Is generative AI bad for the environment? A computer expert describes ChatGPT’s and its relatives’ carbon footprints

Chatbots and picture generators are powered by cutting-edge generative AI technology. How hot, though, is it making the planet?

As a researcher studying artificial intelligence, I frequently worry about the energy requirements for creating AI models. More energy is required by an AI that is more potent. What will happen to society’s carbon footprint in the future when more potent generative AI models emerge?

A machine learning algorithm is said to be “generative AI” if it can create complicated data. As an alternative, “discriminative” AI is used, which generates just one number by selecting one out of a limited number of choices. Accepting or rejecting a loan application is an illustration of a discriminatory output.

Sentences, paragraphs, images, and even brief videos are just a few of the more sophisticated outputs that generative AI may produce. It has long been used to propose a search term in autocomplete or to produce voice answers in smart speaker apps. But only lately did it learn how to produce human-like words and lifelike images.

Also Read: Learn How To Unpin My AI On Snapchat Without Snapchat Plus

Using more power than ever

It is challenging to determine the precise energy cost of a single AI model, which includes the energy necessary to produce the computer hardware, develop the model, and use the model in production. Researchers discovered in 2019 that building the 110 million parameters generative AI model BERT used as much energy as one person’s round-trip transcontinental journey. The size of the model is determined by the number of parameters, with larger models often being more capable.  

The far bigger GPT-3, which contains 175 billion characteristics, was estimated to have used 1,287-megawatt hours of electricity and produced 552 tonnes of carbon dioxide equivalent, which is the same as 123 gasoline-powered passenger vehicles driven for a year. Before any users begin utilizing the model, that is merely for making it ready for launch.

Carbon emissions can be predicted by factors other than size. Similar in scale to GPT-3, the open-access BLOOM model created by the BigScience project in France has a far smaller carbon footprint, using just 433 MWh of power to produce 30 tonnes of CO2eq. According to Google research, utilizing a more energy-efficient model architecture, CPU, and data center may cut the carbon footprint by 100 to 1,000 times for a given size.

Larger types do deploy with a greater energy need. The carbon footprint of a single generative AI query is not well understood, although some industry insiders believe it may be four to five times more than a search engine query. The volume of questions they get each day may increase dramatically if chatbots and picture generators gain popularity and Google and Microsoft put generative AI language models into their search engines.

Also Read: Know Top 6 AI Courses You Need To Take Today

AI bots for search

A few years ago, few individuals used models like BERT or GPT outside of research facilities. On November 30, 2022, OpenAI published ChatGPT, which corrected this. The most recent statistics available indicates that ChatGPT received over 1.5 billion visitors in March 2023. On May 4, 2023, Microsoft included ChatGPT to its Bing search engine and made it accessible to everyone. The price of implementing AIs might really build up if chatbots become as popular as search engines. However, AI assistants may be used for much more than just search, including document authoring, arithmetic problem solving, and marketing campaign creation.

Another issue is the ongoing requirement for AI model updates. For instance, ChatGPT was only trained on data up to 2021, therefore it is unaware of anything that occurred beyond that date. Although the carbon footprint of developing ChatGPT is unknown, it is most certainly significantly more than that of GPT-3. The energy expenditures would increase if it had to be replicated frequently to update its understanding.

The advantage of asking a chatbot for information over utilising a search engine is that it may be more straightforward. Given that accuracy concerns are addressed, you receive a direct response similar to what you would receive from a human rather than a page full of links. The faster access to the information may make up for the higher energy use compared to a search engine.

Also Read: Benefits Of Using Digital Marketing For Your Business

Ways forward

It is difficult to foresee the future, but huge generative AI models are here to stay, and knowledge will likely be sought after by more and more individuals. A student could, for instance, seek a tutor or a friend for assistance or reference a textbook if they need aid right away with a maths difficulty. Probably a chatbot will answer their question in the future. It also applies to other types of expert knowledge, such as legal counsel or medical information.

The environment won’t be harmed by a single massive AI model, but if a thousand businesses create slightly different AI bots for various uses that are all utilised by millions of consumers, the energy usage may. To make generative AI more effective, further research is required. The good news is that AI can function on clean energy. When compared to utilising a grid dominated by fossil fuels, emissions can be lowered by a factor of 30 to 40 by moving computing to locations with access to more renewable energy or scheduling computation for times of day when it is more readily accessible.

Lastly, public pressure might be useful to persuade businesses and academic institutions to follow some others’ lead and reveal the carbon footprints of their AI models. In the future, customers may even select a “greener” chatbot using this knowledge.


How does generative AI affect the environment?

The environmental effect of generative AI models, such as huge language models, is a critical issue because of how much water and carbon they use. This results in considerable worldwide resource depletion and emissions.

What is the carbon footprint of generative AI?

This has been supported by a number of studies looking at the computational power and carbon emissions produced by AI in recent years. The University of Massachusetts, Amherst researchers discovered that the training of many large AI models may produce more than 626,000 pounds of carbon dioxide.

Also Read: India’S Top Cloud And Data Security Summit Is Almost Here

How useful was this post?

Click on a star to rate it!

Average rating 0 / 5. Vote count: 0

No votes so far! Be the first to rate this post.

Riya Kapoor

Riya Kapoor writes about lifestyle, entertainment, news and gadgets. She has been in this industry for almost 4 years now. She is a graduate from Delhi University with English Hons and had deep connection with writing since her childhood.

Related Articles

Back to top button