Gen Z was originally promised to become a great "equalizer" – a tool that would enable 24-year-olds to work as effectively as seasoned professionals. However, the latest research from David Autor suggests that the outcome might be just the opposite.
In a three-month experiment targeting patent lawyers, AI improved the quality of drafts in all groups. However, when the researchers removed the tool and tested the lawyers' independent judgment abilities, only the more experienced lawyers showed an improvement in average performance; junior lawyers did not.
In a written response to Fortune, Autor stated: "Our findings challenge the notion that AI will automatically act as a skill equalizer. Our data indicates that it is more like a performance equalizer, but also a skill de-equalizer, because only those practitioners who already possess a foundational mental model are able to truly enhance their underlying skills."
This study categorizes lawyers based on experience rather than age. "Senior" refers to those with seven years or more of practice, and the study does not consider Generation Z as a single group. However, it is still the early-career professionals who are most directly affected by these findings.
From "China Shock" to AI
Autor is the head of the Department of Economics at the Massachusetts Institute of Technology and is also associated with Google Technology, and, and Society Visiting Fellow. His most renowned research is the "China Shock" theory co-proposed with economists David Dorn and Gordon Hanson. Their study indicates that China's import competition has caused lasting harm to the U.S. manufacturing community, which did not adjust as smoothly as the optimistic narratives of globalization predicted. In the labor markets most affected by this shock, wages and labor participation rates remained low for at least a decade, and unemployment rates remained high throughout that period.
Recently, Autor and Hanson have once again warned that there may be a "China shock 2.0": China's competition in advanced industries threatens the United States' technological leadership, as well as the high-paying jobs associated with it. Some well-known economists, including Apollo Global Management's Torsten Slok, warn that the impact of the "AI shock" on the U.S. labor force could be similar to that of China's impact; Bhaskar Chakravorti from Tufts University even coined the term "wired belt" as an analogy for the "rust belt (Rust Belt)."
Autor refuted this claim in the podcast Possible hosted by Reid Hoffman, co-founder of LinkedIn. He stated that AI "will not repeat the Chinese trade shock in any sense," as the latter was a "purely negative competitive shock" for American businesses, while AI will exhibit "very different characteristics" in terms of improving productivity.
The debate surrounding AI is based on an assumption that has been overturned by China's advancements: once technology takes over routine tasks, workers will naturally move on to higher-level skills. In a response to Fortune, Autor stated that his concern regarding AI is not about mass unemployment. “The danger we face is not that it will completely eliminate a large number of jobs, but rather that it will eliminate certain professional fields while creating new ones,” he said. “That might sound like a reasonable trade-off, but the workers who lose their old jobs are usually not those who can take advantage of the new ones.”
The patent lawyer's experiment focused on a more narrow issue: whether workers using powerful new tools are truly developing the professional skills necessary for their future promotions.
Better drafts, unbalanced learning
Autor and six other co-authors published this research as a working paper of the National Bureau of Economic Research ( NBER ) in the United States, which has not yet undergone peer review. According to the paper, the study included 133 lawyers from 11 American intellectual property law firms that have ongoing patent drafting collaborations with Google. The researchers randomly assigned access to a customized AI patent drafting assistant, which was a tool that had not yet been released at that time. Two-thirds of the lawyers were granted access, while the remaining lawyers were not allowed to use it until the end of the study.
The role of Google is not limited to providing tools. The six co-authors listed in the paper are all employees of Google. The paper also states that Google carried out this research and covered the direct costs, while the Human Research Committee at the Massachusetts Institute of Technology determined that MIT was not involved in the research. Google stated in an email to Fortune that this paper was an independent study conducted by Autor during their tenure as a visiting scholar.
After 10 days and 90 days, lawyers drafted patents based on the simulated invention materials. According to the paper, a patent lawyer from an independent law firm scored these works without knowing which lawyers had used AI, with scoring criteria including feasibility, accuracy, strategic ambiguity, completeness, and clarity.
The paper found that after 90 days, lawyers with the AI permission improved their drafting scores by 0.38 standard deviations. Autor and co-author Tanya Rodchenko described this result in a accompanying Google blog post as an increase of 11 percentage points relative to the control group. The paper states that this improvement came from fewer low-quality works, rather than more excellent works.
The more difficult tests come at the end. Lawyers must make modifications to a hypothetical patent; according to a blog post, this patent contains many substantive and stylistic errors. The paper states that this is a task that patent lawyers “usually complete without assistance.” The use of AI is not allowed here.
