Natural language processing (NLP), a form of deep learning, analyses human language to create everything from more accurate predictive texting to refining search engine results. Yet, as the technology has developed, its processes have become increasingly energy intensive. A new study from the University of Massachusetts reveals that the process of training and optimising one large NLP model emitted more than 626,000lbs of carbon dioxide – nearly five times the lifetime emissions of the average American car, including those from its manufacturing. Of note, the most energy intensive part of the NLP modelling process is the optimisation – the part where the machine repeats and refines different tasks to alter the overall algorithm.
In environmental terms, such insights are alarming. Tech companies are being tasked to address this impact and take a more considerate approach to innovation. As Emma Strubell, lead researcher on the University of Massachusetts study, notes: ‘I’m not against energy use in the name of advancing science, but I think we could do better in terms of considering the trade-off between required energy and resulting model improvement’. She adds: ‘Large tech companies that use AI throughout their products are likely the largest contributors to this type of energy use.’
As a solution, Strubell highlights that big tech companies should not only source or create more renewable energy, they should provide ‘equitable access’ to their data for academic researchers, allowing them to independently train better algorithms that could then be used in turn to help others optimise their processing, all with less environmental impact.
This more moral mindset is becoming increasingly necessary for tech companies in an era when consumers are more aware of the ways in which the products, services and even their connectivity impacts planetary health. And with consumers expectant of companies to be transparent about their operations, digital supply chains will soon come into focus.
Looking to a future of brand experiences led by AI and intuitive solutions, companies must therefore take the steps to ensure their algorithms are sustainable, not only in their creation, but their refinement. In this way, there is an opportunity to openly demonstrate their positive contribution to optimising and safeguarding our world, while actively reducing the carbon footprint of innovation.
Camara Clarke is The Future Laboratory’s graduate trainee. For more on how tech companies are transforming themselves into climate-positive industry leaders, read our Sustainable Data Centres microtrend.