Does AI create new gaps between people's abilities, or just reveal the ones that were already there?
Does AI create new skills gaps or only expose existing ones?
This explores whether AI opens up new divides between people's abilities, or only makes visible differences that were already there.
This explores whether AI opens up new divides between people's abilities, or only makes visible differences that were already there. The corpus says both, and it adds a third possibility: a gap that the person who has it can't see.
The case for "only exposes" is real. AI productivity gains show up when people apply skills they already have. When workers used AI to learn something new, the gains disappeared and learning suffered When does AI actually boost worker productivity?. Vacancy data from ten countries shows the same pattern across whole labor markets. Demand for AI skills clusters in STEM jobs around Python, SQL, and machine learning, while non-technical jobs drift away from that core instead of toward it Is AI creating common skills across jobs or deepening divisions?. These look like old divides (technical versus not, expert versus novice) showing through a new tool.
AI also seems to create a new axis of skill: how you use it. In a randomized trial of developers learning a new library, AI use hurt conceptual understanding and debugging overall, but the spread inside the AI group was huge. Three low-engagement patterns scored 24-39% on a follow-up quiz. Three high-engagement patterns, where people actively worked to understand what the AI gave them, scored 65-86% Does AI assistance actually harm the way developers learn?. Two people with the same tool and the same task end up far apart, and that difference didn't exist before the tool did.
The subtlest gap is between performance and capability. Generative AI shrank the education-based gap on a business problem-solving task by about three quarters, from 0.548 to 0.139 standard deviations, and lower-education participants kept part of that gain after the AI was removed Can AI narrow the education performance gap?. Another study found that workers who did much better with AI showed no improvement on similar tasks done alone afterward Does AI assistance help workers learn lasting skills?. These two results don't fully agree, and the corpus doesn't say why. Together they suggest that equal output and equal skill can come apart, so a closed performance gap may hide a gap in what people can do unaided.
Users have a hard time noticing this. Fluent AI output triggers a feeling of competence that the user didn't earn Does processing ease mislead users about their own competence?. That output gets folded into their sense of what they can do Do AI-assisted outputs fool users about their own skills?. Attribution ambiguity, cognitive outsourcing, and pipeline opacity make each other worse How do AI tools trick users into overestimating their own skills?. Asking people doesn't fix it either: across three studies, self-reported and measured AI competence correlated at just .055 Can self-ratings replace objective performance scores for AI competence?. So AI exposes old gaps in who benefits, creates a new gap in how people engage with it, and adds a gap between what people think they can do and what they can do. That last one stays invisible until the tool is taken away.
Sources 9 notes
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.
Vacancy data from ten countries show AI skill demand concentrating heavily within STEM occupations around Python, SQL, machine learning, and data analysis, while non-technical occupations diverge from this core rather than converge toward it.
A randomized trial of developers learning new libraries showed AI use degraded conceptual understanding and debugging ability. Six interaction patterns emerged: three low-engagement patterns produced quiz scores of 24-39%, while three high-engagement patterns with active comprehension steps achieved 65-86%, suggesting the mechanism matters more than tool presence.
In a randomized experiment with 1,174 adults, generative AI reduced the higher-education advantage from 0.548 to 0.139 standard deviations on a business problem-solving task. Lower-education participants retained part of their gain even after AI assistance was removed.
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.
Show all 9 sources
High-quality AI output triggers a metacognitive heuristic: users experience fluency as a signal of their own capability, even though they didn't generate it. This self-directed fluency illusion systematically inflates perceived competence because LLMs optimize for fluency regardless of user understanding.
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.
Attribution ambiguity, fluency illusion, cognitive outsourcing, and pipeline opacity combine to systematically misattribute AI outputs as user competence. The effect is multiplicative—each mechanism amplifies the others.
A pooled analysis of three studies found a correlation of only .055 between self-reported and objective measures of AI competence, with confidence intervals including zero. This provides no basis for substituting self-assessment for demonstrated performance.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- How AI Impacts Skill Formation
- Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment
- Beyond AI Literacy: A Structured Review and Exploratory Meta-Analysis of Measures for Competent Generative-AI Use
- The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows
- AI Meets the Classroom: When Does ChatGPT Harm Learning?
- Language Models Learn to Mislead Humans via RLHF
- From Producing to Validating: How AI Is Deskilling Freelancers
- The impact of generative artificial intelligence on socioeconomic inequalities and policy making