You are currently viewing GPT-3: an artificial intelligence that is both breathtaking and disconcerting.

GPT-3: an artificial intelligence that is both breathtaking and disconcerting.

You have just read the title “GPT-3: an artificial intelligence that is both breathtaking and disconcerting” and probably without any doubt you thought that this article would be written by a student. What if I tell you that I am not the only one capable of writing this article, but that maybe this article could have been written by artificial intelligence. Is this in the realm of science fiction? No, this belongs to the field of GPT-3 (Generative Pretrained  Transformer Version 3).

NDLR; First, let’s get this clear, I can assure you that this article was written by a human being (1m81,  blond hair and a bit charismatic).

GPT-3 is an algorithm developed by Open AI; a company co-founded by Elon Musk in 2015. The biggest achievement of this algorithm is that it can translate text (for instance English words) into a programming language. If you are neither a nerd nor a computer science expert, you were probably always lost with a programming language such as JavaScript or Python. As a language translator, GPT-3 can now help you to program by translating your wishes in plain text into a computer program.

Pixabay, CC BY-SA

GPT-3 is more than a translator, it is a lawyer, a medicine, a  journalist and even more.

With more than 175 billion parameters, GPT-3 is the largest artificial neural network ever created and thus detains more neural network than what an average human being has. This leads GPT-3 to do mind-blowing tasks. Indeed, this algorithm can solve complex cases, provide legal advice or even medical diagnostics. The average of accuracy is however unknown, but several experiments have shown that GPT-3 is capable of a high reasoning.

This is made possible by the absorption of billions of documents. To measure, the GPT-3 knowledge represents only 0,6% of the Wikipedia online sources.

A mind-blowing algorithm with non-negligible side effects. 

Environmental concerns.

From an environmental perspective, GPT-3 is not ideal. According to the  Massachusetts Institute of Technology (MIT), training the GPT-3 requires the carbon equivalent of a flight over the United States. This is because the extensive capabilities of the algorithm require a high-power consumption. If its use were to become widespread, imagine the colossal carbon impact it would have. The founders of Open AI are aware of this problem, but so far no tangible solution has been found to reduce the carbon footprint of the GPT-3  algorithm.

Unacceptable outcome.

Even though the GPT-3 can handle complex cases, the algorithm sometimes produces wrong outcomes. These wrong outcomes occur as a result of training on dirty data. Dirty data are false, incomplete or non-integrated data. The predecessor, GPT-2 was removed from the market in part because it created unacceptable outcomes. The activity of GPT-2 was of real concern because it produced discriminatory outcome and released disinformation over the web. As an example, GPT-2 helped Trump supporters to spread wrong information during the American elections in 2016 to favour its election.

To address this issue, the Open AI’s developers decided in their third version of GPT, to limit its access to actors who make sure to use the algorithm for a fair purpose without input of dirty data.

A threat to employment?

Finally, we often hear that AI will destroy millions of jobs with the strong argument that it has already put thousands of factory workers out of work. It is a question that makes sense too here. Indeed, if we have an algorithm that can solve legal issues, why should we have lawyers? If we can rely on AI to diagnose a disease, couldn’t we reduce the medical workforce by half? This is the common issue for all new technologies that perform human tasks.

The consultancy firm PWC (Price Waterhouse Coopers) confirms that millions of jobs will be destroyed. However, they do not predict a massive unemployment rate because at the same time as AI destroys jobs it creates new ones. We should therefore speak of a transformation of work rather than an episode of mass unemployment. In a way, it is history repeating itself, as industrial change occurs jobs are destroyed to create new ones. Those who know how to adapt to these changes will not have to fear unemployment.

            Cailin Van der zijden

To know more about the subject.

– “OpenAI API”, June 11, 2020, <https://openai.com/blog/openai-api/>

– David Larousserie, “GPT-3, l’intelligence artificielle qui a appris presque toute seule à  presque tout faire”, Le Monde, 03 novembre 2020, <https://www.lemonde.fr/sciences/article/2020/11/03/gpt-3-l-intelligence-artificielle qui-a-appris-presque-toute-seule-a-presque-tout-faire_6058322_1650684.html>