• Big tech and green AI – policies and approaches.
  
• Closed-loop operations make environmental impact hard to measure.
• Can the technology industry alter its course?

There are several problems affecting the technology that prematurely, we call artificial intelligence.

Among the problems besetting AI in this, the tech’s nascent phase, are:

• the use by models of copyrighted material or other assets where licensing terms are ignored at the point of ingestion,
• the generally low quality of results (especially relevant in the context of these pages) produced by LLMs,
• so-called hallucinations that have to be corrected after-the-fact at the inference stage,
• a lack of practical solutions available generally for organisation- or industry-specific uses.

However, the biggest threat to AI’s progression to a stage where at least some of the above are solved or improved is the inconvenient truth that in just a few decades, without clear, widespread decisive action, AI’s human users will be distracted by climate issues. Being blown away by a typhoon, having to live underwater or suffering from a disease contracted by dirty drinking water tends to take people’s minds off whether or not their search results contain inaccuracies.

As companies quietly roll back carbon neutrality deadlines and obfuscate their environmental statements in a race to throw more and more resources at highly power-intensive AI technology, some individuals and organisations in the industry are doing what they can to keep the likely climate catastrophe at the front-of-mind of technologists.

Speaking to Business Insider, Sasha Lucioni, the AI and Climate Lead at Hugging Face referred to a “race to secrecy” by the companies that stand most to lose if progress on artificial intelligence were to slow.

The placing of models and, as important, information about the learning corpora of those models, into proprietary corrals, means that quantifying the ecological impact of all things machine learning is incredibly difficult. Green AI may exist, but we may never know about it. (N.B. ‘Green AI’ can mean the use of AI to solve environmental puzzles – not the intention here.)

While there’s nothing wrong per se with the consumption of electricity by AI models’ training and use, it’s the source of electricity that is having the most impact on the carbon footprint created by the likes of Google, Microsoft and Nvidia’s smart offerings.

“The issue is that vast supercomputers, be it Google Cloud, Azure, or AWS, are located in places that are not powered by renewable energy – mostly natural gas and coal, and that makes a huge difference,” Lucioni said.

It’s easy to understand, therefore, why operators of large machine learning arrays have practices that make measuring carbon emissions from their activities difficult. Of course, those practices also have the effect of protecting intellectual property and ensuring that any advantages big tech may develop in-house stay secret. But releasing key metrics like, for example, the amount of carbon released in the creation of a two-minute, AI-generated video may stop paying customers want to make them.

Software developers are clearly capable of making tools that at least give best-guess estimates of the impact of their code. But big AI companies don’t create versions of similar functions and make the results public. Doing so would run contrary to their desire to grow to to trillion-dollar scale.

There is a market niche for an AI company that can prove it operates in ways that are environmentally friendly, and several have appeared that claim to be just that. The problem is that none of the big names – the big polluters – are among them.

Having a series of bogey-men in the form of OpenAI, Google and Anthropic is analogous to having Chinese coal-fired power plants: there are always bigger culprits for climate disaster and the actions of the individual are a drop in the ocean. But the excuse of pointlessness in any pro-climate action doesn’t necessarily wash in the technology industry.

As Corey Doctorow is fond of pointing out when discussing the enshitification of technology in general: the people who work in the tech industry have the power to change it. It was technologists who built the modern world way back when, and it’s technologists who can change what they build today.

There is also power in the hands of AI users. As consumers of the technology, we could chose to stop using ChatGPT, etc. – just like everyone could chose to walk to work, take public transport or cycle instead of jumping in a car. In technology and transport, some choices are impractical, however.

But as consumers of AI, we can choose providers more carefully so that our individual actions make a difference. And those consumers who also happen to work in technology have the wherewithal to change the norms of the industry, and therefore the viability of a planet with a human population.