AI environmental cost the ‘elephant in the room’

Artificial intelligence requires a lot of computing power and current data centres are consuming vast resources, especially water, as the impact of AI is being felt in the environmental burden it brings. 

Why AI’s energy problem matters

Sustainability scrutiny hasn’t gone away just because AI is now here. Sam Altman, CEO of OpenAI, has in fact warned about a coming energy crisis that could hamper the technology’s development. Expensive and environmentally costly systems create headaches not just for AI providers but also for users. Businesses ought to be pushing for AI to not just be potent but environmentally efficient, for everyone’s sake.

In context 

Altman’s admission – made during the World Economic Forum conference in Davos – can be seen as part of a habit for preemptively stating uncomfortable truths about the technology for which he is a figurehead. In this way, the former Y Combinator executive gets ahead of the narrative and casts OpenAI (and, of course, himself) as leaderlike and clear-eyed. 

But his comments about the need for a technological “breakthrough” in energy to power AI speaks to the other big issue for the technology: it’s really, really expensive, and is only getting more expensive because of bottlenecks in supplying chips, with the few companies making AI-capable chips counting soaring profits.

This is to say nothing of the energy costs, which were “eye-watering” back at the time of ChatGPT’s debut in the autumn of 2022. In short, AI will bring huge advances to industries like marketing, but not without significant costs. 

What’s going on

New academic research cited in Nature, a scientific journal, indicates that AI is driving big environmental costs, especially when it comes to the amount of water required – what one calls the “elephant in the room”:

• Google and Microsoft’s environmental reports showed increases in water use of 20% and 34%, respectively, as they trained their LLMs. 
• Another paper suggests that the global demand for water from AI systems by 2027 could be half of that used by the entire United Kingdom.

What to do about it

Just regulating the output of AI will not be enough; customers of AI and computing companies need to be aware of and demand lower-energy systems lest it weigh on their own increasingly scrutinised environmental footprint. Transparency is vital, and it is most likely to come about if those holding the purse strings demand it. 

Sourced from Nature, WARC

SPT