Is AI an eco-savior or an eco-disaster?

If one ChatGPT query consumes the same energy as toasting a slice of bread, should we be more aware of our AI use??

When blockchain technology went mainstream, critics pointed to its country-sized energy use. Now it’s AI’s turn in the climate spotlight, and although I’m a huge advocate of AI, some recent conversations I’ve had have made me think more carefully about the environmental cost.

First, the argument for AI, regardless of energy use. AI is already having a positive impact in many areas of life. It’s revolutionising medicine, from diagnostic precision to drug development. This week, it has even helped unlock the mysteries of a 2,000-year-old carbonised scroll from Vesuvius.?

I’ve written much about how I think it will change the way engineering teams develop software, and it may soon touch every team in every new business. I spoke to someone recently who had seen a demo of Adobe’s Firefly, a generative AI tool that can effectively create a new business in minutes – a website, commerce backend, and marketing materials.?

Every new application I read about confirms that AI is for me the most exciting area in tech. But what do we really know about the costs????

A new study this month suggested AI could be a major contributor to global carbon emissions in the future. Data centres managed by Google, Amazon and Microsoft are already consuming a lot of energy training large language models. OpenAI is spending vast sums operating ChatGPT, and I saw one claim that hundreds of millions of daily chatGPT queries could consume the daily equivalent energy usage of 33,000 U.S. households. Even at 10 million daily queries, chatGPT would be a significant energy consumer.?

There are many more models being developed beyond OpenAI and Google’s Bard, and with enterprise use of other AI technology increasing rapidly, how worried should we be?

Essential use only?

I’ve seen some people call for AI to be used sparingly due to its environmental cost. I’m not convinced, for some of the reasons I’m about to mention. Then again, maybe I should speak to more experts, given the accuracy of some of my other assumptions around tech and sustainability.?

Anne Currie is co-founder of Strategically Green and one of the leading voices in green tech. She was a guest on my podcast recently and set me straight on a misconception I had about data and energy consumption. I’d always felt that much of our data – emails, photos, WhatsApp messages – was being stored unnecessarily and that we should get in the habit of deleting files we no longer need, to help data centers reduce their energy consumption.?

According to Anne, the opposite is true. Dormant data uses no energy at all, but the moment you delete a file, you’re effectively bringing it back to life. That requires processing power and therefore energy.??

So if the world had listened to me and deleted its redundant data, it would have triggered an ecological disaster!?

I may not be sure about the personal use of AI, but surely companies can do their bit to minimise the environmental cost of AI while maximising its impact. For most companies and organisations – a marketing agency or a local council - technology won’t be a vast consumer of energy, especially now cloud technology is so commonplace.?

It is a big issue for big tech though, and I do give them credit for their record on sustainability. In the past, there was huge wastage as people bought more storage and processing power than they needed, and server farms ran at half capacity. Thanks to various innovations, tech firms have enabled us to use only what we need, when we need it.?

Firms that supply our AI infrastructure are also on the hook. Nvidia is responsible for over 90% of the hardware that powers AI, and it says it’s aware of the need to reduce energy consumption by continuously optimising the efficiency of its chips and servers. I know that OpenAI is also conscious of the sustainability challenges related to AI, and has said in the past that its infrastructure runs on Azure, and Azure will be running on 100% renewable energy by 2025. Its CEO is a clean energy advocate and an investor in nuclear fusion.

Data centres are a big issue, and another solution may come from the technology sector. Simon Phillips is the CTO at Oxford Quantum Circuits (OQC) , and he told me recently how his firm is applying quantum computing power to make data processing magnitudes more efficient in massive data centres.?

“Quantum computers can do the same tasks as classical computers, but for a lot less power consumption. You can have a few aisles of high-performance computers, doing some optimization routine for hundreds of kilowatts of power, if not megawatts. You can replace all that with a quantum computer that uses 10 kilowatts of power.”

For all the valid concerns about energy efficiency, as an investor and advisor to tech firms, I still want companies to double down on their investment in AI. It should be top of their agenda, and there might be ways they can do that while being aligned with their sustainability strategy at the same time.?

Smaller AI models like Meta’s Llama 2 are now emerging that can perform most tasks with much less processing power. As one AI CEO said recently, you don’t need a Ferrari to drive to work, and a greater choice of AI tools in the marketplace can only be a good thing. Perhaps an eco angle could be an interesting niche for smaller AI firms to target.?

The natural technology cycle in the marketplace may also create its own solution. Until now, the market has been flooded with free AI applications. Everyone from development and design teams to small business owners and students has been playing with the tools, but they may not be free forever. At some point, we will see a commoditization of AI that will naturally dampen demand and strip away some of the less essential use. That friction should accelerate the commercialisation of AI, and that return on investment should accelerate the efficiency of products too.

As for me and ChatGPT, it may or may not be true that two queries have the same carbon footprint as my toaster, so perhaps I will start writing my own best-man speeches in the future.?

What’s your take on the balance between tech innovation and sustainability? Should we use AI more sparingly until we reach a point where it’s carbon neutral? I’d love to hear your opinions in the comments below.