Recent developments from the turmoil at OpenAI to the UK’s AI safety summit to the draft EU AI law all signal the high stakes – commercial, social and environmental – of AI technologies. These high stakes were reflected at the sprawling COP28 venue on the edge of Dubai, where AI featured in climate negotiations for the first time and was the focus of a large number of side-events.

Nations are already using AI for climate objectives, from mapping the carbon absorption capacity of forests in Indonesia to providing communities in Malawi with flood warnings fifteen days in advance, enabling them to evacuate. But much more remains to be done to respond to UN Secretary-General António Guterres’ call to ‘develop AI that is reliable and safe and that can … supercharge climate action’.

That’s why the UN Climate Technology Mechanism has this year created a dedicated Initiative on Artificial Intelligence for Climate Action (AI4ClimateAction). The Initiative provides space for policy discussions on climate and AI, supports capacity-building for developing countries to harness AI and will develop regional networks of actors that can support AI for climate action.

We developed this initiative because the potential impact of AI on climate change is significant and not yet broadly understood. Serving the near-universal membership of the UN Climate Convention and Paris Agreement, the Technology Mechanism is well placed to provide a transparent and inclusive process to explore AI’s opportunities and risks and to bolster international collaboration within a UN framework. Also, for institutions tasked with advising the COP and assisting developing countries on technology, proceeding as if AI is not a thing would not have made much sense.

In Dubai, the initiative launched an AI Innovation Grand Challenge, in partnership with non-profit AI community Enterprise Neurosystem. This competition invites submissions of AI and machine learning applications that can help reduce emissions or build climate resilience in developing countries. It is intended to result in a free AI climate application hub, available for all nations to use.

The jury is still out on whether AI will reduce more greenhouse gas emissions than it will cause. From the host nation United Arab Emirates, world’s-first AI minister Omar Sultan Al Olama argued that AI is the ‘only solution’ for meeting the 1.5 degrees target due to its ability to ‘crunch incredible amounts of data’, but also cautioned that AI would contribute to emissions due to the huge amounts of power it requires. This is the climate-AI dilemma. Resolving it will not be simple, but is clearly going to require massive investment in the greening of data and therefore of electricity.

The COP28 outcomes reflect this mix of AI hope and concern. Addressing AI for the first time in the decision it finally adopted on Wednesday, the Conference of Parties to the Paris Agreement (the CMA) noted that the AI4ClimateAction Initiative has the aim to ‘explore the role of artificial intelligence as a technological tool for advancing and scaling up transformative climate solutions for adaptation and mitigation action in developing countries, with a focus on the least developed countries and small island developing States’.

The CMA highlighted that the Initiative would also be ‘addressing the challenges and risks posed by artificial intelligence’. Challenges and risks including ‘energy consumption, data security and the digital divide’, as the CMA noted in a separate Dubai decision.

With these statements, the supreme governing body of the Paris Agreement underscored the need to focus on what AI can do for the poorest and most climate-vulnerable nations. It also emphasised that the AI4ClimateAction Initiative is not promoting AI for its own sake or pushing it as a climate solution, but rather taking a balanced approach that includes consideration of both pros and cons.

With AI capabilities currently concentrated in a handful of countries, there is a clear need for a multilateral approach that gives all nations a seat at the table. When it comes to climate change, the goal must be to place AI at the service of all humanity rather than letting it become a new generator of inequality. COP28 has set the direction and the AI4ClimateAction Initiative will explore the potential.