The AI boom spells bad news for the planet: The training of AI chatbots and image generators like Open AI’s ChatGPT requires tremendous amounts of energy and adds tons of CO2 to the atmosphere, Bloomberg writes. Training AI programs is a months-long process that sees researchers use power-intensive chips known as graphics processing units in data servers to help large language models like ChatGPT analyze data.
How much energy are we talking? Tech giant Google says artificial intelligence ate up between 10-15% — or 2.3 TWh — of its overall electricity consumption in 2021, the business information service notes. A single training of OpenAI’s GPT-3 reportedly generated some 502 tons of CO2 and consumed enough energy to power 120 US homes for a year, according to research. While some companies use the AI programs to map out courses to optimize energy efficiency as in the case of Google — which is using AI to control cooling and heating at its facilities — the need to source the energy it eats up from renewable sources is paramount, Bloomberg notes.