Skip to content skip to sidebar Skip to footer

 

ChatGPT, OpenAI's text generator, highlights the growing use of AI in many professions. While some do with the emergence of these new tools, others oppose their adoption.

Source: © Artificial intelligence: illustrators, coders and translators already shaken up by this technology


You can share an article by clicking on the sharing icons at the top right of it.
The total or partial reproduction of an article, without the prior written authorization of the World, is strictly prohibited.
For more information, see our terms and conditions.
For any request for authorization, contact [email protected].
As a subscriber, you can gift up to five articles per month to one of your loved ones using the "Gift an article" feature.

https://www.lemonde.fr/pixels/article/2023/01/21/intelligence-artificielle-illustrateurs-codeurs-et-traducteurs-deja-bouscules-par-cette-technologie_6158756_4408996.html?random=1058896422

ChatGPT is, technically, a prodigy: this service offered by the OpenAI company is able to produce texts with such a sophisticated syntax and with so many variations that it is difficult to differentiate these creations from texts written by humans. However, ChatGPT is just the latest manifestation of a years-old technological evolution.

Read also: Article reserved for our subscribers ChatGPT, the software capable of writing confusing little texts

More and more sectors are faced with the emergence of these artificial intelligence (AI) tools which, after studying billions of fragments of sentences, images or computer codes, are able to perform tasks that we would have believed, a few years ago, inaccessible to the machine. This evolution is not without clashes or questions.

Automated illustration

Stable Diffusion, Midjourney or DALL-E are the latest in the sector: appearing in 2022, these services make it possible to transform a short sentence or a few keywords into an image representing, with sometimes surprising accuracy, the objects or representations requested by the user. Many tasks once performed by illustrators are becoming accessible to people without drawing skills who, instead, have learned to master the art of drawing. "prompt", that is, the ability to produce instructions sent to the machine to allow it to create an image from scratch.

Read also: Article reserved for our subscribers ChatGPT: "Welcome the technological advance, but understand the limits of this type of system"

For the past few months, we have seen the flowering of comic book projects produced entirely thanks to AI, prototypes of video games or even book covers. In October 2022, the Shutterstock image bank announced a partnership with OpenAI, in order to offer automatic illustration generation features directly on its platform using AI.

Ombremonde, illustrator member of the Union of Artist-Author Workers (STAA) CNT-SO, remembers his first impressions of these new products: “At first, many found it rather funny. We had a little fun with it. But, when artists began to realize that these services were using their works, it sparked a lot of debate. »

Beyond individual initiatives and calls for a boycott, a collective action was thus opened, on January 13, 2023 in the United States, by a group of authors and artists. against Stable Diffusion, accused of training his model on images without obtaining the consent of their authors. In the medium term, some illustrators fear that part of their work will be delegated to machines. “I fear that in the future we will be dispossessed of our work, worries Shadowworld. For example, that we are increasingly asked to correct images generated by artificial intelligence. »

Assistant or substitute?

A concern that is already a reality in other sectors, such as IT development. In 2021, the collaborative development company GitHub, owned by Microsoft, announced the launch of Copilot, a tool based on OpenAI technologies. Much more elaborate than the existing code completion tools, it aims to support developers in their daily production, by suggesting computer code. It's starting to look like a real questioning of the job of computer developer: if a machine automatically produces quality code, why bother learning a programming language?

Read also: Article reserved for our subscribers Beyond artificial intelligence, the ChatGPT chatbot owes its speaking skills to humans

An anxiety that Ryan J. Salva, vice president of product management at GitHub sweeps away: « We don't expect Copilot to replace developers. On the contrary, we want it to support them, increase their capacities and productivity, reduce manual tasks and help them focus on the most interesting aspects of their work.» GitHub is also pleased with the adoption of the tool on its platform: « In files where Copilot is enabled, almost 40% of the code is written by Copilot in popular programming languages ​​such as Python. Over the next five years, GitHub expects this rate to increase to 80%.."

Newsletter
“Pixels”
Social networks, cyberattacks, video games, manga and geek culture
Sign up

Developers have had to contend with AI for more than a year, and many prefer to adapt to these technologies rather than fight against them. Adrien Brunet, Technical Director at the start-up May, quickly took the fold: “I have gained in productivity, but my job has changed. He went from just “writing some code” to writing a function name that would let Copilot figure out what I want, write me ten lines of code, fix one, and move on. I don't think this type of tool will cause us to lose our jobs. At present, it always takes someone to correct it: we'll see [if this will always be the case] in the future, if the results continue to improve. »

If the developers finally adapt quite easily to artificial intelligences, the question of the data used to train them causes, on the other hand, much more excitement.

He stresses, however, that these tools may complicate the task of new generations when it comes to learning the basics of code. “Without going through the hard years where you gradually learn the mysteries of the profession, it is impossible to be able to correctly correct the outputs of artificial intelligence. I wouldn't like to be a junior fighting against an AI, even if it's also an opportunity to learn at high speed. »

Read also: Article reserved for our subscribers “Deploying an ethical artificial intelligence will be the challenge of the coming century”

If the developers finally adapt quite easily to AI, the question of the data used to train them causes, on the other hand, much more excitement. In the United States, a class action launched in November 2022 thus targets Copilot, also accused of having trained his tool on code without having obtained the authorizations of the original authors. Copilot is also suspected of offering code directly copied from other projects, without giving the source or respecting the user license. A dissatisfaction that pushed GitHub to correct the shot, working on a new feature to better source the origin of the code offered by Copilot.

To translate or to betray?

There is one last sector which is no longer adapting: the profession of translator, which has already been transformed. From 2018, the poll annual survey conducted by the European Union of Translation Business Associations showed that, for the first time, more than 50% of professional translators used an AI-powered machine translation tool like Google Translate or DeepL .

"It's an evolution of our skills that we see more and more, with translators who position themselves today as linguistic consultants, to analyze the relevance of using these technologies, explains Guillaume Deneufbourg, professional translator and teacher at the University of Lille. The profession has changed a great deal with the appearance of these technologies, but this also makes it possible to translate more, [provided that] to be well trained. Translators who translate like robots, who cannot do better than the machine, are the ones who are in danger of disappearing. »

Read also: Article reserved for our subscribers "The EU's artificial intelligence regulation plan does not seem to be up to the risks posed by the breakthrough of tools like ChatGPT"

Laurent Vannini, literary translator and member of the STAA CNT-SO, rejects this model. In addition to his regret at seeing, according to him, professionals reduced to mere “humanizers of machine creations”, he deplores the loss of meaning imposed by the use of these automatic translators: “The simple transcoding of text offered by these tools is not translation. » Mr. Vannini is also alarmed by the precariousness induced by the use of AI: “This is a sector where we already had little price security. This additional threat will not strengthen the position of translators in the face of publishing houses unwilling to negotiate. »

Leave a comment

CJFAI © 2023. All Rights Reserved.