by The arrival of ChatGPT has sparked interest in Artificial Intelligence (AI)-based chatbots, and as such, we’ve seen a few emerge, as well as integrations of what was developed by OpenAI. Although endowed with impressive capacity, these systems have, of course, one drawback: they pollute – a lot.
They are not at their peak yet, but they are already surprising and helping people with their quick responses. Despite the popularity that ChatGPT already has, as well as other similar AI systems, it still needs improvement, either because of the mistakes we know it makes or because of the pollution that results from its activity.
Building an AI-based search engine requires massive computing power. In addition, the engines are fed with huge amounts of data in order to be able to guarantee more and more accurate answers to the questions put to them by the users.
This computing power increases the amount of energy used by the companies that develop them and leads to an exponential increase in carbon dioxide emissions. If this was already a problem that companies wanted to solve, with the arrival of ChatGPT and the appearance of similar chatbots, the issue could take on a more worrying dimension.
Forming these models requires a huge amount of computing power. Right now only big tech companies can train them.
Carlos Gómez-Rodríguez, a computer scientist at the University of La Coruña, confirmed to Wired.
Carlos Gómez-Rodríguez, computer scientist at the University of La Coruña
For companies like Microsoft and Google, which aim to achieve negative emissions between 2030 and 2050, investing in ChatGPT and similar AI could delay or even harm their plans.
AI gives quick answers under a patch of smog
As Xataka mentioned, OpenAI did not disclose the computational or energy costs of ChatGPT. However, studies published by Google estimate that the preparation of GPT-3, on which the current chatbot is partially based, consumed 1,287 MWh and produced the emission of more than 550 tons of carbon.
At the same time, other independent studies conducted by some universities, such as Massachusetts, concluded that a single learning session produces as many emissions as five cars over their entire life cycle. Now, if a model is continuously trained, emissions will grow exponentially.
According to computer scientist Carlos Gómez-Rodríguez, to the amount of emissions produced by training a system like ChatGPT must be added those resulting from "the engine that works to serve the requests of millions of users".
Read also:
