Google 资助研究：AI 让初级专利律师起草更快，但独立审查能力未见提升

The study separated two things by having attorneys first work with AI assistance and then evaluate results without it. This design let researchers measure both the quality of AI-assisted work and how well the attorneys' own judgment held up afterward. Blinded patent attorneys at an independent firm scored all results across five categories: enforceability, technical accuracy, strategic ambiguity, completeness, and clarity.

Two-thirds of participants in the Google-funded study got access to "InFlow," an unreleased AI assistant from Google Labs, while the remaining third worked without any AI. In two patent claim drafting tasks, only the treatment group could use InFlow, and the control group completed the same tasks unaided.

In a third task called redlining, all participants had to critically review an existing patent draft without AI. As the authors explain, drafting measures immediate AI-assisted performance, while redlining measures internalized judgment, a task patent attorneys routinely perform without any tools.

AI improves drafts mainly for junior attorneys

In the drafting tasks, the treatment group clearly outperformed the control group at both the 10-day and 90-day marks, with effects showing up across all five scoring dimensions. Higher scores came mostly from fewer bad ratings and more mid-range ones, while top scores barely moved.

Gains were much larger for juniors with fewer than seven years of experience than for seniors. According to the researchers, this aligns with a growing body of studies showing that AI narrows performance gaps between less and more experienced workers, an effect already observed in writing, customer service, and software development.

Only seniors keep AI advantages

Redlining produced the opposite result. The treatment group scored better overall, but the advantage came entirely from senior attorneys, and juniors saw no benefit at all, even though they had posted the biggest gains during AI-assisted drafting.

Instead, junior results split apart, with far fewer mid-range scores replaced by more bad and more good ones, and no growth at the top. The authors read this as a sign that AI sped up skill-building in some juniors while replacing it in others, though the experiment doesn't explain why the effects diverged.

The attorneys who gained the most from AI were the same ones who retained the least lasting skill, which leads the researchers to suggest that baseline expertise may be a prerequisite for turning AI-assisted practice into durable ability.

A supplementary text analysis offers one possible explanation. Senior attorneys more often restructured the flawed draft and grounded their changes in legal principles, while juniors frequently spotted serious flaws without fixing them, leaving notes that read like instructions to an AI assistant. Because this behavior appeared in the control group too, the authors don't attribute it to an AI-trained habit and instead see it as a baseline deficit at the junior level that AI use didn't fix.

What the results mean in practice

The authors point to several limitations, including the small sample size driven by top attorneys' high billable rates, the fact that participating firms aren't representative of professional services as a whole, the lack of a baseline skills measurement, and the three-month window being short compared to the years it takes to build patent expertise.

The tension between immediate productivity gains and lasting skill development matters as long as professional work still requires judgment without AI, the researchers argue, and it likely will for a long time. Even if AI eventually supports redlining, it probably won't be available to attorneys in real time during client meetings and courtroom proceedings, where stakes are high and judgment calls happen on the spot.

Similar patterns show up in other fields. In software development, an Anthropic study found that younger developers who learned a new programming library with AI help scored 17 percent worse on a follow-up knowledge test than those who learned without it. In essay writing, the MIT Media Lab found signs of cognitive debt from outsourced thinking. And in management consulting, a BCG study showed back in 2023 that weaker consultants improved far more with GPT-4 than strong performers did.

Stanford researchers explain the decline in entry-level jobs across multiple industries by arguing that AI can mostly replace textbook knowledge from school and college, which is the main thing early-career workers bring to the table, while experienced professionals also carry hands-on expertise that AI can't easily substitute.

The patent attorney study feeds into a broader debate about the erosion of professional expertise, what some researchers call the "tragedy of the cognitive commons." Individual firms benefit from automation while the cost of stunted talent development spreads across an entire profession, and this study adds more experimental evidence from real-world practice to back up that concern.

Anthropic recently said that junior positions in its engineering departments dropped from 21 percent to 14 percent since early 2025.