โAs an Artificialย Intelligenceย (AI) researcher, I often worry about the energy costs of building artificial intelligence models,โ Kate Saenko, associate professor of computer science at Boston University, wrote in an post at The Conversation. โThe more powerful the Artificialย Intelligenceย (AI), the more energy it takes.โ
Althoughย while the energy consumption of blockchainsย teck like Bitcoinย (BTC) and Ethereumย (ETH) has been studied and debated from Twitter to the halls of Congress, the effect of the rapid development of Artificialย Intelligenceย (AI) on the planet has not isย still received the same spotlight.
Professor Saenko intendsย to change that, but acknowledged in the post that thereย is limited data on the carbon footprint of a single generative Artificialย Intelligenceย (AI) query. Nonetheless, she stated that research puts the number four to 5 times higher than a simple search engine query.
Reportsย by a 2019 report, Saenko stated a generative Artificialย Intelligenceย (AI) model wasย known the Bidirectional Encoder Representations from Transformers (or BERT)โwith 110 Million parametersโconsumed the energy of a round-trip transcontinental flight for one individual using graphics processing units (or GPUs) to train the model.
In Artificialย Intelligenceย (AI) models, parameters are variables learned from data that guide the modelโs predictions. More parameters in the mix often means greater model complexity, requiring more data and computing power inย doingย so. Parameters are adjusted during training to minimize errors.
Saenko noted in comparison that OpenAIโs GPT-3 modelโwith 175 Billion parametersโconsumed an equivalent amount of energy as 123 gasoline-powered passenger vehicles driven for one year, or around 1,287-megawatt hours of electricity. It likewise generated 552 tons of carbon dioxide. She alsoย mentionedย that the number comes from just getting the model ready to launch before any consumers started using it.
โIf chatbots become as trending as search engines, the energy costs of deploying the AIs could really add up,โ Saenko stated, citing Microsoftโs addition of ChatGPT to its Bing web browser earlier this month.
Not helping matters isย that increasingly Artificialย Intelligenceย (AI) chatbots, like Perplexity Artificialย Intelligenceย (AI) and OpenAIโs wildly trending ChatGPT, are releasing mobile applications. That makes them even easier to use and exposes them to a much broader audience.
Saenko highlighted a study by Google that found that using a more efficient model architecture and processor and a greener data center can considerably reduce the carbon footprint.
โ Althoughย while a single large Artificialย Intelligenceย (AI) model is not going to ruin the environment,โ Saenko wrote, โif a thousand corporations develop slightly different Artificialย Intelligenceย (AI) bots for different objectives, each used by millions of customers, then the energy use could become an issue.โ
Inย theย end, Saenko concluded that more research is required to make generative Artificialย Intelligenceย (AI) more efficientโbut sheโs optimistic.
โThe good news isย theย factย that Artificialย Intelligenceย (AI) can run on renewable energy,โ she wrote. โBy bringing the computation to where green energy is more abundant, or scheduling computation for times of day when energyย fromย renewableย sources is more available, emissions can be reduced by a factor of 30 to 40 compared to using a grid dominated by fossil fuels.โ
Interested in learning more about AI? Check out our latest Decrypt U course, โGetting Started with AI.โ It covers everything from the history of Artificialย Intelligenceย (AI) to machine learning, ChatGPT, and ChainGPT. Find out moreย here.
