ESG Initiatives Experiment with Artificial Intelligence

When India’s government, states, and union territories launched the 24 x 7 Power for All program, an initiative to provide countrywide access to electricity, state officials faced a challenge: They had to buy electricity power blocks in advance, which required them to accurately predict demand. Pen-and-paper calculations proved insufficient.

Fortunately, artificial intelligence models came to the rescue. Models help predict electricity consumption, saving budgets and wasted resources in the process.

The basic premise of machine learning, a kind of AI, is that algorithms can be trained to recognize patterns and images. The algorithms can then use that knowledge to identify similar occurrences in large volumes of data. 

In forecasting electricity consumption in India’s states of Uttar Pradesh and Bihar, for example, shared historical weather data from IBM’s The Weather Company laid some of the foundation for AI models. By studying historical weather patterns preceding electricity consumption spikes and dips, among other variables, electricity companies could proactively forecast demand and states could buy allotments accordingly.

This weather data is part of IBM’s Environmental Intelligence Suite, which is used in a variety of other ESG initiatives, as well. Companies use the AI-powered suite to predict potential effects of climate change and weather across their businesses, said Rishi Vaish, CTO and vice president of IBM AI applications.

Related:New Tools Measure Green IT, Sustainability Success

“You can think of [the Environmental Intelligence Suite] as a digital twin of Planet Earth,” Vaish said. The suite layers petabytes of satellite data from the Weather Company along with time and geolocation data to deliver business insights. An example in action: helping utilities companies figure out where to use their vegetation management budgets based on plant growth patterns.

AI Technologies in Environmentally-adjacent Programs

Consulting firm Capgemini is developing machine learning models for an image-recognition project that differentiates human-made trails from natural ones in the Mojave Desert. The Mojave, which sprawls more than 47,000 square miles, is home to many wildlife species. The desert also attracts recreational sports, such as all-terrain vehicle and off-road dirt bike riding, that can adversely affect the environment. Working pro bono for The Nature Conservancy, Capgemini has completed the first phase of the project, developing AI models that use satellite imagery to study the desert terrain over time. The project ultimately aims to measure the impact of humans on at-risk species, and to suggest alternatives for recreational activities.

Environmentally focused AI projects can also take the form of operational improvements. Organizations can shave inefficiencies from everyday business operations, thereby preventing waste. IBM’s Maximo, for example, uses AI for predictive maintenance of assets so they last longer.

ESG Initiatives Confront AI Data Challenges

ESG initiatives that use AI will run into problems that all fields face with AI deployments. These include, according to Vaish, “the need for good data, being able to train relevant models, and making the results explainable to people driving the process.”

As Capgemini found, decades-old satellite imagery does not have enough resolution -- pictures are taken from too far above the earth -- for detailed, ground-level insight and identification. Patterns might be missed.

“Since photos are just a shot in time, they sometimes don’t reveal too much,” said Dan Simion, vice president of AI and analytics at Capgemini.

Capgemini used synthetic data to complement their existing sets. “We used AI to generate synthetic data, so we could still train the models if the quality of images was inadequate or if we didn’t have enough instances of a particular kind of image like [images of animal] nests,” Simion explained.

Getting user buy-in can also be a challenge for ESG initiatives that use AI. “It's critical to earn the trust of business users, detect and remove any bias or drift in the data, and operationalize the development-to-deployment lifecycle for the AI models,” Vaish said. “This can be especially complex in the space of asset and supply chain management and environmental intelligence, as they each have unique data sets (some in the public domain, some within an organization, some spanning organizations), as well as unique integrations between systems and applications. All of this needs to be solved for an AI solution to be effective and consumable.”

Even if you address these limitations, AI alone can’t drive an ESG initiative forward. “There has to be organizational commitment to defining ESG goals, investing in and empowering teams to achieve these goals, tracking progress, eliminating roadblocks, and measuring and accounting for success,” Vaish said.