The rapid expansion of artificial intelligence has triggered a growing debate about its environmental cost. Researcher Sasha Luccioni, a leading voice in the field of AI sustainability, argues that the industry currently lacks two critical pieces of information: reliable emissions data for AI systems and a clear understanding of how these systems are actually used.
Luccioni, who works at the intersection of machine learning and climate science, has spent years analyzing the carbon footprint of large language models and other AI tools. Her research suggests that the environmental impact of AI is often underestimated because emissions reporting remains inconsistent and incomplete.
Why Emissions Data Matters
Without standardized metrics, companies and researchers cannot compare the energy consumption of different models or identify the most efficient approaches. Luccioni emphasizes that transparency is essential for accountability. If developers do not know how much energy a specific AI task requires, they cannot make informed decisions about where to deploy resources.
Several major technology firms have begun publishing environmental reports, but the data often covers only the training phase of a model. Luccioni points out that inference, the process of running a trained model on new data, can consume far more energy over time than training. This imbalance is overlooked in many sustainability assessments.
Usage Patterns Remain Opaque
Beyond emissions, the researcher highlights a second blind spot: the lack of detailed usage telemetry. Many AI services do not disclose how frequently their models are queried, for what purposes, or with what data. This opacity makes it difficult to allocate computing resources efficiently or to identify unnecessary usage that drives up energy demand.
Luccioni argues that understanding usage patterns is just as important as measuring emissions. For example, a highly efficient model that is used millions of times per day may have a larger aggregate carbon footprint than a less efficient model used sparingly. Without usage data, policymakers and engineers are essentially working in the dark.
Toward a Sustainable Framework
Luccioni advocates for a combination of regulatory standards and industry self regulation. She supports mandatory disclosure of energy consumption and carbon emissions for AI systems, similar to fuel economy labels for vehicles. She also encourages developers to design models with efficiency as a core requirement, not an afterthought.
Several research groups have developed tools to measure AI energy use, but adoption remains voluntary. Luccioni believes that until measurement becomes universal and transparent, meaningful progress toward sustainability will be slow.
The challenge is not purely technical. Economic incentives currently favor raw performance and speed over efficiency. Changing this dynamic will require pressure from consumers, investors, and regulators alike.
Looking ahead, Luccioni expects that the conversation around AI sustainability will intensify as legislation such as the European Union’s AI Act comes into effect. These rules may include provisions for energy reporting and lifecycle assessments. The AI industry, she notes, will need to adapt quickly to these new expectations or risk facing stricter mandates.
