ChatGPT is not environment friendly, here’s why?

As the development and integration of artificial intelligence (AI) advance and transform our daily lives, environmental consequences

By groundreportdesk 06 May 2023

New Update

These AI tools are not secure, could lead to cyber attacks: Research

As the development and integration of artificial intelligence (AI) advance and transform our daily lives, there is growing concerned about the potential environmental consequences of these technologies.

According to the research paper, Microsoft used enough water to cool its US data centres during GPT-3 training that could have been used to produce "370 BMW cars or 320 Tesla electric vehicles."

How much does ChatGPT impact the environment?

A recent preprint study published on arXiv suggests that a conversation with an AI chatbot consisting of 20 to 50 questions could consume the equivalent of a 500ml bottle of water.

The research includes a framework for estimating the amount of freshwater needed to power data centre servers and cool servers for AI model training.

For example, the study estimates that Microsoft may have used 700,000 litres (185,000 gallons) of water to train its GPT-3 model, which is equivalent to the water needed to produce 370 BMW cars.

The study calls on companies operating AI models, such as Google's LaMDA, which can use millions of litres of water, to take responsibility for their water footprint in the face of global water scarcity.

Microsoft, which has partnered with OpenAI and invested billions of dollars in building supercomputers for AI training, has revealed that its latest supercomputer contains over 10,000 graphics cards and more than 285,000 processor cores.

This high water consumption, combined with the popularity of chatbots like ChatGPT, could have a detrimental effect on water supplies, particularly in areas already experiencing drought and environmental instability.

Water consumption in ChatGPT

A recent study has revealed that the water usage of the data centres that power the AI chatbot ChatGPT, which serves billions of users around the world, is alarmingly high.

Researchers from the University of California Riverside and the University of Texas Arlington reported that data centres powering ChatGPT consume 500 ml of water for a single conversation of 20 to 50 questions.

The study also reveals that training Microsoft's GPT-3 model requires the same amount of freshwater needed to produce 370 BMW cars or 320 Teslas.

Experts suggest that these impressive abilities have come at a considerable environmental cost, as the chatbot's large-scale water consumption has been overlooked, unlike the carbon footprint that has been studied previously.

Researchers, including scientists at the University of California Riverside in the US, have shed light on this previously unexplored aspect of the environmental impact of AI models.

OpenAI addresses sustainability concerns

In light of concerns about the environmental impact of large language models, OpenAI has emphasized its commitment to finding efficient ways to use computing power.

In response to a Bloomberg query, the company said it takes its role in tackling climate change very seriously and is collaborating with Microsoft to improve efficiency in running large language models on Azure.

However, due to limited transparency and a lack of reporting, the total electricity use and carbon emissions attributed to AI remain unknown.

While some researchers have estimated emissions, data centers are not required to report their carbon footprint, leaving the industry in a black box compared to other industries. Without explicit figures documenting the power used by language models like ChatGPT, it is difficult to assess the full extent of their environmental impact.

ChatGPT's Carbon Footprint

According to an article by the specialist in sustainable data science, Kasper Groes Albin Ludvigsen, published in Medium, to calculate the impact of these platforms, one must consider the amount of electricity they consume and their carbon intensity.

The page developed by OpenAI consumed 1,287 MWh, emitting 552 tons of carbon dioxide (CO2).

There is still not enough data to estimate how much the platform developed by Alphabet pollutes, as a first-time competition, since it was presented just a few days ago. However, it is believed that the trend is the same.

For her part, science journalist Sarah DeWeerdt wrote in an article published in Anthropocene magazine that this chatbot needs an amount of energy similar to the annual consumption of 126 homes in Denmark.

While creating a carbon footprint equal to 700 thousand kilometres of car travel in a single training session of this artificial intelligence.

Researchers at Stanford AI Lab and UC Berkeley, including Ying Sheng and Lianmin Zheng, have developed FlexGen, a rendering engine for running large LLMs with limited GPU memory.

In a recent study titled "High-Performance Generative Inference of Large Language Models with a Single GPU," researchers demonstrated the ability to run Meta AI's OPT-175B LLM on a single 16 GB GPU.

How Do We Reduce the Environmental Impact of AI?

To address the issue of power consumption in machine learning, the academics suggest promoting transparency in the development and operation of such systems.

Frameworks have been developed to help researchers report their energy and carbon use, to encourage accountability and responsible practices in the field.

To compare energy use, online tools are available that encourage researchers to test green areas, monitor energy and carbon use, and assess trade-offs between energy use and performance before implementing energy-intensive models. energy.

People also play a vital role in promoting AI accountability. One approach is to reduce the hype surrounding flashy new AI systems like ChatGPT and recognize the limitations of language models.

By putting their achievements in context and recognizing the trade-offs, we can encourage new avenues of research that are not solely based on developing larger and more complex models. This not only promotes responsible AI practices, but also paves the way for more sustainable AI.