University of California professors Don Patterson and Bill Tomlinson, along with MIT Sloan School of Management scientist Andrew Torrance, conducted a study on the environmental impact of AI systems. Their paper, published on arXiv, highlights the lower carbon dioxide emissions generated by generative AI tools like ChatGPT and Midjourney compared to humans performing the same tasks. The study has sparked discussions among scientists, both supporting and refuting the findings. Given the widespread implementation of AI technology in various aspects of our lives, it is crucial to understand its environmental implications for the future. This article delves into the methodology used by scientists from MIT and the University of California to measure ChatGPT’s energy consumption and explores the reasons why some doubt these findings.
To begin, VentureBeat reported on the recent ChatGPT energy study conducted by these scientists. Although the paper was originally published in March, it is currently undergoing peer review. The study reveals that an AI system like ChatGPT emits 130 to 1500 times less carbon dioxide equivalent (CO2e) than humans. Similarly, image generators like DALL-E 2 and Midjourney produce 310 to 2900 times less CO2e. The researchers analyzed existing data on the environmental impact of AI programs, human activities, and text and image production. They relied on information from various studies and databases to estimate the effects of AI and human activities on the environment.
One interesting data point from their analysis is the online ChatGPT estimate based on the traffic of 10 million queries, which generates approximately 3.82 metric tons of CO2e daily. Additionally, the training process of ChatGPT contributes to a carbon footprint of 552 metric tons of CO2e. To provide some context, the researchers compared these figures to the annual carbon footprints of the average US individual (15 metric tons) and India (1.9 metric tons) to gauge the per-capita emissions over the estimated time needed to generate a text page or image.
The scientists emphasize the importance of measuring carbon emissions from AI activities to inform policymaking on sustainability issues. Professor Don Patterson explains, “Without an analysis like this, we can’t make any reasonable kinds of policy decisions about how to guide or govern the future of AI.” He stresses the need for grounded information to move forward.
However, the study does have some flaws, as acknowledged by scientist Andrew Torrance. He points out that complex systems, such as climate, society, and AI, introduce unpredictability into the results. This recognition highlights the inherent challenges of studying AI and its environmental impact.
Some experts in the AI community, like Meta’s chief AI scientist Yann LeCun, have used the research to support their claims. However, this has caused skepticism among researchers like HuggingFace AI member Sasha Luccioni. Luccioni argues that it is unreasonable to attribute an individual’s total carbon footprint estimate to their profession. She also questions the comparison between human footprints and AI energy footprints in terms of life cycle assessment and overall energy consumption.
One of the main reasons for doubting ChatGPT’s energy consumption figures is the lack of transparency from OpenAI and other AI companies. AI researcher Verónica Bolón criticizes OpenAI for not sharing exact information about energy consumption, claiming that it must be substantial due to the large amounts of data and neural networks required.
Additional studies have also raised concerns about ChatGPT’s water consumption. Inquirer USA references a study from Johnson Controls suggesting that training the GPT-3 large language model consumed 700,000 liters of water, equivalent to the water needed to manufacture 370 BMW automobiles. Moreover, an arXiv study mentioned by the website states that ChatGPT consumes a “500ml bottle of water” when answering 20 to 50 questions.
In conclusion, a recent study indicates that ChatGPT’s energy consumption is lower than that of human activity. However, skeptics have pointed out various factors that cast doubt on the findings, including a lack of transparency from AI companies like OpenAI and the complex nature of AI systems. Despite these uncertainties, the research conducted by MIT and the University of California contributes to our understanding of how AI can benefit humanity while minimizing harm to the environment. To stay updated on the latest digital trends, visit Inquirer Tech.
Denial of responsibility! Vigour Times is an automatic aggregator of Global media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, and all materials to their authors. For any complaint, please reach us at – [email protected]. We will take necessary action within 24 hours.
