Urgent questions related to the need for global governance on Artificial intelligence (AI) have been raised during the AI Action Summit in Paris, from 10 to 11 February 2025. Golestan Sally Radwan (Egypt), Chief Digital Officer at United Nations Environment Programme (UNEP), explained to UNRIC what the dangers are but also the potentials of AI in the field of environment.
What was your main advocacy during the AI Action Summit in Paris?
One of the biggest issues around AI governance is addressing its environmental impact. While energy consumption is often the focus—and rightly so—the entire value chain requires thorough examination. At the moment, we don’t have agreed-upon standards for what should be measured, how to measure it, and what are acceptable levels of environmental impact.
There was an essential need for a dialogue at the global level on those aspects and this is why, together with the Government of France and our colleagues from International Telecommunications Union (ITU), we launched at the Paris summit a new Coalition for Sustainable AI, bringing together stakeholders across the AI value chain for dialogue and ambitious collaborative initiatives. The Coalition aims to strengthen sustainable AI’s place in the global discussion around AI in much the same way AI security or AI ethics are studied.
AI is often criticised for its environmental footprint, and its potential role in climate disinformation. Are you worried about these negative aspects at UNEP, and how do you address them?
We are very concerned about AI’s environmental footprint. Direct impacts include everything from the extraction of raw materials, rare earth elements and minerals, for hardware that go into manufacturing hardware used for AI, to data center construction and operation, energy consumption, greenhouse gas emissions, water used for cooling but also waste and e-waste.
We also need to have a look at the wider picture, with indirect impacts such as unsustainable production oroverproduction and consumption, but also misinformation and disinformation. This problem occurs when AI gets proliferated, especially generative AI, which can generate content that wasn’t there before.
Large Language Models, one of the currently most popular forms of AI, have an inherent hallucination problem. In addition, Generative AI is often used to create fake content, whether it is text, images, or videos, which is contributing to an increase in misinformation and disinformation. Unfortunately, climate and environmental topics in general have been very negatively affected by misinformation and disinformation in the past.
At UNEP, we want to limit those risks and ascertain the value of global environmental law and related scientific work. For this reason, we are developing a tool based on existing models but trained exclusively on UNEP-approved scientific reports, ensuring accuracy while minimising hallucinations. It provides scientifically based and authoritative information on climate change and air pollution, for now, citing its sources for full traceability—unlike standard Large Language Models, which present risks of hallucination.
The tool is in beta at the moment, but once completed, will be released to the public so that it can be, among other things, a way to fight misinformation.
When you say that generative AI can hallucinate, what do you mean exactly?
If you train an AI on the whole internet, it becomes excellent at constructing grammatically correct sentences in English, French, or Italian. However, it doesn’t understand their meaning.
As a result, it can produce fluent, beautiful sentences that are factually incorrect or entirely fabricated. AI can generate things in the voice of Shakespeare, for example, but it can be things that Shakespeare never wrote or things that don’t even exist. It lacks a way to verify the veracity of the information. This leads to a high rate of “hallucinations,” especially in models like GPT.
The newer generation of AI, called reasoning models, includes content checks before responding. We expect their environmental footprint, especially the energy used, to be much higher on that “inference” stage, when users start to make it work. While we hope these reasoning models will be more accurate than large language models, hallucinations can still occur.
That is why I always advise colleagues to carefully review AI-generated content and ensure it is accurate — never assume it is entirely reliable. Chances that it could make things up are quite high.
Can you tell us about a specific innovative AI-powered platform helping monitor climate change?
The International Methane Emissions Observatory (IMEO), launched by UNEP in 2021 under the name “An Eye on Methane”, is a good example. It’s an AI-powered tool that detects methane emission hotspots around the world.
If it sees unusual plumes, it will send an early warning signal to affected governments enabling them to implement mitigation plans.
What is the most interesting potential of AI in the field of environment?
AI can support smarter investment decisions by integrating diverse data sets on what is happening in countries environmentally, regarding their SDG progress, and global issues like climate change and biodiversity loss. Imagine being able to combine all of this data and then adding to that question that a country has the following challenges and plans, such as a UNEP project to replace 500 diesel buses there with hydrogen-powered ones.
With AI, you can assess not just monetary costs but also the time, resources, and projected impacts on the environment, public health, and transportation efficiency. To go further, whatif the project is scaled to 5,000 buses? AI can calculate the incremental costs and benefits, creating a detailed business case for each intervention.
This forward-looking approach can guide better decisions and provide valuable insights for donors and stakeholders.
Is UNEP developing its own tools to monitor progress towards the SDGs with, for instance, the Freshwater Ecosystems Explorer or the UN Biodiversity Lab?
The Freshwater Ecosystems Explorer is very specific to SDG 6.6.1, “Change in the extent of water-related ecosystems over time”. It is tracking changes in water-related
ecosystems over time by monitoring freshwater ecosystems and their changes to identify
encroachments and the impacts of pollution and climate change on freshwater.
The UN Biodiversity Lab, a partnership between UNEP and UNDP, tracks biodiversity data, allowing users to explore the distribution of plants and animals worldwide over time. This
tool can also overlay with other data types to detect changes.
Next, we are developing the World Environment Situation Room, which will integrate data
from those different applications to provide holistic insights. To give you a specific example, it
would be able to process and analyse multiple variables, such as how floods in India could
impact the global rice supply chain over a given period of time.
What is Environmental Data Governance and why is it important?
Environmental data includes satellite and earth observation data for monitoring climate change and biodiversity, as well as data collected on the ground to measure water levels and pollution levels with sensors. It can also cover statistical data, such as progress on SDGs related to the environment.
This data is increasingly vital for decision-making and scientific assessments. However, it is often poorly managed. For example, there are no standardised protocols for securing environmental data, or frameworks for exchanging these. If you have a biodiversity dataset and I have another, we can’t always automatically exchange or combine them, as we may be using different standards or methods for collecting and storing the data.
Data classification and quality levels are also key issues that need to be tackled. Without an authoritative body, like the Intergovernmental Panel on Climate Change (IPCC) for climate data, we struggle to determine which dataset is reliable and has highest standard, and which is suitable for specific applications. This hinders our ability to fully leverage environmental data.
Should the average citizen fear IA or embrace it as it is the next digital revolution?
I wouldn’t say fear AI but use it with caution. We should understand what AI can and cannot
do and use it for the intended purposes.
Always verify the output of generative AI, for example. Also, be mindful of its environmental impact. Ideally, we’ll one day have a clear estimate of the carbon footprint of the individual use of AI, similar to how we calculate emissions from flying. This would help us make more responsible choices.
Overall, I would say: know what you’re using AI for and never rely on it too much—always apply human judgment.
USEFUL LINKS
- Artificial intelligence: the UN at Paris summit
- The Governance Challenges of Artificial Intelligence (UNU-CRIS)
UNEP:
- New Coalition aims to put Artificial Intelligence on a more sustainable path.
- ÉlyséeArtificial Intelligence (AI) end-to-end: The Environmental Impact of the Full AI Lifecycle Needs to be Comprehensively Assessed
- Digital Accelerator Lab
- A global foresight report on planetary health and human wellbeing