![DOE's Pacific Northwest Laboratory near Richland, Wash.](https://cdn.govexec.com/media/img/cd/2024/08/22/pnnl/860x394.jpg?1724356063)
DOE's Pacific Northwest Laboratory near Richland, Wash. Photo courtesy Department of Energy
Energy Department wants to use AI to speed up permitting
The hope is that helping feds dive into historical environmental documents creates efficiencies and better outcomes.
As generative artificial intelligence and associated data centers drive up demand for electricity, the Energy Department is testing if AI can help the government speed up clean energy production.
The department is spending nearly $20 million to build and test AI-powered tools meant to speed up often years-long permitting processes.
The tools could help government employees tap into historical permitting and environmental data as part of its three-year project dubbed VoltAIc. The hope is that AI drives better outcomes and expedites decisions so that clean energy infrastructure can be built more quickly.
The Pacific Northwest National Laboratory, one of Energy’s national labs, is fielding the core project, called PolicyAI.
The focus is developing AI-powered software to augment federal reviews under the National Environmental Policy Act — a landmark environmental law that requires agencies to scrutinize significant environmental impacts of major actions, including issuing permits.
The department is currently testing prototypes internally and has built a data lakehouse — which combines the elements of data lake and a data warehouse architectures — of historical NEPA documents, said Sai Munikoti, PNNL data scientist and co-principal investigator for the project.
That lakehouse has information from over 28,000 PDFs from the Environmental Protection Agency aligning with nearly 3,000 environmental impact statements, one of three potential outcomes of the NEPA process.
These documents detail the potential impact of a proposed project on the environment, as well as potential alternatives. They can range from 150 to 300 pages, according to PNNL.
“We are trying to add value by making them more accessible,” Munikoti said. The effort includes standardizing the data and adding management and governance structures so that the documents are easier to search and can feed other AI applications.
For now, the lakehouse stores federal environmental impact statements from 2012 to 2023, although the hope is to eventually add environmental assessments and categorical exclusions — other documents that can result from a NEPA review.
While EPA houses all environmental impact statements, there’s currently no central database of these other types of documents, so getting them will require working across agencies that may be in various stages of digitization, said Davie Nguyen, deputy director for state, local, tribal territorial policy in Energy’s Office of Policy.
Even “within DOE, we don’t have one spot where you can download all of them,” said Nguyen. “It’s pretty difficult.”
Adding new documents as they’re created would also require working across agencies with various types of tech, but “we would like to see that functionality eventually,” said Keith Benes, a senior advisor in Energy’s Office of Policy,
So far, the team has built out a generative AI-driven semantic search function for the lakehouse that can find relevant documents and summarize search results, said Munikoti. The PNNL team is currently using open-source tech, but is working to create a model fine-tuned with NEPA data.
The search function not only shows users the documents themselves, but also gives a broader overview of a given project, such as its related documents, timeline and agencies involved, said Munitoki.
The team is also testing out two related pilots with Energy users as it works out agreements for employees at other agencies to also test the tools.
Meant to assist federal employees ushering projects through NEPA reviews, one of the tools lets users ask questions and get answers about a single document in a chat interface. The second also uses generative AI to allow users to “chat” with a group of documents.
The department is using proprietary models including Gemini and GPT-4 for these, although the goal is to eventually move to fine-tuned models based on open source technology for certain tasks, said Munikoti.
The VoltAIc team is also looking into using AI to help state and local reviewers as well, said Benes, potentially by doing things like inventorying all the relevant rules for EV chargers, for example. Using AI to sort through public comments garnered during permitting and environmental reviews is another project the team is considering.
The long-term future of any tools created in the effort — which Energy is envisioning as an agile sprint to explore possibilities — isn’t set. But the agency has also “actively started those conversations internally with other agencies about, ‘How would we make this a going-forward resource,’” said Benes.
If successful, the work has implications for the Biden administration’s efforts to implement milestone pieces of infrastructure and climate legislation, said Nguyen.
“The administration is really focused on, ‘How are we going to get these clean energy investments that we are making across all of the different pieces of legislation right now — How are you going to actually build those?” he said.
A less clunky, faster permitting environment would also support the department’s priorities, like modernizing the United States’ transmission system and electricity infrastructure, he noted.
“The grid infrastructure that we need, the clean energy projects that are going to help deliver on building energy security, combating climate change, advancing environmental justice, lowering energy costs for families… You can't do all those things if you're not actually able to build the things that are helping to deliver that clean energy,” he said.
The Permitting Council put $6.1 million into the project as part of a larger set of investments to improve permitting tech across the government. As the White House’s Council on Environmental Quality has reported, currently agencies use disjointed technologies to power their permitting processes.
Although the Energy Department is not technically a big agency player in terms of actually permitting projects — federal permitting usually involves multiple agencies, but the specifics differ depending on the project — it’s taking on the problem because of its AI expertise, said Nguyen.
The AI executive order issued last fall also directed Energy to develop foundation models to streamline permitting and environmental reviews.
Part of the work has involved speaking with potential users of these products, Benes said.
Broad stakeholder engagement will be key to ensure that whatever comes out of the effort is actually helpful, said Jessie Mahr, director of technology at the nonprofit Environmental Policy Innovation Center.
Efforts to streamline permitting should also include the removal of the bureaucratic hang-ups that slow down clean energy or restoration work in the first place, she said, noting that “permitting is a means to that end.”