INSIGHT: How Artificial Intelligence Will Change Environmental Compliance

The broad subject of artificial intelligence makes headlines on a daily basis, whether in the form of self-driving vehicles, virtual assistants like Apple’s Siri and Amazon’s Alexa, dire predictions of automated systems replacing even more human jobs, or woven into pop culture from “Blade Runner” to “Westworld.”

Yet absent from all the attention to AI is any discussion of the Clean Air Act’s permitting requirements or implications under the air law, Clean Water Act, and other environmental laws. Admittedly, these issues aren’t exactly attention-getting topics for the average person or particularly catchy themes for a new movie or TV show. But in the world of environmental law, artificial intelligence is actually an emerging issue of great potential interest.

Philosophers and computer scientists can argue over where sophisticated programming ends and AI begins. From the standpoint of environmental legal requirements, the distinction isn’t important. Or at least not yet. Instead, the real focus is on how this technology interacts with and affects regulatory requirements.

For the purposes of the discussion here, any computer software program that has a modicum of machine learning capabilities will be considered to be AI.

More Efficient, But More Pollution

The leading edge of this issue is air-permitting requirements under the Clean Air Act. A primary feature of the law is a program known as New Source Review.

Under New Source Review, major manufacturing or production facilities that undergo a physical change or a change in the method of operation must evaluate whether their potential for air emissions will increase as a result. If there is a significant increase in air emissions, or even the potential for such an increase, then a permit may be required. The permitting process triggers emissions control requirements and limits on how much pollution can be emitted.

Typically, an upgrade to a software system wouldn’t even raise a hint of a concern under New Source Review. But using AI systems to control aspects of a manufacturing operations introduces a new and distinct issue. AI systems will be used not just to collect information and provide it to operating personnel but also be tasked with using that information to make decisions that were formerly made by real people.

In order to do this, it is likely that sensors and monitoring equipment of various kinds will have to be installed to provide the necessary informational inputs to the AI system so that it can carry out its mission. The use of AI is intended to provide efficiencies in production, which leads straight to the likelihood that production increases and so do air emissions.

Given this situation, is there a potential for New Source Review permitting requirements to be triggered either as a physical change or a change in the method of operation? The answers are: “yes” in the first case, and “maybe” in the second case.

Guidance on Physical Change

First, installing new sensors and monitoring equipment is a physical change, and that in and of itself, can fairly be considered a triggering event. Environmental Protection Agency guidance and supporting court opinions generally instruct that “physical change” is to be interpreted broadly.

This conclusion, and indeed even the very question, will likely come as a surprise to a typical plant operator, who isn’t used to considering such minor physical changes, let alone new software, as potentially triggering air permitting requirements. But where such physical changes are in service to an AI system that brings with it operating efficiencies that introduce the real potential for emissions increases, suddenly a legitimate permitting question is introduced.

Indeed, the operating efficiencies for a production line are often a key motivating factor in the decision to use artificial intelligence software in the first place. As such, the installation of an AI system, even if seemingly involving primarily a virtual change can trigger New Source Review if accompanied by physical changes necessary to implement the AI.

Does AI Change Operations?

Second, there is even a potential that the use of AI could be considered a “change in the method of operation” such as to trigger permitting concerns even without any physical changes. This potential interpretation is far more speculative since “change in the method of operation” isn’t well-defined in the Clean Air Act or its regulations. Arguably, from the perspective of a “change in the method of operation,” use of AI is no different from a change in personnel, which has never been considered a triggering event.

Take the following example, involving Joe and Sally. Joe is our long-term production line supervisor. He is a great guy but getting up there in years and prone to moving slowly, taking long breaks, and occasionally napping. Sally is about to take his place; she is younger, full of energy, very attentive to her job responsibilities, and keen to earn bonuses for high production levels.

