This article was published in Lawfare’s research paper series. The canonical version for citation is available here.
This article was co-authored by Justin Curl. He is a third-year student at Harvard Law School. Previously, he was a Schwarzman Scholar at Tsinghua University and earned a degree in computer science from Princeton University, where we first worked together. You can find more of his writing on AI and law here.
Many AI leaders believe the technology will reshape knowledge work. OpenAI CEO Sam Altman has predicted that by 2030 there will be AI systems “smarter than humans,” 1 while Anthropic CEO Dario Amodei has compared future AI models to “a country of geniuses in a data center.” 2
Researchers believe legal services are especially vulnerable to AI disruption. 3 And since GPT-4 passed the bar exam, 4 many in the industry seem to agree. Law schools have already begun incorporating AI into their curricula, 5 and partnering with AI-focused legal tech companies to help future lawyers adapt to a changing profession. 6 One prominent lawyer has even argued that AI can already replace clerks and oral argument. 7 Another predicted that by 2035, AI may “replace the traditional lawyer.” 8
This enthusiasm for AI comes against the backdrop of an already expensive legal-services market. Millions of individuals are priced out of legal help, 9 and corporate legal spending is steadily rising, with hourly rates for partners at large firms now exceeding $2,300. 10 Unsurprisingly, many observers believe AI has the potential to make legal services more accessible by delivering outcomes at lower cost. 11
Our central argument is that advanced AI will not automatically help consumers achieve the legal outcomes they want at lower cost. We examine the bottlenecks standing between improvements in AI capability and the kind of positive transformation in legal practice that some have envisioned. 12 For AI to usher in a world of abundance in legal services, the industry would have to overcome three bottlenecks: regulatory barriers, adversarial dynamics, and human involvement.
First, regulation around the unauthorized practice of law (UPL) may limit consumers’ and, to some extent, lawyers’ use of AI. These laws prohibit non-lawyers from practicing law. 13 If courts determine that certain systems have crossed the line into practicing law, individuals and organizations could face steep fines and criminal liability, 14 forcing prospective service providers either to constrain the legal functionality of their AI tools or to risk enforcement. Entity-based regulation — which limits the types of entities that can own stakes in businesses providing legal services — would constrain how legal services are delivered, and in turn limit the ways lawyers and consumers can use AI. Without reform, AI will not help people achieve their legal goals if consumers cannot access its capabilities or if lawyers have no incentive to use it effectively, no matter how advanced it becomes.
Second, even if AI is adopted effectively and widely, the adversarial structure of the U.S. 15 legal system may prevent advanced AI from reducing the cost for clients to achieve outcomes. 16 Because legal outcomes often depend on relative quality rather than absolute quality, when both sides become more efficient, the equilibrium simply shifts upward overall. 17 In a world with advanced AI, achieving the same outcome — such as a favorable settlement or a win at trial — will require more and better legal work. So even if productivity rises and the cost of individual legal tasks falls, both sides may still be locked in an arms race that increases the amount of legal work required to reach the same result.
As a historical analogy, digitization could have lowered discovery costs by making document review easier. 18 But litigators operating within an adversarial framework used the proliferation of digital documents to shift costs onto their opponents, and total litigation costs remained high. 19
Although transactional work, such as contract negotiation, is not as obviously adversarial, it may exhibit similar dynamics: Lawyers compete to control the scope of disclosure and to suppress the other side’s counsel when drafting and negotiating agreements. 20 Of course, some legal outcomes — such as effective estate planning — do not depend on adversarial processes, and this bottleneck would not apply to them.
The third and final bottleneck we discuss is human involvement in legal work. In a world where AI capabilities improve faster than the total volume of legal work grows, our desire to have humans decide cases and understand the contracts they sign creates a final constraint. In litigation, if AI causes legal work to surge, judges will likely respond by taking longer to resolve disputes (delaying outcomes) or by delegating more work to clerks (reducing adjudicatory quality). 21 And in transactional work, even if AI can draft an entire contract, human lawyers will still need time to understand what those terms mean for the organization’s interests. human