3 ways to manage AI’s environmental impact

More than 90 percent of companies are stepping up their use of artificial intelligence, but just 39 percent of chief information officers are confident their organization is managing AI’s environmental impact. That’s according to a survey of 1,000 CIOs conducted by IT service provider Logicalis.

Roughly the same number of respondents are confident that energy efficiency is prioritized as their company embraces AI services and infrastructure, leaving plenty of room for improvement.

Indeed, almost three-quarters of CIOs are concerned about “unchecked” AI adoption, the survey found.

Other takeaways from the survey: Sustainability leaders can support their company’s technology strategists by suggesting ways to reduce electricity consumption; helping them revise purchasing proposals and contracts to request emissions data; and educating teams across their company about AI’s potential environmental side effects on water and energy consumption.

“Keep the conversation alive, make it proactive,” said Logicalis Global CEO Bob Bailkoski. “I’ve never met a CIO professional that isn’t open to that conversation.”

Here are three ways to manage AI’s environmental impact:

Request emissions data as part of contracts and bids

Sustainability teams should apply the same supplier responsibility standards they use with other vendors or partners to those providing their AI software, said Marshall Chase, director of sustainability for Micron Technology. 

“We often have imperfect information, but there’s enough ‘good enough’ information out there to take a stab at estimating our AI usage impacts,” he said. “My take is that unless you’re a [software as a service] company, the environmental impact of AI in your supply chain — while growing — is probably going to be pretty darn small.”

A growing number of Logicalis’ clients are seeking emissions data about their AI-related technology, which it calculates with an algorithm that converts electricity usage into a carbon footprint. Typically that information isn’t shared widely outside of sustainability teams, but they can play a role in establishing a benchmark and interpreting trends for the departments using AI, Bailkoski said.

“The AI future is going to come,” he said. “We have an obligation now to try to minimize the impact of this.”

Encourage AI vendors to set reduction targets

Sustainability teams can also work alongside their corporate procurement colleagues to help them evaluate whether potential AI vendors appear willing to manage and reduce their electricity and water consumption and other ESG impacts — now and in the future. 

That could include requiring prospective suppliers or partners to disclose such information as part of the bidding process and including reduction targets in the terms of long-term contracts, said John Mennel, U.S. sustainability and cleantech leader at consulting firm Deloitte.

For strategic vendors, that might mean introducing a scenario in which helping your organization meet emissions reduction commitments could create incentives for better contract terms over time.

For second-tier partners, it could mean collecting this information with an eye to the future. “You need to be able to pick battles and tell procurement how they should make tradeoffs,” Mennel said.

Nudge employees toward options that use less energy and water

Sustainability professionals can also win support from their counterparts in technology by demonstrating how measures that reduce AI-related energy and water consumption can translate into cost savings, said Boris Gamazaychikov, former AI sustainability manager at Salesforce. 

For example, decommissioning underutilized computer servers and consolidating applications and digital workloads onto hardware that is more energy efficient is one effective strategy. Paring down the volume of data that your company stores is also essential, because it reduces the amount of energy required to train algorithms and surface answers.

“Explain that the sustainability of AI is not just an environmental concern but that there are other risks that you have in common, including costs,” he said. “It’s about finding common ground and helping them de-risk rather than adding more to their plate.” 

Sustainability teams can bring important perspective about which AI services are the best options for certain sorts of applications, said Ryan Bogner, digital sustainability leader at EY. 

The amount of energy that AI uses depends on where the “tokens” — the words and characters used to generate an answer or train an algorithm — are processed. 

Companies are unlikely to choose a particular AI application based on its energy consumption, but they may be able to control which cloud computing service or data center delivers it. “If you are able to distribute where the work happens, you can have an impact,” Bogner said.

This is something that companies can control in the background, by automating where certain types of AI queries are directed based on energy consumption, which usually translates into cost savings, said Deloitte’s Mennel. This can be invisible to employees. “[Chief information officers] don’t want to tell you not to use the tools, but they are managing on the backend to keep the footprint low.”