Sara C. Bronin (George Washington University – Law School) has posted Keeping AI in its Zone (Yale Law & Policy Review, forthcoming) on SSRN. Here is the abstract:
With zoning-reform fever sweeping local governments, state legislatures, and even Congress, there is an urgent push to understand exactly how zoning works. Unfortunately, zoning is a particularly opaque body of law, and zoning codes are arguably the most complex legal documents in the American legal canon due to their length, inconsistencies, interpretation challenges, and regulatory scope. Given this complexity, it may be tempting to believe that we should use artificial intelligence (AI) to make sense of zoning. Indeed, some recent scholarship has asserted that AI can understand zoning codes as well as, or better than, humans. This Article aims to temper such claims. By honestly acknowledging AI’s limitations, we can redirect our energy to harnessing AI’s actual capacity to help us understand zoning, the hidden power that shapes our lives. The Article draws from the author’s experience building the National Zoning Atlas (NZA), the largest geocoded zoning-data repository ever created, to posit that zoning codes will likely continue to stubbornly defy AI interpretation—even if the models improve and the relevant training corpus expands. Part I identifies relevant regulatory attributes, including inconsistent structure and terminology, the prevalence of alternative regulations, and mismatches between text and map. These attributes confirm that zoning is a dynamic body of law that demands interpretation, contextualization, and other practices of human reasoning, and is not a static dataset waiting to be harvested. Part II goes on to provide a taxonomy of recent, problematic scholarship that uses AI to characterize zoning codes, predict results from zoning codes, and extract information from zoning codes. This scholarship’s overestimation of AI’s capabilities will only achieve one thing: helping us misunderstand the law faster. Part III expands on the argument that zoning codes challenge AI by revealing the disappointing yet foreseeable results of AI trials conducted by the NZA. Part IV suggests recalibrating our expectations about how AI can be successfully applied to zoning, offering specific characterizing, predicting, and extracting applications that avoid direct code interpretation and require only limited human intervention.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum