The concerns about increasing amount of carbon emissions and global warming due to AI boom haven’t emerged without plausible reasons. Namely, to be functional and respond to users’ queries adequately, every new chatbot (and image generator, too) needs to be trained.
The process of developing and training requires mammoth amounts of electrical energy – more precisely, 1.287 GWh. That amount, which was wasted for training ChatGPT 3.5, equaled 1,287,000 kWh or approximately the power that 20 US homes consume per year, Bloomberg reports.
Bitcoin (BTC), which has been constantly criticized for gargantuan energy consumption, wasting the same amount of power as Argentina, seems to have got a worthy rival.
The AI sector is growing incredibly fast – perhaps even faster that we’re ready for. The potential issue with such rapid growth, in terms of electrical power, is the absence of transparency. No one can pinpoint the exact amount of power and carbon emissions attributed to artificial intelligence.
The emissions may also vary greatly depending on the type of power plants that supply that electricity. For instance, a data center that obtains power from a coal or natural gas-fired plant will produce far more emissions compared to the one that gets its electricity from solar or wind farms.
Some companies have indeed submitted data on their electricity consumption. Also, academics have totaled the emissions from the development of a single model; still, they lack a broad assessment of the total amount of electricity the device consumes.
Sasha Luccioni, a researcher at Hugging Face Inc, an AI company, compiled research in which the carbon footprint of her company’s BLOOM, a competitor of ChatGPT-3 was calculated. On the basis of a small collection of publicly accessible data, she has also attempted to predict the same for OpenAI’s popular ChatGPT system.
We’re talking about ChatGPT and we know nothing about it. It could be three raccoons in a trench coat.
Sasha Luccioni, Artificial Intelligence Researcher
The need for greater transparency
According to Luccioni, researchers need transparency on energy consumption and carbon emissions for AI models. With that in mind, governments and businesses may conclude that employing ChatGPT and other AI models for disease diagnosis or prevention and preserving indigenous languages is worth the electricity and emissions. However, that might not be the case with writing useless scripts or books that will never live up to be published or generating overly basic images.
Why is transparency so important?
In researcher’s view, greater transparency will lead to more examination. Remember the gigantic amount of electricity that Bitcoin mining consumed? Well, the insatiable thirst for power urged China to ban mining and New York to pass a moratorium on new licenses for cryptocurrency mining powered by fossil fuels.
We’ve already mentioned that training ChatGPT consumed 1.287 GWh of electricity. In addition to devouring so much energy, the training produced 520 tons of carbon emissions. The same amount is generated by 110 cars in the USA in one year. Also, it’s worth highlighting that these figures are related to training only one AI model.
What’s an even more sobering fact is that AI model training is not a one-time job. Quite the contrary, AI tools need to be retrained; otherwise, they would miss some contemporary information. This further implies that more energy will be needed and wasted on retraining processes. As a result, more electricity brings about more carbon emissions, which further lead to increased global warming. Sounds like a vicious circle, doesn’t it?
With the models are growing stronger, renowned AI companies are continually working on ways to make them function more effectively. Though three tech giants, Microsoft, Google, and Amazon, have all made commitments to be carbon neutral or negative, none of them bothers to reveal how much carbon emissions their AI models emit.
Google might be among the first AI companies who introduced an approach to carbon emission reduction. The goal is to considerably reduce the carbon footprint produced by machine learning workload.
Currently, Google completely relies on renewable energy to meet its needs for electrical power. The company has promised to decarbonize energy consumption by 2030 by constantly operating on carbon-free energy.
AI as a game changer
While AI may pose a threat to global warming, it can also be a game changer when it comes to greenhouse gas (GHG) emission.
It offers a viable path to accelerate sustainable transformation and decreasing costs in a time of need thanks to its capacity to generate in-depth insights into many facets of a company’s carbon footprint. Large firms are in an especially good position to take advantage of AI capacities owing to their size, which offers them access to enormous data sets—a crucial success element for adopting AI.
According to the study, global GHG emissions amounted to 53 gigatons of CO2e (carbon dioxide equivalent) in 2021. In 2023 and on, it is only projected to rise. To accomplish the objectives set in the 2016 Paris Agreement – limiting the rise in the average global temperature to 1.5°C – it’s necessary to reduce the emissions by 50% by 2030.
But how can AI help?
The biggest strength of artificial intelligence is its capacity for experience-based learning, enormous data collection from its environment, intuitive connections that humans miss, and the recommendation of appropriate actions based on its findings.
Companies can employ AI-powered data engineering to monitor emissions throughout their carbon footprint. They have an opportunity to collect data from various activities, operations, and value chains. Artificial intelligence, on the other hand, can utilize data from novel sources like satellites. It can also produce approximations of missing data and calculate the degree of certainty of the results by adding intelligence to the data.
Furthermore, predictive AI can foresee future emissions across the carbon footprint of a company related to ongoing attempts to reduce emissions, novel carbon reduction approaches, and anticipated demand. As a consequence, they will be able to define, modify, and meet reduction targets accordingly.
By offering deep insight into every step of the value chain, prescriptive AI and optimization can increase efficiency in production, transportation, and other areas, thus lowering costs and reducing carbon emissions.