The double relationship between AI and climate change
A few days ago ChatGPT turned one year old, although I don't know if it would be more correct to use a different measure. After all, a few weeks after its launch, it became the fastest growing application in history, with more than 58 million visits in mid-December 2022 and exceeding 1.8 billion monthly visits in mid-December 2022. 2023, with OpenAI stating that its website is one of the most visited on the planet.
If you are reading these lines and you are not overwhelmed by the popularity achieved by artificial intelligence in general and by Generative AI in particular in these 12 months, you may be disconnected from the world. Be that as it may, we are witnessing live and direct a brutal revolution not only because of its potential positive and negative effects, but also because of its speed.
A few days ago I was participating in the 1st Andalusian Artificial Intelligence Congress that was held in Granada. Two very intense days in which we mixed concepts such as Generative AI, Quantum Computing and Private AI; We talked about R&D use cases in multiple sectors and discussed ethics, the transformation of the workplace and a thousand other things. Well, of all the presentations I was able to attend, I chose one by Jesús González, “Demystifying GenAI: What is behind this great revolution?” I did not know the speaker, although upon listening to him I learned that he has been working in the world of AI for more than 20 years and I was impressed by his way of demystifying AI by explaining how we got here, concepts such as neural networks and supervised Machine Learning, not supervised and reinforcement learning. In Granada I saw several image and video generation models with impressive results, from ChatGPT's lesser-known brother, Dall-e (pronounced Dalí), to Emu Video (from Meta), to name a couple. I was especially amazed by the models that generate video from text.
Well, on the return trip I was reflecting on the energy consumption of these models and the thousands of GPUs that make them possible. Indeed, Generative AI, Large language models (LLMs) have a secret: they require large amounts of energy to train and function. In fact, the CO2 emissions generated by these models are still a mystery. Although I searched a little, I found a report from Hugging Face and Carnegie Mellon University that proposes estimating the emissions produced throughout the entire life cycle of the model instead of just during its training. The study provides insight into the carbon footprint of AI by offering hard numbers and reveals some worrying upward trends. The problem is that emissions are adding up quickly and the rise of generative AI has led big tech companies to integrate powerful AI models into many products. Although they are not always necessary and sometimes smaller AI models designed for specific tasks would be enough. One of the authors of the document, Sasha Luccioni, gives us a clarifying example, explaining that using a generative model to classify movie reviews according to whether they are positive or negative consumes around 30 times more energy than using an adjusted model created specifically for that task. The reason generative AI models use so much more energy is that they try to do many things at once, such as generating, classifying, and summarizing text, rather than a single task, such as classification.
While researching this topic, I discovered Professor Lynn Kaack, who heads the AI and Climate Technology Policy Group at the Sustainability Center and Data Science Laboratory at the Hertie School in Berlin. I delved into the debate on the double relationship between artificial intelligence and climate change; and I realized that we must be clear that AI is like the rest of technology, it is neither good nor bad on its own. It is important to consider both the positive ways in which artificial intelligence can be used to combat the climate crisis and explore its environmental impact.
领英推荐
The conclusion is that it is advisable to focus on developing AI models in a sustainable and responsible manner. After all, data-driven technologies can provide important climate solutions and play an important role in decarbonizing the planet.