Researchers at Harvard University and the Stevens Institute of Technology have developed a new artificial intelligence model designed to improve understanding of potentially harmful byproducts in drinking water that result from common disinfection processes. Chemicals such as chlorine and chloramine are widely used to treat tap water, but when they interact with organic matter naturally present in water, they can form byproducts that may pose health risks, including links to cancer and effects on fetal development. Traditional regulation does not cover all of these compounds, so scientists trained the AI on toxicity data from more than 200 known chemicals and used it to predict the dangers of over 1,100 related byproducts.
The AI model helps identify which of these countless disinfection byproducts could be more toxic than those currently regulated, giving researchers and regulators a broader view of potential risks that have been difficult to study at scale. While the findings do not suggest that everyday drinking water is unsafe, the enhanced predictive capability can inform deeper research and potentially strengthen safety standards. By pointing scientists to byproducts that merit closer investigation, this approach could support improvements in monitoring, regulation, and treatment practices to protect public health.
Overall, the work highlights how machine learning can accelerate environmental health research, especially in complex systems where traditional testing would be slow or incomplete. As researchers refine the model and expand the scope of analyses, AI could become an important tool for safeguarding water quality and helping regulators make more informed decisions about contaminants of concern.
