates temporary capability extensions—workers produce skilled-looking output while AI is present but revert to baseline performance when access is removed. This differs fundamentally from true skill, which persists independently.
Wu et al. found that workers using generative AI performed substantially better on content tasks, but when performing similar tasks independently afterward, their performance showed no improvement. The capability did not transfer across contexts.
Studies showing AI productivity gains measured tasks within workers' existing domains. When workers used AI to learn new skills, productivity gains disappeared and learning suffered, suggesting prior findings do not generalize to skill acquisition.
Research identifies a systematic cognitive attribution error where individuals integrate AI-generated outputs into their capability identity, believing they possess skills they don't actually have. This occurs when task output is seamless and fluent, obscuring the human-AI boundary.
Research shows the LLM Fallacy operates through misattribution of AI outputs to personal capability, independent of output accuracy or reliance behavior. It requires interventions that clarify human-machine contribution boundaries, not just better system accuracy or forced verification.
A four-month EEG study of 54 participants found that brain connectivity systematically scaled down with AI reliance—LLM users showed weakest neural engagement, poorest memory retention, and impaired ability to recall their own recent work.
AI excels at structured, externally verifiable tasks like literature retrieval and drafting, but fails sharply on novel ideas and scientific judgment. The boundary consistently tracks whether an external oracle can verify the output—a principle that remains stable even as specific task assignments shift.
Well-intentioned AI suggestions can damage reasoning performance by severing cognitive immersion, forcing users to rebuild focus before continuing. Evaluation must measure flow preservation across entire tasks, not just local suggestion accuracy.
Research shows AI doesn't reduce total task time; it reallocates it away from active work toward composing prompts and understanding outputs. This shift changes the cognitive demands and learning outcomes, making time-on-task a poor productivity metric.