Harnessing AI for Corporate Sustainability Requires Precautions

Artificial intelligence (AI) has the potential to play an integral role in corporate sustainability. However, the massive amount of energy needed to run the technology threatens to overshadow potential advantages. Widespread use is expected to cause a 160 percent surge in data center energy demands by the start of the next decade, undoing previous gains in operating efficiency.

A recent survey from the software company Salesforce and the market research firm YouGov reflects this dilemma, with 81 percent of sustainability professionals agreeing that lowering AI’s carbon footprint is important. At the same time, 58 percent think it can be used to have a net positive impact on the climate crisis.

How are sustainability professionals using AI?

“The top use cases are for improving energy efficiency,” Boris Gamazaychikov, the senior manager of emissions reduction at Salesforce, told TriplePundit. “So things like monitoring energy consumption [and] predicting when energy use is high, carbon emissions modeling, and then ensuring compliance with environmental standards and regulations. Those are kind of the top three.” 

Of those using AI in their sustainability programs, 65 percent claim it is transformative. But only 20 percent of the survey’s respondents completely integrated AI into their sustainability programs. Another 29 percent are still testing its use.

Not everyone is convinced that AI will be a boon for sustainability. While approximately half of respondents use AI in their sustainability efforts, the other half do not. Almost 40 percent said they are concerned that it will do more harm than good for their company’s sustainability.

“AI does have this very sort of wide-ranging and potentially extremely beneficial range of uses, but I do not think that we are quite there in terms of the corporate space,” Monica de Bolle, senior fellow at the Peterson Institute for International Economics, told 3p. She expressed concern that — while AI can be used to track energy use and other beneficial monitoring activities — it’s all too common for the tech to be used for inane activities, like drafting emails, that save minimal amounts of time and use excessive energy.

Where should the line be drawn on AI?

AI is also used for forecasting and reporting, Gamazaychikov said. “It's great that we have so many [environmental] regulations,” he added. “It's really important. But now there's so much time spent by sustainability professionals filling in forms, basically, and writing the same narrative in a slightly different way to satisfy a slightly differently worded question. So generative AI is really well suited for that.”

Likewise, AI can look at millions of data points and make forecasts in a fraction of the time it would take a human. Still, it’s important for sustainability professionals to be cognizant of the carbon footprint of their AI use versus how effective it is and how much time they save.

“The kind of energy consumption that we're talking about for different types of AI algorithms is equivalent, or higher than, what cryptocurrencies consume in terms of energy,” de Bolle said. Policymakers and grid planners continue to raise concerns about cryptocurrency mining’s energy use and emissions. It’s likely responsible for 0.6 to 2.3 percent of the United States’ electricity consumption, according to the U.S. Energy Information Administration.

That’s a big deal considering 75 percent of knowledge workers use AI in the workplace. And many of them are providing their own tools, which means the full breadth of environmental harm caused by AI use in the office may be unknown.

“I think using AI to help people write emails is definitely a waste,” de Bolle said. “If you're a corporate leader, and you're looking at the uses of AI within your own company, and you're somebody who's environmentally minded, and you care about the impact that your company's having, then you … should be delineating those activities where AI is going to be used and where AI is not going to be used.”

Instead of using it to do basic tasks like draft emails or simple online research, de Bolle said she would rather see AI reserved for things like scientific projects, medical care or energy monitoring,  where it can make the kind of difference that justifies its emissions and energy use.

The need for transparency and intentional energy efficiency

Gamazaychikov and de Bolle seem to agree that AI has a transparency problem. Despite, or perhaps because of, its eruption in popularity, users are largely unaware of its environmental consequences.

“It's critical that this sector be regulated,” de Bolle said. “These [AI] companies are very opaque in terms of how they operate. And they're very opaque, in particular, with regard to disclosing how much energy is actually consumed … So there needs to be regulation that says, ‘Look, you need to disclose this information.’ Because otherwise, how can companies that are actually using your product … how can they know the costs and benefits?”

Gamazaychikov would like to see a rating system for AI similar to Energy Star, the federal program that administers the familiar blue star label on products and buildings that meet strict energy efficiency standards. That’s something Salesforce is working on with the AI firm Hugging Face, in hopes of bringing transparency to the industry. 

In the meantime, he suggests that companies look at the underlying algorithms when choosing a model to use for sustainability programming. The largest models have trillions of parameters that create the foundation for their decision-making abilities and are dramatically less efficient than smaller language models that use billions of parameters, Gamazaychikov said. The smaller traditional machine learning models are sufficient for the monitoring, predicting and forecasting that corporate sustainability teams do.

“Next it's important to think about the hardware that's actually running these models … There's a lot of efficiency improvements that have been happening with AI hardware, so it's important to be mindful of what's being used to train and then deploy your AI,” Gamazaychikov said. “And then finally, the type of energy that's being used to power these data centers. How much of that is fossil fuel? How much of that is renewable energy? Between somewhere like France, with a very clean grid, and India, with a lot more fossil fuel, you can see a 95 percent carbon emissions difference. [It’s the] same exact workload, but just talking about a different type of energy usage.”

Do the benefits outweigh the harm?

It may sound like a lot for sustainability professionals to consider, but which model they choose will determine the environmental impact of their company’s AI use. After all, the whole point of using the tech in sustainability programming is to enable sustainability — not to solve one efficiency problem by creating another. 

But for AI to have a net positive impact, companies need to limit its use to where it can make a real difference. “I think you have to use AI responsibly,” de Bolle said. “And using AI responsibly means: Is it really helping to get to whatever result you want to get to? Or are you just cutting down three minutes of a person's time by using AI to write an email?” 

Overall, AI appears to have the potential to make a huge difference in corporate sustainability if it’s used responsibly. “We're at a really critical inflection point,” Gamazaychikov said. “AI innovation has been picking up dramatically, so we're at this point where we are seeing a potential future where AI could exacerbate climate change if precautions aren't taken.”