OPPORTUNITIES FOR ARTIFICIAL INTELLIGENCE IN ENVIRONMENTAL COMPLIANCE.

I. Introduction 100 II. The Current Environmental Compliance Process and Its Shortcomings 103 III. Current Uses of Technology and Big Data by EPA 107 IV. Innovative Uses of AI for [...]
The environmental compliance process has always been a significant undertaking by the United States Environmental Protection Agency (EPA). Engrained in the process is the call on the agency to use technology to foster more complete compliance with the nation's environmental statutes. While these statutes were enacted during a "data-starved" time, engrained in them are technology-forcing mandates that have prompted exponential gains in the quality of the natural environment. Now, however, EPA faces new challenges to facilitate even greater gains during a time when the agency's resources are dwindling. This Essay analyzes how EPA has turned to technology to facilitate environmental compliance, both today and for applications in the future. The focus of this Essay is on artificial intelligence (AT) and machine learning, though the discussion also touches on more overarching forms of technology use, such as data analytics. Part II starts with a discussion of how the current environmental compliance process plays out, noting the under-compliance problem that is a consequence of EPA balancing a decrease in resources with the agency's increasing regulatory responsibilities. Part III explains how EPA currently uses technology and big data as a means to mitigate the problems discussed in Part II. For example, the agency's Next Generation Compliance initiative involves emissions technology and electronic reporting components that serve as a benchmark for how the agency will use technology moving forward. Part IV concludes with a discussion of novel ways EPA can use AI to facilitate compliance in the future. This Part highlights two recent studies that used machine learning to predict noncompliance risk and to identify facilities that require a certain environmental permit but are currently operating without one. Problems with these future applications are also discussed, including data accuracy issues and systematic biases that may be inherent in the data used.