Defining and deploying responsible AI
AI has the potential to serve a dual purpose: as a catalyst for sustainability and as an enabler of sustainability innovation. But it carries its own environmental footprint.
Bell asked panellists how they are using AI to support sustainability and what measures they implement to ensure responsible, resource-efficient deployment.
Rik Irons-Mclean, Sector Lead Manufacturing & Energy and Enterprise CTO at Microsoft Australia and New Zealand, said the company takes a balanced approach.
“We have to look at the possibilities of AI and, if we look at energy transition in Australia, we can’t do that without being able to model, simulate, and operate the next generation grid. You look at climate resilience and modelling short-term weather patterns – you need AI,” he said.
While recognising the environmental impacts, he emphasised Microsoft’s mitigation efforts, including consideration of data centre construction and operations to minimise impacts. “We look at low carbon steel, low carbon cement to mitigate…the data centre runs fully 100% on renewable energy, low carbon source,” he said.
“The actual models themselves are getting more efficient,” he said. “So, what we’re seeing as this AI usage goes up, at least the energy consumption is starting to come down but there is still an impact”.
Agentic AI in practice
Salesforce operates at the level of the technology stack where agents and agentic applications live, using AI to integrate with enterprise workflows.
For Justin Tauber, General Manager of Agentic Technology, Trust & Adoption at Salesforce, this represents a distinct set of responsibilities – different from owning data centres or supplying AI models – focused on responsible use within the organisation itself.
“Part of that responsibility is really about us being our own customer zero for our own technology and ensuring that our own sustainability goals are met by our own products, in the way we become an agentic enterprise ourselves,” he said.
For Salesforce, this means prioritising smart demand efficiency and clean data sources. “From a smart demand perspective, it’s just about using the right tool at the right time,” Tauber explained. “So, not using AI for things you don’t want to use AI for, or using the right kind of AI in the right context.”
Salesforce’s internal research teams are also focused on developing smaller, more efficient models.
“We’re trying to create the smallest possible model that’ll work, leveraging the kind of context in which we know that model’s going to be used and make it as small as possible and as efficient as possible,” he said.
“We also try and create defaults in the platform to create a path of least resistance that’s most sustainable, which we use ourselves and for our customers.”
Intel, as a manufacturer of the hardware that acts as the building blocks for AI, takes a complementary approach. Amy Pendergast, Chief Sustainability Officer at Intel ANZ, highlighted sustainability efforts across manufacturing and supply chains, but said operational carbon is as big a consideration as embodied carbon.
“We don’t want to ship something that is then going to use a whole lot of power,” she said.
Practical impacts and responsible innovation
When it comes to AI’s role in advancing sustainability, Tauber emphasised the need to rethink traditional approaches. He pointed to the benefits of agentic AI, particularly in applications such as sustainability reporting, where data comes from multiple systems, teams, and sources, often with gaps or inconsistencies.
“Delegating some sort of responsibility to an AI system to be able to take…discretionary action, is what makes all the difference,” he said.
Bell reinforced the need for collaboration.
“It’s tempting to say ‘mine, my AI, my tech, my kit’, but we can’t accelerate the sustainability challenge without collective action,” she said.
AI’s potential is amplified when organisations share insights, adopt common frameworks and act together.
Irons-Maclean noted that Microsoft’s participation in the Australian Climate Leaders Coalition (CLC) is an example of this. “The CLC…there’s over fifty organisations coming together to take collective action around climate change and getting towards net-zero,” he said.
Irons-Mclean highlighted practical applications of responsible AI, including using AI to adjust wind turbine speeds in real time, protecting migrating birds and bats while maintaining energy output – demonstrating that AI can simultaneously safeguard ecology and operational performance.
Guidance when applying AI
The panel concluded with actionable guidance for organisations integrating AI into sustainability strategies. Irons-Mclean urged immediate action, noting, “People wait for perfection far too often.”
Pendergast recommended using AI selectively and adopting an ‘AI not for everything, AI for the right things’ philosophy, including running models on renewable energy and considering AI’s end-of-life impact.
Tauber suggested measuring environmental impact via an AI energy score: “Ensure that the vendor is actually being transparent enough to develop the AI energy score for that model and check it out. Start a conversation between your Chief Information Officer and Chief Sustainability Officer about that”.
