AI Can Strengthen Technical Judgment, Not Replace It

Like any tool, the value of artificial intelligence depends on who is using it and how.

As a geochemist, I’ve spent many years working with chemicals and elements in the environment. Experience and established scientific principles give me a foundation for understanding their behavior, but no one can be an expert on everything.

When I encounter a question about an unfamiliar contaminant or issue, AI can help me quickly get up to speed on what others have learned, what research exists, and what factors might be important.

This doesn’t mean accepting whatever AI produces, of course. I still need enough background knowledge to determine whether the information makes sense based on the scientific principles that govern how chemicals behave in the environment. AI can accelerate the process of gathering and synthesizing information. Technical judgment determines what information to trust and how to use it.

The same principle applies to data analysis.

Scientists and engineers can spend hours writing scripts, analyzing complex data, and creating graphics to communicate results. Increasingly, AI can help with the coding and automate portions of that process. This can fundamentally change where we spend our time.

Creating the graphics is necessary. Understanding what these graphics are telling us is more important.

If we can spend less time on the mechanics of producing an analysis, we can spend more time examining the results, confirming that they are sound, identifying what matters, and telling the story contained in the data.

Using Machine Learning to Look Ahead

Machine learning can take that idea a step further by helping us look not only at what has happened but at what may happen next. In the context of environmental monitoring, this can turn years of data into a tool for understanding how a system behaves, anticipating where it may be headed, and helping us decide what we should consider doing next.

Here’s one example: Working with Gunnar Guddal and other colleagues at Anchor QEA, we developed a physics- and chemistry-informed neural network (PCINN) model that we have been applying at a remediation site where groundwater is contaminated by metals and organic compounds. Years of monitoring have produced a substantial groundwater chemistry dataset that includes the contaminants of concern as well as key geochemical indicators that provide insight into the processes occurring at the site.

Following implementation of reagent injection for in situ groundwater treatment, the site has been moving toward its cleanup goals. But after years of monitoring, important questions have emerged: When will different areas of the site reach those goals? Where might cleanup take longer? Where can monitoring potentially be reduced and over what time frame? Where might additional action eventually be warranted to achieve site closure?

By analyzing how groundwater quality at the site has evolved over space and time, the model can identify patterns and make projections about future conditions. We can then interpret those patterns in the context of the geochemical processes occurring at the site. Each year, we can compare those predictions with actual monitoring results, update the model with new data, and evaluate how well its projections are holding up.

This adds another dimension to long-term environmental monitoring by extracting greater predictive value from the data we already collect.

Retaining the Human Element

The objective isn’t to abdicate decision-making to an algorithm. It’s to give scientists, engineers, and our clients another tool for making better decisions. For example, a model may indicate that one portion of a site is progressing as expected while another could take considerably longer. The technical team still has to evaluate why, determine how much confidence to place in the prediction, and decide what that means for the path forward.

This is where I see some of the greatest potential for AI and machine learning in environmental consulting. These tools can help us find information faster, automate time-consuming analytical tasks, and extract more value from the data we collect.

But the most important parts of the process remain human: understanding the science, asking the right questions, checking the answers, and determining what the data are actually telling us.

AI doesn’t eliminate the need for technical expertise. Used well, it helps us focus that expertise on what matters most.

Dimitri Vlassopoulos is a Principal Scientist at Anchor QEA.

Contact Dimitri to learn more.