Double-edged
Using generative AI to help enterprises keep tabs on their greenhouse gas emissions, as Salesforce plans to do, can be a double-edged sword, as building and tuning the large language models (LLMs) they run on is energy intensive, and not all data centers use clean energy. In other words, using generative AI can increase greenhouse gas emissions.
One way to avoid that dilemma, Norman said, is to avoid using AI for its own sake, ensuring that we only use AI models that are fit for purpose. “That’s the number one thing we can all be doing as an industry,” she said.
IDC’s Cravens welcomed Salesforce’s focus on fitness for purpose.
A useful question to ask, she said, is, can the use of AI deliver results that human analysis would not generate? “I think in the instance of ESG and sustainability, there are use cases that it absolutely makes sense,” she said.
Norman added it’s not just about clicking (or not) on that generative AI button, though.
“There are a lot of other exciting levers we can pull” to reduce greenhouse gas emissions from AI operations, she said.
Those include using cleaner energy — Salesforce sources 100% clean energy for its global operations, she said — and more efficient hardware in data centers. Nvidia, in particular, has been hammering home the message that GPUs are more efficient for training LLMs.
But amid all this technology, CIOs in search of better energy use should never lose sight of the human factor.
“Developers can produce more efficient code,” said Norman, whether that’s the base code running an application or model, or “even UI designers, who are creating a better UI that’s more efficient and gives a better experience to the end user.”
