By: Marla Orenstein
My father-in-law called it “despicable deskwork.” He was a neurologist, and wanted to spend his time either talking with his patients or—preferably—hiking in the mountains. At the bottom of his list was sitting at his desk, filling out forms and transferring notes from one piece of paper to another.
I actually like sitting at my desk, and am not generally a fan of hiking (at least uphill). But I also don’t enjoy spending hours, days, or weeks on repetitive and uncreative tasks that I know can be done more efficiently. Who does, really?
There are many parts of the impact assessment (IA) process that fall into the “despicable deskwork” category of repetitive, data-heavy tasks, which can be a great fit for AI.¹ When combined with AI’s ability to synthesize huge amounts of information and make predictions, the parts of IA that used to eat up weeks can start taking hours, leaving more time for the work that actually requires a human in the room.
Here are the five use cases for AI in IA that I am most excited about at the present moment, as well as some cautions against an overly-rosy picture.
- Counting the birds and the trees
An impact assessment starts by establishing baseline environmental conditions. Historically, this has been done by people going out in the field and conducting surveys. But in recent years environmental monitoring has undergone a data explosion. Advances in technology have resulted in a profusion of data from remote networked sensors, satellites, drones, and airborne LiDAR as well as from field studies, citizen reports and historical records. The problem now is not obtaining enough environmental data, but analyzing and integrating it. It’s no wonder that AI is being used to help with data management—this is pretty well a textbook example of how to leverage AI’s strengths.
- Making predictions about alternative scenarios and cumulative effects
Weather forecasting will never be perfect—but it’s gotten remarkably better, remarkably fast. Two things have driven performance gains: vastly more data, and AI models that outperform traditional approaches. A modern 5- to 6-day forecast has roughly the same accuracy as a 1-day forecast had in the 1980s.
IA faces a similar challenge in the need to predict how complex, interconnected systems will respond under different conditions. This might mean comparing impacts across alternative project designs. Or predicting cumulative effects—how the impacts of the proposed project will combine with those of other existing or planned projects nearby. Or even or tackling the puzzle of multiple simultaneous chemical exposures, which is something traditional toxicology methods struggled to address.
As in weather forecasting, impact assessment now has much more data available, and AI modeling is a tool that can substantially improve how that data is used. This area is still in its infancy, but I am watching with great interest as it opens up capabilities that simply didn’t exist before.
- Enabling different parties to engage with IA information
Impact assessments generate more information than any one person can read in full. The regulatory application for the Trans Mountain pipeline expansion, for example, ran to over 20,000 pages – two meters of binders! AI chatbots trained on project documents can help make this more manageable. Regulators, rightsholders, community members, or other interested parties can query the chatbot to quickly locate and understand information about the project and the topics they care about—in any language.
An unresolved problem, at least currently, is that chatbots may invent incorrect but authoritative-sounding answers rather than admit ignorance when there is a gap in knowledge.²
- Collating stakeholder feedback
Stakeholder input—opportunities for people to comment on a proposed project—arrives in messy, unstructured formats: emails, meeting notes, social media posts, survey data, written submissions, comment letters, etc. And in an impact assessment, the volume of stakeholder feedback can be overwhelming, particularly if the project is contentious. AI’s ability to quickly synthesize and organize different forms of data can really help in rapidly summarizing what stakeholders have to say.
A potential hazard is that AI struggles to distinguish what truly matters from what merely appears most often. Without the contextual expertise to know what it is looking at, AI risks missing what is most important.
- Preparing regulatory application documents
Regulatory submissions are enormously time-consuming to produce. They require pulling data from multiple sources, checking its quality, and structuring it in the specific way each regulator expects. AI can take a strong first pass at this: assembling and organizing information, flagging gaps, and verifying compliance with required methodologies and formats. A human still needs to validate the output and catch any errors before submission, but letting AI do the heavy lifting first can save significant time and effort.
AI holds enormous promise for the field of impact assessment, but there are important questions still to be worked out around accountability, transparency, bias, accuracy, safeguards, and regulation. In addition, there are parts of the IA process, particularly around decision-making and building trust with rightsholders and stakeholders, that should never be outsourced to a machine.
But the potential gains in efficiency and data management—as well helping people avoid despicable deskwork tasks—means that a future of IA experts partnering with AI is almost inevitable.
¹ Bond A and Dusik J. 2025. Artificial intelligence in impact assessment: the state of the art. DOI: 10.1080/14615517.2025.2594274
² Jones N. 2024. Bigger AI chatbots more inclined to spew nonsense-and people don’t always realize. Nature. https://doi.org/10.1038/d41586-024-03137-3
