AI was developed with the aim of executing tasks or processes more efficiently. Precisely because of the time savings, AI can be a more sustainable choice – although this, of course, depends on how it is used.
Recent American research shows that individuals using AI perform just as well as teams without AI. In other words, an individual can achieve the same benefits with AI as through teamwork. And teams that use AI perform better and are on average 12% to 16% faster than teams that do not use AI (source: Harvard Business School (opens in new tab)). By working faster, less electricity is consumed, and fewer travel movements are made.
Blooloop (opens in new tab), the British knowledge institute for visitor attractions, introduced an AI chatbot on its website. This helps visitors find answers to their questions faster across the 50,000 web pages. They conclude that energy consumption is lower because the search time is significantly shorter.
AI can also be used to improve processes. Think of optimizing power consumption in large buildings or purchasing food and beverages more precisely to reduce waste.
The Environmental Impact of Generative AI
To immediately dispel a misconception: no one knows exactly how large the environmental impact of (generative) AI is. Tech companies do not make this data public. All the figures you read, such as five questions to ChatGPT cost half a liter of water (opens in new tab) and AI is responsible for 11 to 20 percent of global data center electricity consumption (opens in new tab), are reasoned estimates by scientists. Based on this, it is clear that the environmental impact of generative AI is significant. And it is expected to only increase in the coming years.
The use of AI in work situations in the US has almost doubled (source: Gallup (opens in new tab)). Since technological trends in America are often translatable to the Dutch situation, it is expected that this will also impact energy consumption by AI in the Netherlands.
In describing the environmental impact of generative AI, it is important to distinguish between three aspects:
Hardware (embedded emissions): The computer chips used for AI systems have a relatively short lifespan because they are used so intensively. This causes pollution during equipment production and generates physical waste. As a user, you have no visibility or influence over this.
Training generative AI systems: Training so-called Large Language Models for generative AI systems requires a lot of time and computing power. Tech companies benefit from reducing development time – and thus costs and environmental impact. Researchers at TU Delft are exploring how the environmental impact of generative AI systems can be reduced by using smaller, task-specific AI models instead of large language models. New language models are already being made faster and more efficiently than three years ago, and this trend will continue in the coming years.
Use of generative AI systems: The use of generative AI systems also consumes electricity and water to cool data centers. Popular applications like ChatGPT are estimated to consume 10 to 15 times more energy for a similar type of query than a ‘regular’ search engine. Additionally, the size of the language model you use matters: the Deep Research model from Perplexity uses more computing power - and thus energy and water - than the regular language model.
Consciously Working with AI
The European Union encourages, through the AI Act, the development and use of AI in a sustainable and environmentally friendly manner. Research into AI solutions for sustainability and accessibility is also encouraged.
But legislation alone is not enough. If cultural organizations want to achieve their sustainability goals, users play an important role.
What Can You Do?
Increase your awareness of the environmental impact of ChatGPT with the plug-in GreenChat (opens in new tab). This plug-in estimates the CO2 emissions per prompt.
Develop AI guidelines or a code of conduct to provide your employees with clear frameworks for using AI. Include ethical and sustainability considerations in this.
Ensure you have the right knowledge and skills to work with generative AI. If your entry level is higher, you can move more quickly toward a valuable application and thus need to ‘experiment’ less.
Make conscious choices All those images and music with Gen AI are fun, but are they really necessary and useful? And remember that one prompt in ChatGPT costs 10-15 times more energy than a search query in a search engine. It is, however, a question of how long this will last: Google is increasingly integrating AI systems into its search engine.
Choose the right model: Not every question requires a large language model (LLM). Smaller, more specialized, and/or local models (SLMs) can often deliver the same accuracy with less energy. Hugging Face (opens in new tab) is the most well-known platform for this. This requires sufficient technical knowledge and, in many cases, a dedicated server.
Check if there are AI systems hosted on sustainable energy, such as the Dutch GreenPT (opens in new tab), hosted with renewable energy. At the time of writing, the tool is only accessible to a test group.
More articles per topic
