Build a Sustainable AI Strategy

Learning Objectives

After completing this unit, you’ll be able to:

• Articulate the value of creating a sustainable AI strategy.
  
• Explain the Salesforce strategy for sustainable AI.
  
• Implement strategies for sustainable AI at your organization.
  

Why Do You Need a Sustainable AI Strategy?

While AI has the potential to be a powerful tool to help address climate change, the rise of generative AI also presents environmental risks. However, 58% of sustainability professionals believe the benefits of AI will outweigh its risks when solving the climate crisis.

The large language models (LLMs) that power generative AI require large compute resources to function, which can result in negative environmental impacts such as carbon emissions and water depletion.

As we’re experiencing record temperatures and the other effects of climate change across the globe, sustainable development has never been more important.

Every business developing and deploying AI needs a strategy to implement ethical and sustainable technology from the start. Let’s explore how to do this.

Strategies for Sustainable AI Development

Sustainability is a guiding principle for the development and deployment of AI here at Salesforce. In partnership with the Salesforce AI Research, Sustainability, and Office of Ethical and Humane Use teams, Salesforce developed a blueprint for sustainable AI focused on three main components: choosing right-sized models, utilizing efficient hardware, and prioritizing low-carbon data centers.

Check out our learnings and these three tips to help shape your sustainable AI strategy.

1. Choose right-sized models.
   

• While there has been a rise in popularity of general-purpose LLMs, bigger models aren’t always better. Smaller models built for specific use cases require less data and compute power compared to general-purpose large language models.
  
• In addition to having smaller environmental footprints, smaller models are more affordable to operate, easier to train, and often outperform large language models.
  
• Salesforce uses domain-specific models—small language models trained on particular data sets designed for a specific purpose. For example, Salesforce unveiled an AI model that “punches well above its weight class” as described in the VentureBeat article, “Salesforce proves less is more: xLAM-1B ‘Tiny Giant’ beats bigger AI models.” The company’s new aforementioned AI model, Tiny Giant, outperforms models up to 7x its size.
  

2. Utilize efficient hardware.
   

• The energy efficiency of an AI model depends on the hardware it’s trained and deployed on. AI hardware manufacturers are working to deploy new versions of their hardware that prioritizes both better AI performance and higher energy efficiency. For example, Google’s AI hardware, the Cloud Tensor Processor Unit (TPU), has become more efficient with each new generation.
  
• Initial Salesforce tests found that the new TPU v5p outperforms the previous generation by a 2x efficiency increase (measured by averaging the model parameters per training energy consumed).
  

Source: Salesforce Research

3. Prioritize low-carbon data centers.
   

• Data centers house the hardware, servers, and other infrastructure needed to train and deploy AI models. Data center emissions vary widely from region to region because they rely on local electric grids for power.
  
• The more a grid uses fossil fuels, the higher the local data center’s emissions will be (see graphic below). This makes it crucial to know where the data centers are that are training and deploying AI.
  
• To help reduce AI model emissions, Salesforce intentionally trained models in low-carbon data centers. These data centers are powered by electricity that emits nearly 70% less carbon than the global average—the equivalent to savings of 105 tons of carbon dioxide equivalents (tCO2e).
  

Source: electricitymaps.com

A Holistic Approach to Sustainability and AI

Sustainability is one of the Salesforce core values. This means sustainability guides actions we take across the business, including how we develop and use technology like AI.

In addition to prioritizing sustainable AI development, we’re focused on leveraging AI to create tangible environmental benefits. For example, we’re enhancing ESG reporting with AI innovations for Net Zero Cloud, illustrating the growing intersection between technology and corporate sustainability. Salesforce is also focused on supporting ecopreneurs who are using AI for climate action.

There are so many exciting possibilities for sustainable development and leveraging AI as a climate solution. By sharing learnings on sustainable AI, Salesforce aims to provide best practices to help you incorporate sustainability into your AI strategy.

Resources

• Salesforce+: Virtual Event: Salesforce Climate & AI Summit
• Salesforce News & Insights: Salesforce Paves the Way for More Sustainable and Equitable AI