Understanding the Carbon Footprint of ChatGPT
As artificial intelligence continues to become more and more commonplace, an important topic to consider is the environmental impact of these models. Generative AI, or the ability of an AI algorithm to produce complex data, is the technology behind chatbots and image generators, but how much energy does using these systems take and what is their environmental impact?
Generative AI algorithms often require more energy than their discriminative counterparts, in which the AI chooses from a fixed number of options. For instance, the 2019 study of “BERT” found that creating the model consumed energy equivalent to a round-trip transcontinental flight for a single person. An even larger model, “GPT-3”, would require 1,287 megawatt hours of electricity and produce 552 tons of carbon dioxide equivalent, the equivalent of 123 gasoline-powered passenger vehicles driven for one year.
However, the exact carbon footprint of an AI model is hard to estimate, and using more efficient model architectures and processors can reduce emissions by 100 to 1,000 times. It is estimated that a single query for a generative AI model could take four to five times more energy than a search engine query. This will become more prevalent as chatbots become more popular, and as large companies like Google and Microsoft incorporate AI language models into their search engines.
OpenAI released its chatbot model, ChatGPT, in 2022, and by 2023 it had over 1.5 billion visits. With the move to these AI models, the energy costs increase. In addition, these models often need to be updated over time, increasing the energy and carbon costs further.
On the other hand, chatbots are becoming a more direct way to get information than using a search engine, which can be more efficient in terms of energy use. Additionally, Ai models can be run on renewable energy, potentially reducing emissions by 30 to 40 times. Finally, increased transparency and societal pressure could help companies and research labs publish the carbon footprints of their AI models, providing clarity and potential for improvement.
Kate Saenko, the Associate Professor of Computer Science at Boston University, provided an interesting article on The Conversation website about the implications of the carbon footprint of AI models. Currently on leave from Boston University, she is working with Meta, Inc. to see first-hand what is going on in industry. This experience has enabled her to provide valuable insight into understanding the carbon footprint of chatbot models such as ChatGPT and the need to reduce emissions and prioritize renewable energy as these models become more widely used.