Merely replacing Joe with Sally, even if it increases production and results in more air emissions, wouldn’t be considered an event that would require permitting. Replacing both Joe and Sally, however, with an AI software system that runs the production with greater precision and higher production and emissions than Joe—and maybe even than Sally—could potentially be considered a “change in the method of operation.” This is an unanswered question at this point, and it will be interesting to see how the law on this question develops.

Looking beyond the New Source Review issue, it is notable that both the Clean Air Act and the Clean Water Act impose standards that require the use of best available technology. While these are detailed and complicated regulatory schemes, the essence is permit limits for air or water pollution are developed based on a metric that the facility will use the best available technology.

Typically, best available technology is considered in terms of air control devices such as scrubbers or water treatment methods such as reverse osmosis. AI systems, however, offer the potential to collect and analyze a lot of data very quickly to fine-tune aspects of both the production line and operation of pollution control equipment.

As such, an AI system, appropriately used, should be able to operate such systems more effectively to limit the pollution that is created by the production process and/or emitted or discharged into the environment after passing through pollution control equipment. Once such systems begin to be adopted, one can forecast a future where AI systems won’t only be options but will establish themselves as the best available technology. At that point, use of such systems could become effectively mandated by the technology requirements of the Clean Air Act and Clean Water Act.

More Data, More Violations

Use of AI systems also present a risk of inadvertently making pollution limits more restrictive. Many environmental permit limits are expressed in terms of the mass of pollution (e.g. so many pounds per hour or day) or concentration (e.g. so many grams per liter) of a given chemical or compound. Yet, a practical component of a limit is how often compliance with that limit is checked. Compliance can be measured on the basis of once per hour, per day, per week, per year, etc. Obviously, with such intermittent testing, minor, short-term, or infrequent exceedances have little chance of being detected.

When bringing in an AI system, however, the information needs of the system can result in a more frequent or even essentially continuous data sets. This then provides a method to determine compliance status on a much more frequent basis than previously. It is as if a driver has gone from watching out for occasional speed traps to driving a car that automatically informs the police any time the speed limit is exceeded. The net result is the pollution limits have effectively become much more stringent as a result of compliance information being checked more frequently by an AI system.

Check the Numbers

Another interesting and emerging aspect of use of AI is the relationship to compliance certification obligations. A number of environmental programs require company officials to certify the accuracy of data and information submitted to the government. A false certification is subject to civil and criminal penalties. Criminal prosecutions for environmental violations aren’t that common. But when they do occur, they are frequently based on falsification of information or false certifications.

It isn’t necessary that a company official have direct personal knowledge of the information that is being submitted and certified in a report to the government. Instead, it is acceptable that the company official make reasonable inquiry with persons with knowledge of the information that is being reported if there is a legitimate basis to conclude that the information those persons are providing is complete and accurate.

When a process and the relevant information sources are an AI system or one subject to the control of an artificial intelligence, however, it is unclear how the company official can or should fulfill his or her certification obligations. What happens if a certification is later found to be incorrect? What does a company official need to know, check, or verify in order to be able to rely on conclusions made by an AI system? This is unknown, with no clear established precedent.

Looking ahead, though, responsible company officials would be well-advised, in their own self-defense, to develop criteria and procedures to define a defensible basis to rely on information from AI systems at the time that those AI systems are installed.

Undoubtedly, AI will affect environmental protection and legal requirements in myriad ways not forecast here. But looking into the crystal ball, there are already a number of potential impacts and concerns that should be recognized and strategized as AI system begin to be adopted for activities affected by environmental laws and regulations.

David A. Rockman is a Pittsburgh-based attorney with Eckert Seamans Cherin & Mellott and helps clients manage environmental compliance and environmental risks. He helps clients understand and meet their legal obligations with respect to federal, state, and local environmental laws, regulations, and permits. He also defends clients facing enforcement actions, represents clients involved in environmental litigation, and provides representation with respect to environmental issues in the purchase and sale of businesses and property.

The opinions expressed here do not represent those of Bloomberg Environment, which welcomes other points of view.