What happened
David Autor, head of MIT’s economics department, and six Google co-authors published a National Bureau of Economic Research (NBER) working paper detailing a three-month experiment with 133 patent lawyers at 11 U.S. firms. The study found that while AI access improved drafting scores for all participants, only senior lawyers (7+ years experience) showed improved independent judgment when tested without AI. Junior lawyers showed no average skill gain, challenging the notion that AI is an automatic skill equalizer.
David Autor, known for his 'China shock' research, led a study with six Google co-authors published as an NBER working paper. The experiment involved 133 patent lawyers at 11 U.S. intellectual-property law firms with ongoing relationships with Google. Researchers randomly assigned two-thirds of the lawyers access to a custom, then-unreleased Google Labs AI patent-drafting assistant for three months, while the remaining third served as a control group without access.
The study measured two outcomes: drafting quality and independent judgment. After 90 days, lawyers with AI access produced drafts that scored 0.38 standard deviations higher than the control group, described by the authors as an 11-percentile-point gain. However, the improvement was driven by a reduction in weak work rather than an increase in excellent work. The critical test came when all lawyers were required to mark up a hypothetical patent containing errors without AI assistance.
In this unassisted test, senior lawyers (defined as having seven or more years of practice) who had used AI outperformed their control-group peers by 0.45 standard deviations. In contrast, junior lawyers showed no average gain in independent judgment. The paper notes that junior scores split, with significantly more poor scores and fewer mediocre ones, but no gain at the top. The authors describe AI as serving as a 'springboard for some juniors and a cushion for others,' but failing to fix a 'baseline junior deficit' in independent problem-solving.
The study includes several caveats. It has not been peer-reviewed. Only 91 of the 133 lawyers completed the final AI-free test. The firms declined a pre-study skills test, so researchers inferred skill gains from random assignment rather than direct before-and-after measurement. Additionally, 15 of the 91 final markups were flagged as possibly AI-assisted despite the ban; excluding these, the senior advantage shrank to 0.39 standard deviations but remained statistically significant at a looser threshold, while the overall effect did not hold up as strongly.
Autor and co-authors argue that AI functions as a 'performance equalizer' but a 'skill-disequalizer.' Experienced lawyers used AI as a 'logic auditor' to pressure-test their work and focus on strategic scope, while juniors often relied on the tool to bypass the 'blank page problem,' leading to higher task satisfaction but less development of foundational mental models. Autor warns that if firms automate away formative practice without replacing it with guided learning, they risk severing the apprenticeship that produces future senior experts.
Why it matters
This research provides concrete evidence that AI tools may create a 'skill-disequalizer' effect, where experienced workers leverage AI to enhance strategic thinking while novices risk developing an 'illusion of competence' without building foundational expertise. This has significant implications for workforce development, apprenticeship pipelines, and how organizations structure training in AI-augmented environments. It suggests that without deliberate unassisted practice, AI could widen the gap between junior and senior professionals rather than closing it.
The findings challenge the optimistic narrative that AI will naturally upskill workers by handling routine tasks, allowing them to focus on higher-level skills. Instead, the data suggests that without existing foundational expertise, AI may hinder skill acquisition by providing a 'cushion' that prevents novices from struggling through the necessary learning process. This is particularly concerning for early-career professionals in knowledge-intensive fields like law, where expertise is built through years of deliberate practice.
For employers, the study highlights a tension between short-term productivity gains and long-term talent development. While AI improves immediate output, it may create a dependency that undermines the development of independent judgment. Autor suggests that firms need to deliberately design workflows that include unassisted practice and guided learning to ensure that junior employees continue to build the mental models necessary for senior roles.
The research also has broader implications for the labor market. Autor, who has previously warned about the 'China shock' and its impact on manufacturing jobs, argues that the 'AI shock' will have a different texture, focusing on the transformation of skills rather than mass job loss. However, the risk is that AI may eliminate areas of specialty while creating new ones, with the workers losing the old careers not being the same as those who can take advantage of the new roles. This could exacerbate inequality within professions, not just between them.
The study's focus on patent law is specific, but the mechanisms identified—reliance on AI for routine tasks, lack of independent practice, and the resulting skill gap—are likely applicable to other fields where expertise is built through iterative, feedback-driven learning. As AI tools become more integrated into professional workflows, understanding and mitigating these effects will be crucial for maintaining a skilled workforce.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
What should a useful AI forecast state?
What to watch next
Monitor how law firms and other professional services industries adjust training protocols in response to these findings. Watch for further research on long-term skill development in AI-augmented roles and whether companies implement 'unassisted skill checks' or guided learning environments to mitigate the risk of severing apprenticeship pipelines. Also observe if similar patterns emerge in other knowledge-work fields beyond patent law.
Look for how law firms and other professional services organizations respond to these findings. Will they implement policies that require unassisted practice for junior employees? Will they invest in guided learning environments that help novices use AI as a critic rather than a crutch? The adoption of such practices will be a key indicator of how seriously the industry takes the risk of skill atrophy.
Monitor further research on the long-term effects of AI on skill development. The current study is limited to a three-month period and a specific profession. Future studies will need to examine whether the skill gap persists over longer timeframes and whether it varies across different types of AI tools and professional contexts.
Watch for policy discussions around AI and workforce development. As the evidence mounts that AI may not be an automatic skill equalizer, policymakers may need to consider how to support workers in adapting to AI-augmented roles. This could include funding for training programs, incentives for companies that invest in skill development, or regulations that require transparency about how AI is used in professional settings.
Observe whether similar patterns emerge in other knowledge-work fields. The study's findings are specific to patent law, but the underlying dynamics—reliance on AI for routine tasks, lack of independent practice, and the resulting skill gap—are likely to be relevant in other areas such as software development, finance, and healthcare. Research in these fields will help determine whether the 'skill-disequalizer' effect is a general phenomenon or specific to certain types of work.