Brightcon 2026, hackathon and courses

Artificial intelligence (AI) is increasingly embedded in life cycle assessment (LCA) workflows, supporting inventory completion, literature synthesis, scenario analysis, and interpretation. Yet while AI is often framed as an efficiency gain for sustainability assessment, the environmental impacts of AI workflows themselves remain largely unexamined. Current AI-enabled LCA approaches tend to prioritize accuracy, scale, or speed, with limited attention to energy use, carbon intensity of inference, model size, prompt design, or iteration frequency. This creates a disconnect between sustainability goals and computational practice.

This presentation addresses the absence of LCA-aware design principles for AI workflows. We discuss how AI can be treated as an explicit system component within collaborative LCA practice, rather than as a neutral tool operating in isolation. Drawing on recent applications of lightweight, open-source language models and modular machine-learning pipelines, we explore how design choices in AI-enabled LCA tasks can influence both insight generation and computational burden. Rather than focusing solely on model performance, we highlight practical tradeoffs associated with task-specific model selection, constrained prompting, limited inference passes, shared infrastructure, and transparent reporting of proxy metrics such as runtime, token use, and model scale.

Using a small, reproducible case study using Brightway for life cycle inventory tasks, we present an emerging framework for designing low-carbon AI workflows that balance decision relevance with computational cost. Open-source code (Jupyter, Python) will accompany this presentation for adoption by others. By reframing AI adoption as a collective, tool-sharing design challenge, this work encourages reproducible, efficient, and environmentally coherent collaboration across researchers, consultants, and software developers using AI in LCA.