AuthorDenney, Robert
  1. 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 Environmental Compliance in the Future 111 V. Conclusion 113 I. INTRODUCTION

    In 2015, the German car manufacturer Volkswagen was involved in the "Dieselgate" scandal. (1) The company had used software in its diesel-powered cars that was able to recognize when the affected vehicles were being tested for nitrogen oxide emissions. (2) Using this software, the vehicles would reduce emissions during testing and then increase emissions by up to forty times once the cars resumed normal operations on the road. (3) After researchers accidentally discovered this problem, Volkswagen was found liable, and the car company has since paid over $25 billion in fines, penalties, and settlements. (4) Other vehicle manufacturers have also been found liable for similar emissions-cheating technologies, including Daimler AG, which recently settled for $1.5 billion to resolve the claims against it. (5) Dieselgate "is one of the most infamous examples of using artificial intelligence (AI) software with malicious intent," even though the AI used in the cars was relatively simple firstgeneration AI technology where programmers denned the parameters in which the AI system was able to operate. (6)

    In contrast to Dieselgate, the use of ToxCast by the United States Environmental Protection Agency (EPA) provides an example of AI being used to promote the public good. Traditionally, the toxicity of chemicals has been verified in animal testing, but "ToxCast applies machine-learning algorithms--specifically, linear discriminate analysis--to data on chemicals' interactions obtained from in vitro testing to predict their toxicities." (7) In addition to avoiding the controversy associated with testing chemicals on animals, it has been estimated that ToxCast could save the government $980,000 for every toxic chemical identified through the technology. (8) However, even though ToxCast could have monumental benefits, it is not without problems. For example, ToxCast has significant computing costs, as the program involves managing a library of thousands of chemicals. (9) The program also requires that the prospective prioritization of chemicals be tested in a way that minimizes false negatives. (10) If ToxCast incorrectly rules out a chemical as not being toxic, the consequences could be deadly.

    Both Dieselgate and ToxCast demonstrate that as AI becomes more complicated, concerns will arise over how it can be used in environmental compliance, both by regulators and by regulated entities. Moreover, as AI technologies advance, they will have the power to transform the environmental compliance sector. At a general level, environmental compliance concerns the monitoring, inspection, and enforcement involved in carrying out environmental statutes such as the Clean Air Act (11) (CAA) and Clean Water Act (12) (CWA). EPA and its delegated state-run programs largely oversee environmental compliance in the United States, and the process has been plagued by a number of deficiencies since the nation's major environmental statutes were enacted almost fifty years ago.

    One of these deficiencies involves the fact that environmental law became an established discipline in a "data-starved" time when computing power was limited, the use of geospatial data was in its infancy, and the causal chains involved in carcinogenic chemicals were largely unknown. (13) Despite this, the new environmental statutes embraced a compliance system that relies on data collection and new technologies, using words such as "maximum achievable control technology" and "best scientific... data available" to command how regulated entities will comply. (14) While this new regulatory regime was technology-forcing and prompted exponential gains in the quality of the natural environment in its first couple of decades, the regime now faces hurdles as EPA learns how to use the treasure trove of environmental compliance data it has accumulated in sophisticated ways during a time when agency resources are dwindling. (15)

    This Essay explains how EPA and its associated state-run programs have turned to technology to facilitate environmental compliance, both today and in potential applications in the future. The focus of this Essay is on AI (a catchall term for machines that can replicate human capabilities, such as problem-solving) and machine learning (a type of AI where a machine is trained to make complicated predictions in a manner more efficient than could be done by humans). (16) However, this Essay also pertains to simpler or more overarching forms of technology use, such as data analytics.

    Part II starts with a brief discussion of how the current environmental compliance process normally plays out, along with problems associated with this process. Namely, environmental compliance in the United States is plagued by a massive under-compliance problem during a time when EPA must balance a decrease in resources with the agency's increasing regulatory responsibilities. Part III explains how EPA currently uses innovative technologies and "big data" (i.e., large volumes of available data) as a means to mitigate the problems discussed in Part II. For example, the agency's Next Generation Compliance ("Next Gen") initiative involved innovative emissions technology, electronic reporting ("e-reporting"), and data analytics components that continue to drive the agency even after the initiative concluded in 2017. Part IV concludes with a discussion of innovative ways EPA can use the compliance data it has accumulated in AI technologies in the future. This Part highlights two recent studies that used machine learning to predict facilities' risk of noncompliance and to identify facilities that require a certain environmental permit but are currently operating without one. Problems with these potential future applications are also analyzed, including data accuracy issues and systematic biases that may be inherent in the data used.


    The environmental compliance regulatory system that EPA uses is based on the deterrence model, which is a model that involves the following two policy levers: 1) "frequency of inspections to enhance the probability of detection and" 2) "magnitude of sanctions." (17) If a facility needs to emit certain pollutants into the air, ground, or into a waterbody, it applies to EPA (or a state environmental agency) for a permit. If the permit is granted, monitoring devices are installed in the facility to measure the pollution, the agency conducts regular inspections of the facility, and adjudicatory enforcement actions are taken if the facility exceeds the emissions allowed under the permit or violates other relevant permit conditions. (18) EPA uses the two policy levers of the deterrence model to try to achieve an outcome that complies with the goals of environmental statutes such as the CAA and CWA, but this process is far from perfect.

    To start, the deterrence model does not map seamlessly onto environmental compliance initiatives. The model "assumes that (1) all expected benefits and costs of taking an action by the regulated entity are known and (2) collecting such information is costless." (19) In the environmental compliance arena, these assumptions do not hold. Entities regulated by environmental statutes include complex wastewater, industrial, and utility facilities that "may be uncertain about legal requirements, precluding cost-benefit analyses of possible compliance activities, or might lack the internal management capabilities necessary to undertake such evaluations." (20) Moreover, the traditional process for collecting information on these facilities is not costless--it involves considerable effort in terms of reporting and inspection requirements. (21)

    The imperfections of the deterrence model show, at a high level, why effective environmental compliance is difficult to achieve. This problem can also be discerned by looking at EPA specifically and the shortcomings the agency has dealt with over the years. For one, EPA and its state-run environmental compliance programs suffer from major data gap issues resulting from the difficulty of linking pollution to environmental degradation from a causation perspective. (22) "Data gaps haunt every scale of regulatory interest in environmental law," (23) and "these gaps affect problem identification, causal specification, evaluation of health and environmental impacts, valuation of harm, identification of rights, the nature of policy intervention, implementation, monitoring and enforcement, and updating and refinement." (24)

    By looking at specific EPA environmental compliance programs, these data gaps are easy to see. For example, one of the most prominent permitting programs under the CAA is the new source review (NSR) process, which requires regulated entities to install pollution control equipment if they build or modify facilities in a way that would create a significant increase in emissions of a regulated pollutant. (25) Comparatively, one of the most important regulatory programs under the CWA is the section 404 permit, which "regulate[s] the discharge of dredged or fill material into waters...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT