This article is published in the April 2024 issue.

CCC Responds to RFI on DOE’s Responsibilities on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence

In: /

By Haley Griffin, Program Associate, CCC

On April 1, 2024, the CCC submitted a response to the Department of Energy (DOE)’s Request for Information (RFI) Related to Responsibilities on Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. In the solicitation, DOE sought advice on their plan to carry out some of the responsibilities outlined for them in the October Executive order (E.O.), “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence”. The RFI response was written by the following CCC Council Members and staff: Nadya Bliss (Arizona State University), Haley Griffin (CCC), Michela Taufer (University of Tennessee, Knoxville), and Adam Wierman (California Institute of Technology).

The authors were impressed with the DOE’s list of areas of focus and opportunities in AI application, but they also cautioned that AI can have significant detrimental environmental impacts. There are also many opportunity areas in computing beyond AI that they urge DOE to consider, as explained in the CCC Whitepaper, “Computing Research for the Climate Crisis.”

The CCC RFI Response was organized into the three major topics identified by DOE: (1) AI to improve the security and reliability of grid infrastructure and operations and their resilience to disruptions, (2) AI to improve planning, permitting, and investment in the grid and related clean energy infrastructure, and (3) AI to help mitigate climate change risks. Below are recommendations CCC provided under each area.

Topic 1. AI to improve the security and reliability of grid infrastructure and operations and their resilience to disruptions

  • Large scale compute/grid infrastructure operations: The authors would like this to be added to the list of opportunities. There is a greater need for data centers than can realistically be supported in many areas of the US, and especially with the addition of AI to the grid, there is a massive construction load anticipated in the near future. The authors recommend (1) investing in software solutions designed to reduce the costs associated with accessing and using data remotely, and (2) programs that work jointly with utilities and datacenter hyperscalers to incentivize developing compute infrastructure outside of peak regions.
  • Understanding carbon impact: This is an underfunded area of AI research especially at the local, regional level. This could entail using AI to predict renewable energy availability and to determine where marginal carbon contributions are on different parts of the grid in order to modulate operation charging, discharging, usage, economic incentives, skiffing, etc. This is especially important at the regional level where the impact of decisions can be felt more immediately and strongly.
  • Quantification and consideration of risks of AI use: There is a need for context-specific uncertainty quantification tools to understand when and to what degree you can trust the advice/prediction from AI. There also needs to be safeguards integrated to account for the significant security risks associated with an AI system managing grid resilience, and because there would be drastic consequences in society if these systems make even small errors.
  • Importance of AI considering economic impacts: The management of power systems requires integration between economics and engineering. There is significant danger in optimizing just the engineering of these systems because of the potential to impact market structures negatively.
  • Knowledge transfer of similar AI deployment: Integrating AI into existing infrastructure is a strategy successfully applied in various sectors, such as rail infrastructure. These sections share critical objectives with the proposed plan for enhancing energy grid infrastructure, including the need for continuous monitoring, improving operational efficiency, and bolstering resilience against disruptions.
  • Need for digital twins: One of the key areas that needs to be supported is publicly accessible, robust, and detailed digital twins. Making use of AI in the domains listed requires detailed digital twins to be available for research and testing/validation. Digital twins would also be very useful in addressing Topics 2 and 3.

Topic 2. AI to improve planning, permitting, and investment in the grid and related clean energy infrastructure

  • Understanding the impacts of AI-recommended clean energy investments: There needs to be a continuation of research into the impacts of following recommendations made by AI in using wind and solar energy. AI is being used to project, predict, and optimize the use of these technologies, and there needs to be an understanding of how well it did and if it was worth using.
  • Strategic placement of data centers to limit harm to communities and the environment: Data centers have a significant impact on the power grid, and their location, orientation, large-scale storage generation placement, etc. needs to be heavily considered. These factors are only going to be more significant with AI increasing the energy demand. There are already disproportionate impacts on some communities, and in certain areas, a data center coming in would have devastating consequences on the prices of water and electricity in addition to environmental concerns like pollution. 
  • Need for specific use case AI: AI can have disastrous consequences if it is deployed in a setting that it was not trained for. There needs to be significant investment in LLMs for specific purposes that have specialized training data that is related to the area where it is going to be used. 
  • Need for cyberinfrastructures for data accessibility and use: The convergence of High-Performance Computing (HPC) with Cloud Computing is critical for broadening equitable access to data. The convergence underscores the importance of investing in infrastructure and technologies that allow for seamless remote data access, thereby minimizing the energy footprint associated with data movement. Integrating AI can optimize this convergence, enabling more efficient data processing and analysis methods.  

Topic 3. AI to help mitigate climate change risks

In this section of the RFI Response, the authors referenced the “Computing Research for the Climate Crisis” CCC Whitepaper that highlights the role of computing research in addressing climate change-induced challenges. In the Whitepaper, the authors describe four broad areas of computing research (AI/robotics/algorithms, devices & architectures, software, and sociotechnical computing) that could make progress in six key impact areas: Energy, Environmental Justice, Transportation, Infrastructure, Agriculture, and Environmental Monitoring & Forecasting. They emphasize the need for interdisciplinary teams that bring computing researchers together across these areas along with engineers, as well as professionals from the social, behavioral, economic, and physical sciences.

In the Whitepaper, the authors specifically provide the ways that AI, robotics, and/or algorithms can help make progress in each key impact area:

  • Energy: Planning, optimization, and decision support for production, distribution, and consumption of energy; AI-enabled materials science for renewables.
  • Environmental Justice: Modeling and decision-support strategies that leverage those data and manage cascading risks.
  • Transportation: Spatiotemporal planning strategies to optimize the routing of flows in the network.
  • Infrastructure: Optimization and decision support of flows of energy, goods, water, vehicles, people, power, etc.; AI-enabled materials science for green materials.
  • Agriculture: Algorithms that leverage rich sensor data, together with real-time information about economic factors and transportation networks, for planning and risk assessment.
  • Environmental monitoring & forecasting: Uncertainty quantification; system-level, risk-sensitive modeling, planning, and optimization strategies for climate variables, at all scales.

The CCC RFI Response wraps up by encouraging DOE to (1) engage in community outreach efforts to explain how AI is being used and the way it will improve planning, the security risks, and mitigations being taken, etc. and (2) to collaborate with other federal agencies on these efforts, especially those that have significant security implications like grid resilience.

Read the full CCC RFI response here.