Using AI for Transparent Policymaking

Artificial Intelligence (AI) may help bridge the gap between scientific research and policy in the plastics industry.

Better Data, Better Decisions

Data is a key driver of evidence-based policy. Communicating the intricacies of the plastics industry to decision makers is crucial for effective governance. To support the interface of science and policy, researchers developed an AI-driven framework using plastic lifecycle environmental impacts (LCEI) data. This framework can help researchers and policymakers synthesize large-scale datasets to make informed decisions.

You can also read: Smart Plastics: How Data Intelligence Is Reshaping Production

LCEI data comprises interconnected elements, such as the studied product, system boundaries, methodological choices, and underlying data sources. This complex data can be difficult to parse using a large language model (LLM) without customization. Researchers used a customized LLM-driven workflow to build a Plastics LCEI database with over 65,536 data points.

Developing a Knowledge Base

To create the custom LLM-based framework, researchers developed a plastics lifecycle knowledge base. This knowledge base served to improve prompt precision and contextual understanding for the LLM. It focused on various areas vital to understanding the plastic lifecycle, including the fundamental properties of plastics. The knowledgebase also reviewed over 200 common production pathways for plastic resins and 130 typical recycling technologies.

Creating a Structured LCEI Database

With the foundation of the knowledge base, researchers created an automated workflow combining prompt engineering and LLM-driven information extraction. This process resulted in a structured database for LCEI in the plastics industry.

The researchers conducted multi-dimensional statistical analyses to visualize the data. Figure courtesy of Artificial intelligence-driven framework for science-policy interface on global plastic life cycle environmental impacts.

Overcoming Bias in Data

A database is only as useful as its source data. Research literature may introduce bias into the workflow, underscoring the importance of data quality control. Uncertainty analysis is a pathway to improving LLM results. Out of three tested machine learning (ML) models, researchers chose eXtreme Gradient Boosting (XGBoost) for further training and analysis. Afterwards, they screened the data from this model, which resulted in the retention of higher-quality records. These records were only 12.76% of the original data volume, highlighting the need for data quality control.

Learning from the LCEI Database

The resultant plastics LCEI database provides insights into research priorities. By plotting gaps in the data, researchers can visualize where research is lacking. For example, emerging technologies such as chemical recycling have scarce data. For policymakers, this lack of evidence may be a barrier to incentivizing the diversification of plastics end-of-life strategies.

Visualizations resultant from this LLM-based framework can give a top-down view of research gaps. Figure courtesy of Artificial intelligence-driven framework for science-policy interface on global plastic life cycle environmental impacts.

Another large gap is evident for engineering plastics, such as polyphenylene oxide, and specialty engineering plastics (liquid crystal polymer, polyacrylate). Research around additives in the context of LCEI also remains limited. Data deficiencies may affect policy outcomes due to a lack of scientific evidence. Reliable LCEI data can help policymakers identify intervention points, appropriate substitutions, and design targeted incentives for industry stakeholders.