TOWARDS BUILDING AN ENERGY SUSTAINABLE AI FUTURE

ARTIFICIAL Intelligence (AI) has the potential to revolutionise efficiencies across the healthcare, finance, transportation, and manufacturing sectors.

- Amit Kapoor & Nabha Joshi

Another domain, sustainability, has emerged as an urgent priority, tackling pressing concerns such as carbon emissions, environmental degradation, and climate change in general. Prima facie, these two domains may seem to be disparate realms. Yet, their intersection reveals a telling narrative that the energy-intensive nature of AI development necessitates embedding sustainability into its framework.

Artificial Intelligence has become an integral part of technological and governance mechanisms. The edge AI has over other technologies lies in its ability to handle, process, train, and analyse data at a speed incomparable to any other prevailing technologies. Its interoperability and adaptability have made it an effective tool in delivering services in core sectors like healthcare, education, agriculture and governance. Data centres that provide uninterrupted high-density power are central to these high-performing AI technologies. This high demand for power makes AI an environmental challenge. The International Energy Agency (IEA) reports that these data centres account for 2.5-3.7 per cent of global greenhouse emissions and account for 0.9-1.3 per cent of global electricity demand.

The Economic Survey 2023-24 states that the global power demand from data centres can surpass 1000 Terawatt hours (TWh) by 2026, twice France and Germany’s current net power demand, which stands at 500 TWh each. According to a study by Cornell University, training large language models, such as GPT-3, results in electricity consumption comparable to 500 metric tons of carbon dioxide, roughly translating to 1.1 million pounds. An estimate by the International Energy Agency posits that each query made through ChatGPT consumes ten times more energy than a similar search on Google.