Three experts enter into debate on a pressing issue. This time:
Text Kim Bakker
© Sam Rentmeester
Are you concerned by the environmental impact of AI?
Sallou “Some studies estimate that by 2030, IT will account for 20% of our total energy consumption, with AI playing a leading role, as its integration into technology keeps picking up pace. We have to ask ourselves carefully whether AI is always needed.”
Verdonk “Absolutely. It’s paramount that we develop responsible AI practices, and I’m heartened by the solutions already emerging from the field. While the models got exponentially larger at first, developers are now looking for ways to shrink the models without compromising on quality.”
Van der Klauw “I’ve always been a tech optimist, but we have to think carefully about where we go from here. AI systems so far have been highly centralised, while a more distribution local approach could be more robust, safer and more energy-efficient.”
Is there enough awareness of the environmental impact of AI?
Verdonk “I’m aware that I live in a bubble that is probably more concerned than the general population. I think only a handful of people realise that every whimsical poem generated by ChatGPT costs lots of energy. We take a lot of minor AI tools for granted without thinking about the large data centres churning in the background.”
Sallou “I think we do realise it has a significant impact, without truly knowing how big it really is. I’m an expert in Green AI, and even I struggle to see the whole picture because there’s so much we’re yet to discover.”
Van der Klauw “I believe that most people are aware that things will have to change, although I do wonder whether it’ll work out in our best interest.”
Couldn’t AI also help combat climate change?
“AI is part of both the problem and the solution. Climate change is a vast and highly complex phenomenon that can only be solved and understood with the help of AI.”
“Absolutely. We desperately need AI to solve problems. Without it, we’d be fighting a losing battle in the energy transition.”
“The opportunities are virtually infinite. We could harness AI to make our energy grids more dynamic, for instance, which could play a big role in solving the climate problem.”
In many ways, AI makes our lives easier and more efficient. To what extent do its benefits outweigh its massive carbon footprint?
“I’m afraid there’s no one-size-fits-all answer to that question. Every time we design or use AI, we need to ask ourselves why. It’s important that the bottom line is positive. If AI adds a lot of sustainable social, economic or technical value, it could be worth the climate impact.”
“We shouldn’t see AI as an environmental threat. The potential contribution it can make to the energy transition far exceeds its costs.
That said, you do have to factor in the climate impact. Some AI draws such vast amounts of power that we have to be judicious in what we spend our scarce energy on. Are TikTok video’s and Instagram pictures worth building enormous data centres for?”
“In my opinion, the advantages far outweigh the disadvantages. AI opens up new directions for us to solve difficult problems. The small-scale developments emerging today make me hopeful for the future: I think AI will be embraced by society.”
What responsibility do developers like OpenAI and Microsoft bear?
“Developers certainly have a responsibility, as they’re in a prime position to drive change. Pawning off all responsibility on developers, however, is a little too facile. As far as I am concerned, it’s a shared responsibility of all parties involved, including consumers. They’re the ones buying the products and services: without demand, there would be no supply. Still, people can only make an informed choice if the provider is transparent.”
“We’re in our rights to expect big corporates to operate responsibly, but we shouldn’t forget that they need to turn a profit. What’s more, we need to avoid greenwashing and having big tech chart the course.
It is important for the government to increase awareness of the side effects of AI, including climate impact. I’m not saying governments should regulate everything, but they should take a leading role.”
“The providers of these kinds of services have to be held accountable and audited. Fortunately, there’s already an incentive for these companies to become more sustainable, as increasing grid congestion means they’re already running up against their limits. As a result, they’ve already had to turn down opportunities. If companies succeed in making their operations more energy-efficient, the impact of each individual operation would decrease and they’d be able to serve more customers.”
To what extent can the users of AI programmes like ChatGPT be held responsible for the climate impact of generative AI?
“I think people would change their behaviour if they knew the impact of every ChatGPT query. Maybe not across the board, but the average user would. It will get trickier as AI becomes more intertwined in our lives, but even then I think we’ll be able to shift down a gear if necessary: People are excellent at adapting.”
“Consumers have a bad habit of choosing the cheapest, tastiest and most convenient options. On the other hand, more and more people are deliberately deciding to go vegetarian, for example, which only goes to show that social norms can push people to make more responsible choices. Financial incentives would also work: you could adopt a dynamic pricing model for AI use, offering cheaper rates when green energy floods the grid.”
“Users have a responsibility to be informed. Some airlines now report the carbon emissions of every flight, and you could roll out something similar for AI. Many users may just close the pop-up and go on with their day, but at least they’ll know the impact of their choices.”
‘Consumers have a bad habit of choosing the cheapest, tastiest and most convenient options’
Are we right to put the environmental impact of AI, specifically, under a magnifying glass? Would it be more appropriate to look at tech in general?
“Yes and no. By definition, AI has high energy consumption compared to other technological developments. At the same time, AI will ratchet up the energy consumption of tech in general as it becomes more intertwined. Besides, developments are moving so fast that we need to regulate it before it gets out of hand. AI is a potent accelerator.”
“I think focussing on AI is justified, but we shouldn’t lose sight of the bigger tech picture. AI is not an isolated system and it makes little sense to carve out the energy consumption of AI alone, as it replaces energy that would otherwise be consumed elsewhere. It’s part of a bigger entity.”
“The impact of AI on climate and other parts of our lives is so great that we’re right to highlight it. The Scientific Council for Government Policy compared AI to the invention of the steam engine in terms of the transformative potential. Facing such an exponential development, it pays to think ahead. While AI is starting from humble beginnings, it will soon snowball to the point that we can no longer ponder important issues such as climate and safety carefully, which has to be avoided at all costs.”
‘Thankfully, building the biggest, toughest model out there won’t give you bragging rights anymore’
Should we halt AI developments to prevent further climate damage?
“That would be very difficult, because society depends on innovation. Restricting completely new AI developments in favour of focussing on what we already have would, however, be a good idea, because there are major efficiency gains to be had.”
“That would be a bad idea. The advantages of AI far outweigh the disadvantages. That being said, AI has to be used judiciously. In China, you can’t walk down the street without being recognised by cameras, which is not something we want to happen.
At the same time, AI has the potential to expedite the energy transition, which does merit further investigation. Actively distinguishing desirable from undesirable developments will let us unlock the full potential of AI, preventing a lot of environmental impact and more.”
“We need to push forward. Climate impact should remain a topic of conversation and it should be easy for people to find transparent information about it. My hope is that AI developments will prioritise efficiency, resulting in smaller models with the same kind of outcomes. Thankfully, building the biggest, toughest model out there won’t give you bragging rights anymore.”