AI engineers don’t feel they’re not empowered to tackle the sustainability crisis, new research suggests

A study conducted by King’s College London of 23 machine learning (ML) practitioners in the UK and other parts of the world highlights alienation from model sustainability and suggests that environmental qualifications are not considered part of AI performance.

Despite the increasing number of tools to track the environmental impact of AI, responded with comments such as “As an individual, I don’t think you can really do much,” suggesting more needs to enable developers to create more greener models.

Visualization author and lecturer Dr. Georgia Panagiotidu said, “It underscores the fundamental lack of agency in the heart of AI and the conversation of sustainability. Despite tracking information and information about the environmental impact of ML, many developers feel that their industry is not focused on sustainability and that what is not important.

“This work collaborates to identify blockers to achieve spatial change rather than defeating individuals. By integrating sustainable thinking into all AI practices, we can address the lack of knowledge practitioners and provide the tools to make sustainable decisions in the face of the climate crisis.”

ML, a subset of AI, has seen major developments in recent years as AI tools play a bigger role in the global economy. This was brought about at a significant environmental cost. Global greenhouse emissions from the ICT industry have doubled over the past decade, with resource needs for ML model training training delaying the retirement of coal power plants.

A scope of ICT companies that train AI To address concerns about three emissions, tools like Code Carbon have been developed to estimate the emissions generated when running code or training AI, giving developers a window into the sustainability of the model.

Research analyzing information given by smart meters influences how consumers make more sustainable decisions, but little work has been done on how eco-feedback tools affect ML practitioners’ decisions.

Participants in the paper presented at the 2025 ACM Conference on Fair, Accountability and Transparency (2025) highlighted technical, personal, regulatory and cultural approaches to reducing carbon emissions, including the use of eco-feedback tools and the need for workplaces to streamline the energy use of AI training.

Nevertheless, most people felt limited responsibility to address environmental concerns, and the responsibility lies with large technical providers of large language models like ChatGpt.

Furthermore, participants described how their field, whether in academia or industry, considered sustainability as “not one of our outcomes, not an indicator of performance,” and was secondary to cultures that appreciated the accuracy and speed of high models in the production of papers and new products.

PhD students described the sustainability of taking the backseat in a competitive environment for research publication research. “I need to do my research. If I tell the supervisor NO, I’m not going to use HPC (high-performance computer) because I feel bad for Antarctica penguins.

Sinemgörücü, the first author candidate for PhD King’s and The Paper, said, “Qualitative interviews are essential to capturing how each individual thought about their work, and we were amazed at how empowered people are.

“Responsible AI has taught us that something around a conversation can actually be embedded deep in. Our research highlights that there is something to do to repeat that success story with sustainability.”

In the future, the team hopes to conduct large-scale quantitative research into machine learning practitioners’ perception of environmental sustainability.