According to the paper, senior lawyers with AI privileges performed 0.45 standard deviations better in this test than their counterparts in the control group. Junior lawyers did not show any improvement on average.
The paper also lists some limitations. Only 91 out of 133 lawyers completed this test without the use of AI. Initially, law firms refused to participate in the skills test; therefore, researchers inferred skill improvements through random assignment rather than by measuring before and after. The authors also identified 15 of the 91 revised drafts as possibly having received assistance from AI, despite the rules prohibiting the use of AI. Moreover, these instances occurred with similar frequency among lawyers in the control group. If these 15 drafts were excluded, the paper states that the advantage of senior lawyers narrowed to 0.39 standard deviations, and this pattern still held true, although the statistical significance threshold became more lenient. However, the overall effect was not significant.
"The Illusion of Competence"
The paper found that there was a differentiation in the score distribution among junior lawyers: significantly more low scores, fewer medium scores, and more high scores, but the highest scores did not increase. The authors wrote that AI “served as a springboard for some junior lawyers and acted as a buffer for others.”
The blog post describes the working methods of junior lawyers. They tend to approach their work from top to bottom, first polishing the introductory text and then addressing the core arguments. Some people identified serious flaws but merely left comments to point them out without making any actual corrections. Junior lawyers in the control group also followed the same approach, which the author referred to as “basic flaws at the junior level.” Moreover, three months of using AI did not fix this issue.
However, junior lawyers have indeed noticed a difference: they like this tool. In a response to Fortune, Autor stated that the satisfaction rate of junior lawyers using AI in reporting tasks has increased significantly, which is the most notable effect of the study. He said that they find it convenient to skip the “blank page problem” and be able to directly assume the role of reviewer.
"Young professionals need to be wary of the illusion of competence," he said. "The only way you can truly know whether you are improving your skills is by completing tasks without the assistance of AI and evaluating your own performance."
Why Senior Lawyers Benefit
According to the blog post, experienced lawyers described AI in subsequent interviews as a kind of “logical auditor,” rather than the final product. The author wrote that AI weakened their reliance on the existing text and forced them to explain more clearly the reasons and methods behind the structural modifications.
Autor told Fortune, "It can be said that only when you have a sufficient understanding of the law and are able to identify issues with legal texts, can you use AI as a 'logical auditor.'" He believes that outsourcing the basic drafting work allows senior lawyers to focus their efforts on strategic aspects, which is like "seeing the forest rather than just the trees."
In an unassisted test, a blog post stated that senior lawyers who relied on the assistance of AI skipped low-risk textual embellishments, reconstructed their arguments from scratch, removed wording that could potentially narrow the scope of legal protection, and linked many of these modifications to legal principles.
This does not mean that junior lawyers should abandon tools. Autor told Fortune that beginners should work in the same way as more experienced professionals: first construct their arguments without any assistance, and then use AI as a critic to stress-test their logic. He acknowledges that this requires a high level of discipline. “Most of us do not have enough self-discipline to complete tasks manually when we have readily available, low-cost tools that are willing to do the work for us,” he said.
Don't abolish the apprenticeship system.
For employers, better output will incentivize them to hire fewer junior employees. Autor considers this approach short-sighted. “If companies automate formative exercises without replacing them with some kind of guided learning environment, they are cutting off the apprenticeship pipeline for cultivating future senior partners,” he told Fortune.
He suggests that companies decide which tasks can be permanently assisted by AI, and which judgmental tasks require proficient mastery without the use of tools. Companies can also combine the promotion of AI with regular unassisted skill checks, such as offline red-line modification exercises, and have partners responsible for guiding junior employees through technical mistakes. These are just ideas; they are not yet proven effective solutions. “In this field of research and practice, we are still in the early stages,” says Autor, and few solutions have been fully verified.
The paper also lists other limitations. The sample size is small, and it comes from law firms that have been continuously engaged in patent business with a single mature client; three months is also a short period compared to the many years required to build professional patent capabilities. Blog articles add that if this experiment were conducted today, more advanced models and a broader level of familiarity with AI might change the results. It does not prove that AI would prevent junior lawyers from becoming experts throughout their careers.
Autor reconnects this issue to his research on the impact on China. He told Fortune that most technological advantages come not so much from devices or blueprints, but rather from human knowledge, which is built up through a slow and laborious process of mastery. “If we think that AI will free us from the burden of mastering specialized skills, I think we will be greatly disappointed,” he said. “Human intelligence and machine intelligence will be complementary in the long run.”
