Salesforce will begin lobbying for new regulations requiring companies to disclose emissions data and efficiency standards for artificial intelligence.
The company announced the move Monday as part of its new “Sustainable AI Policy Priorities.” It has previously published positions on ethics and equity related to AI, as have other big tech companies including Amazon, Google and Microsoft.
Salesforce’s new policy comes amid growing concern over the amount of electricity required to train and run AI algorithms. Data centers already account for 2-3 percent of annual global power consumption. Demand could triple by 2030, accelerated by AI, according to Boston Consulting Group. That possibility has inspired Amazon, Google and Microsoft to look beyond solar and wind to other non-fossil fuel power sources, including nuclear and geothermal.
Salesforce: 6 ways policy can support ‘sustainable AI’
Like many software companies, Salesforce is vocal about ways AI could improve corporate sustainability such as reducing energy and water usage, monitoring biodiversity and planetary conditions, and speeding data collection for environmental, social and governance policy reporting. Its addition of specific concerns for AI is unique, but not surprising, given that it added “climate” to its public policy priorities in 2021.
“The models that power AI require enormous compute resources to function, resulting in environmental impacts like increased energy demand, carbon emissions, water depletion and resource extraction,” Salesforce said in its brief. “Being mindful of AI’s environmental impact is crucial for ensuring that the technology develops in a way that supports a sustainable future.”
Salesforce identified six priorities for new regulations and incentive programs, in two categories, summarized below:
Measuring and managing environmental impacts
- Require AI emissions disclosure for “all providers of general-purpose AI models.” Salesforce wants standard metrics for reporting on energy efficiency and carbon footprints, as well as impacts such as the amount of water used by the data centers.
- Add environmental impact as a risk factor. The amount of computing power AI training requires should be part of broader risk assessments, the company believes.
- Establish efficiency standards for “high-risk” AI systems. These could include requiring certain hardware or making sure systems are powered by low-carbon grids or clean energy.
Accelerating climate innovation with AI
- Enhance climate modeling, data analysis and predictions. Salesforce wants more open-source environmental data and forecast databases to be free for use by AI applications.
- Encourage public-private partnerships. The company is urging collaboration between governments, business and academia, so that communities can access AI-enabled services regardless of their resources.
- Spur AI-driven innovation. Salesforce would like to see grants and tax incentives that spur environmental applications.
What Salesforce already does
Salesforce began disclosing energy and environmental metrics related to its AI development activities in December. It’s working strictly on AI models that pertain to its core software applications, which “results in high-performing models that are smaller and more sustainable, cost-effective and easier to fine-tune,” said Boris Gamazaychikov, senior manager of emissions reductions at Salesforce, in that disclosure.
The company is using the most energy-efficient hardware available from Google’s cloud services unit to reduce AI emissions and is running these activities in data centers powered by lower-carbon sources, Gamazaychikov said.
Salesforce — like Amazon, Google and Microsoft — is funding organizations and startups developing AI applications that could have climate benefits. On Monday, it announced support for five more nonprofits: Climate Collective Foundation, Good360, Groundswell, Ocean Risk and Resilience Action Alliance, and WattTime.