Yonathan A. Arbel & Shmuel I. Becher
90 Geo. Wash. L. Rev. 83
What does it mean to have machines that can read, explain, and evaluate contracts? Recent advances in machine learning have led to a fundamental breakthrough in machine language models, portending a profound shift in the ability of machines to process text. Such a shift has far-reaching consequences for diverse areas of law, which are predicated on, and justified by, the existence of information barriers. Our object here is to provide a general framework for evaluating the legal and policy implications of employing language models as “smart readers”—tools that read, analyze, and assess contracts, disclosures, and privacy policies.
Synthesizing state-of-the-art developments, we identify four core capabilities of smart readers. Based on real-world examples produced by new machine-learning models, we demonstrate that smart readers can: simplify complex legal language; personalize the contractual presentation to the user’s specific sociocultural identity; interpret the meaning of contractual terms; and benchmark and rank contracts based on their quality.
Nevertheless, the implications of smart readers are more complex than initially meets the eye. Although smart readers can overcome traditional information barriers and empower parties, they rely on black-box models that sophisticated parties can exploit. Smart readers can close some of the gaps in access to justice, but they also introduce concerns about contractual bias and discrimination. And even though smart readers can improve term transparency, they might lead judges and policymakers to relax their guard prematurely.
The current body of doctrine and scholarship is ill equipped to address both the prospects and risks of smart reader technology. This Article narrows this gap. It maps the necessary theoretical, policy, and doctrinal adaptations to the age when machines can automate the reading of contracts.