S 3732 Introduce
AI Environmental Impacts
Earlier this month, Sen Markey (D,MI) introduced S 3732, the Artificial Intelligence Environmental Impacts Act of 2024. The bill would require the EPA to conduct a study on the environmental impacts of artificial intelligence. It would then require the National Institute of Standards and Technology (NIST) to convene a consortium “to identify the future measurements, methodologies, standards, and other appropriate needs, in order to measure and report the full range of environmental impacts of artificial intelligence”. No new funding is authorized by this legislation.
Definitions
Section 3 of the bill provides definitions for four key terms used in the legislation. One of the terms is defined by reference to an existing definition in statute. There are two technical terms of potential interest here:
Findings
Section 2 of the bill provides a series of ‘congressional finding’ that out line some of the environmental impacts that could be expected to be associated with the increase in the use of artificial intelligence systems. These include:
Energy consumption of computer hardware,
Energy and water (cooling) consumption for data center infrastructure,
Manufacturing facility energy and other resource consumption, and
E-waste.
Environmental Impact Study
Section 4 of the bill would require the EPA to prepare a report describing the results of, a comprehensive study on the environmental impacts of artificial intelligence. The report would include an examination of:
The energy consumption and pollution associated with the full lifecycle of artificial intelligence models, including the design, development, deployment, and use of those artificial intelligence models,
The energy consumption and pollution associated with the full lifecycle of artificial intelligence hardware, including the extraction of raw materials, manufacturing, and electronic waste associated with that hardware,
The energy and water consumption for the cooling of the data centers used in the design, development, deployment, and use of artificial intelligence models,
How choices made during the design, development, deployment, and use of artificial intelligence models, including the efficiency of the artificial intelligence models used, the location, power source, and design of data centers used, and the type of hardware used, impact the resulting environmental impacts,
Potential environmental impacts that could be acute at local scales, which may include added power loads that create grid stress, water withdrawals that create water stress, or local noise impacts,
The positive environmental impacts associated with applications of artificial intelligence, which may include optimizing systems for energy efficiency, developing renewable energy, advancing planetary systems research, enabling discovery of new materials, and automatically monitoring environmental changes,
The negative environmental impacts associated with applications of artificial intelligence, which may include rebound effects, behavioral impacts, and accelerating high-pollution activities,
Disparate impacts in the negative environmental impacts of artificial intelligence,
Other environmental impacts, as determined by the Administrator of the Environmental Protection Agency, and
The results of the updated data center study carried out under 42 USC 17112(e)(2).
Subsection (c) requires the EPA to solicit public comment during the development of the report.
NIST AI Consortium
Section 5 would require NIST to convene a consortium of stakeholders, specifically including members from academia, civil society, and industry, “to identify the future measurements, methodologies, standards, and other appropriate needs, in order to measure and report the full range of environmental impacts of artificial intelligence”. The goals for the consortium include:
Facilitating consistent, comparable reporting on the environmental impacts of the full lifecycle of artificial intelligence models, systems, and hardware.
According to technical feasibility, the development or cataloging of open source software and hardware tools and other resources designed to facilitate the measurement of environmental impacts of artificial intelligence models, systems, and hardware.
Providing recommendations on how to mitigate the negative, and promote the positive, environmental impacts of artificial intelligence.
Voluntary Environmental Reporting
Section 6 would require NIST to establish a voluntary environmental reporting program for the full range of environmental impacts of artificial intelligence.
Moving Forward
Markey and one of his five cosponsors {Sen Welch (D,VT)} are members of the Senate Commerce, Science, and Technology Committee to which this bill was assigned for consideration. This means that there may be sufficient influence to see the bill considered in Committee. With no funding, and no regulatory requirements, I see nothing in this bill would engender any organized opposition. I suspect that the bill would receive some level of bipartisan support were it considered. As with most bills introduced in the Senate, this bill is not politically important enough to take up the Senate’s time for consideration under regular order. I doubt that it would be a candidate for consideration under the unanimous consent process as it would draw the ire of a small number of Republicans who are almost knee-jerk opposed to any environmental legislation.
Commentary
I noted in an earlier blog post on this bill:
“Section 2, Findings, of that draft lays out the type effects that Markley, et al, expect to find in the reports required in the bill. While those concerns seem to be legitimate and should be substantiated and quantified by the reports, it is clear from the bill’s language that the AI concerns are really more appropriately concerns about the operation of large server farms regardless of the focus of the individual facilities.”
Singling out one type of server farm, one that is already under the political microscope, is political grandstanding. If this is a legitimate effort to look at environmental effects of large server farms, then it should not matter if they are in the service of AI, crypto mining, or a social media backbone, they have an environmental effect that may need to be regulated. The study and measurement system development would go a long way to determine if regulations is necessary.