Chat-GPT, not as free as you think

How convenient, a bot that writes your thesis, plans your holiday and writes your Sinterklaas poems. In 2022 Chat-GPT took over the world and since then it has improved a lot. A lot of people see it as a replacement for google, but better. You ask it a very specific question and it gives you the exact answer that you need. But at what cost?

Chat-GPT, Generative Pretrained Transformer, is a research preview published by OpenAI. Besides all of the moral questions that this tool proposes it also affects the world's durability. To let ChatGPT get better, it needs to be trained. It gains its knowledge by running data through it. A lot of data. To run all of this data, a lot of energy and water is needed. To give you a bit of perspective, the training of GPT-3 required 1.3 gigawatt hours of electricity, which is just as much as the energy consumption of 121 American households per year (DeGeurin, 2023). This cost approximately 500 tons of CO2 emissions. And this is just the training of the bot alone. A conventional search using Chat-GPT uses sixty times more energy than a regular Google search (Vanderbauwhede, 2023).

Luis Cruz is dealing with the sustainability of AI at the TU Delft and even started a class ‘sustainable software engineering’ for TU Delft students. In an interview on TUDstories he mentioned that there are multiple ways to make AI more energy efficient. His favourite is presented as a favourite. Meaning to have a stricter selection in what data will be used for the training of the bot. Instead of running all available data through it, only the necessary things will be filtered out. 

A completely different approach is undertaken by Sander Bohté, Professor of Cognitive Computational Neuroscience. He is working on a chip that takes inspiration from our brain. Our brain is very energy efficient and he is trying to put this efficiency into an algorithm. This chip would be 20 times more efficient than the current software (NOS, 2023). 

In conclusion, Chat-GPT is a really handy tool to help you with your tasks and it keeps improving. Unfortunately this improvement costs a lot of energy and causes a lot of carbon emissions. So next time you are thinking about using Chat-GPT for a regular search, think about our precious earth for a minute and maybe just Google it. 

 

Sources:
DeGeurin, M. (2023, 10 mei). “Thirsty” AI: Training ChatGPT Required Enough Water to Fill a Nuclear Reactor’s Cooling Tower, Study Finds. Gizmodo. https://gizmodo.com/chatgpt-ai-water-185000-gallons-training-nuclear-1850324249
Duurzame kunstmatige intelligentie: van ChatGPT naar groene AI. (z.d.). TU Delft. https://www.tudelft.nl/stories/articles/duurzame-kunstmatige-intelligentie-van-chatgpt-naar-groene-ai
NOS. (2023, 31 mei). Kunstmatige intelligentie vreet stroom, één opdracht hetzelfde als een uur een lamp aan. https://nos.nl/nieuwsuur/artikel/2477186-kunstmatige-intelligentie-vreet-stroom-een-opdracht-hetzelfde-als-een-uur-een-lamp-aan
Vanderbauwhede, W. (2023, 17 november). Emissions from ChatGPT are much higher than from conventional search • Wim Vanderbauwhede. https://limited.systems/articles/google-search-vs-chatgpt-emissions/