Productive AI can create complex outputs such as sentences, paragraphs, images and even a short video. It has long been used in applications such as smart speakers to generate voice responses or to suggest a search query in autocomplete. However, we see that he has recently acquired a human-like language and the ability to produce realistic photographs.The energy costs of building AI models are a matter of curiosity for some people. The stronger the AI, the more energy it consumes. So how will the emergence of increasingly powerful, productive AI models affect society’s future carbon footprint?
Equivalent to using 123 petrol passenger cars for 1 year
In 2019, researchers found that building a generative artificial intelligence model called BERT with 110 million parameters consumes the energy of a round-trip intercontinental flight for one person. The number of parameters refers to the size of the model, and larger models are generally more resourceful. The researchers estimate that building the much larger GPT-3 with 175 billion parameters consumes 1,287 megawatt hours of electricity and produces 552 tons of carbon dioxide, equivalent to using 123 petrol passenger cars over the course of a year. Because that’s just the energy used to get the model ready for market before any consumer starts using it.
Updating costs even more
Another problem is the necessity of constantly updating artificial intelligence models. For example, ChatGPT has only trained on data up to 2021 and thus has no knowledge of anything that has happened since. The carbon footprint of creating ChatGPT is not public knowledge, but it is very clear that it is probably higher than that of GPT-3. If it had to be regularly recreated to update its information, energy costs would have increased even more.
Social pressure can be useful to encourage companies and research labs to publish carbon footprints of AI models. In the future, consumers may even use this information to choose a “greener” chatbot.
55438
