The effect on the environment has been largely disregarded in the buzz around ChatGPT and the astonishing strength and potential of artificial intelligence (AI).
According to analysts, AI’s carbon footprint may even be greater than that of bitcoin mining, which presently produces more greenhouse emissions than whole nations.
This is the last thing that our vulnerable life support systems need, as seen by the record-breaking temperatures throughout the land, sky, and seas.
Around 2% of the world’s CO2 emissions are currently caused by the IT sector as a whole. According to consultancy company Gartner, the AI sector would use 3.5 percent of the world’s electricity by 2030 if its current growth trajectory is followed.
Fundamentally speaking, Sasha Luccioni, an ethical researcher at the open-source machine learning platform Hugging Face, told The Guardian that if using AI to preserve the earth, one must also take the environmental footprint into account.
“Burning a forest and then using AI to track deforestation doesn’t make sense.”
Open.To provide its chatbot service to more than 100 million customers worldwide, AI spends an estimated US$700,000 daily on computational expenditures alone.
The success of Microsoft-backed ChatGPT has sparked a technology arms race, with Google and Amazon putting money into building their own natural language processing systems.
Many businesses have outlawed ChatGPT use while internally building their own AI.
Artificial intelligence (AI) uses powerful graphics processing units to crunch data, similar to cryptocurrency mining. Huge data centers using tens of thousands of these power-hungry computer chips power ChatGPT.
Calculating the overall environmental impact of ChatGPT and other AI systems is difficult, and many of the data needed to do so isn’t readily available to academics.
According to computer scientist Roy Schwartz of the Hebrew University of Jerusalem, “Obviously these companies don’t like to disclose what model they are using and how much carbon it emits.” he told Bloomberg.
Additionally, it is difficult to forecast how much AI will advance or how energy-efficient it will become in the next years.
The training of GPT-3, the forerunner of ChatGPT, on a database of more than 500 billion words is thought to have used 1,287 kilowatt hours of electricity and 10,000 computer chips, according to researchers.
In the US, the same quantity of energy could supply 121 families for a year.
A 33-time flight from Australia to the UK would have produced about 550 tonnes of carbon dioxide throughout this training session.
The July release of GPT-4, which was trained on 570 times as many characteristics as GPT-3, suggests that it may consume more energy than its forerunners.
When trained on 1.6 terabytes of data, the BLOOM language model was shown to use 433 kilowatt hours of electricity.
If the expansion of the AI industry is anything like the rise of cryptocurrencies, it will only increase in energy consumption over time.
Bitcoin uses 66 times more energy today than it did in 2015; because of this, cryptocurrency mining has been outlawed in China and New York.
Cryptocurrency mining requires complex computations from computers, and it can take a month to earn just one bitcoin.
The annual electricity consumption for bitcoin mining is 137 million megawatt hours, and the carbon impact is almost as big as New Zealand.
Innovation and preserving the planet’s finite resources must be carefully balanced.