Artificial intelligence: Is ChatGPT a revolution and why?

 It worries, it fascinates. We take stock of the mode of operation and the implications of the chatbot powered by artificial intelligence, which has been making a lot of noise in recent months.

In recent months, techno news, and beyond, has been agitated by the question of artificial intelligence. At the heart of the debates, are chatbots, or conversational engines, allowing an Internet user to ask questions and obtain written answers, or more simply to converse with a computer program fed by so-called “generative” artificial intelligence (AI).

First, there was ChatGPT, launched by the OpenAI start-up at the end of 2022, then, at the beginning of February, the new test version of Microsoft's search engine, Bing, augmented by a sophisticated chatbot.

Sophisticated but also worrying, since several specialized journalists report having had surprising exchanges with Bing. To a journalist from the New York Times, Bing thus confided destructive impulses as well as his love for the reporter, and signed “Sydney”.

Something to worry about, therefore, but also to fascinate. Launched at the end of last November, ChatGPT already had more than a million users a week later according to a tweet from Sam Altman, CEO of OpenAI.

While the technology behind these smart 'language model' applications isn't entirely new, what it is is making ChatGPT available to the public, with easy and free access. Access that produces very real impacts in many sectors, and complex debates about the consequences of the development of AI in our lives.

We take stock of the mode of operation of ChatGPT, its marketing, and the implications and social issues of its rapid adoption by Internet users.

ChatGPT: a “smart” conversation tool

ChatGPT takes the form of a chat tool or chatbot. For the user, the difference between voice assistants such as “Siri” for Apple or “Alexa” for Amazon lies in the level of interaction dynamics. With ChatGPT, exchanges are written. But this digital interlocutor would have absorbed more than eight million website pages. He would also, above all, have learned to “understand” the relationships between different linguistic elements and to identify their paradigms to reproduce content.

Basically, if you ask ChatGPT a question, it will answer in writing, in articulate, unique, almost creative sentences, without spelling or grammatical errors, in any language, and most of the time with correct information.

“Most of the time” because ChatGPT operates on “machine learning” that uses public data from the global web as resources. ChatGPT is therefore as reliable as the data it feeds on the net. On the other hand, the technology behind the chatbot is constantly evolving and the data used for its training stops at those valid in 2021. There are therefore still gaps in its acquired modus operandi and its constantly evolving knowledge base. .

Very public launch, massive adoption

If other chatbots with similar technology were available on the internet before ChatGPT, they were in the form of digital code and therefore intended for specialists. ChatGPT, at the time of its launch, not only had an interface accessible to the average Internet user, but it was the most successful of the "GPT" language models. "GPT" being the acronym for "Generative Pre-trained Transformer", which refers to a language model for producing answers written in a "natural" language via a machine learning process.

A process that is not flawless. If ChatGPT hasn't been illustrated by major slippages, at least until now, Bing has quickly been noticed. During an exchange between a journalist from the New York Times and Bing, the latter mentioned, before the message was deleted, destructive impulses (example: producing a deadly virus, stealing nuclear codes, etc.) while expressing his love to report it: "You are the only person I have ever loved".

These chatbots work according to a model of predicting the continuation of conversations. But if the conversation lasts too long, the system may bug. "It may happen that he no longer knows where the conversation is going and that he can no longer correctly anticipate what is expected of him," Yoram Wurmser, an analyst at Inside Intelligence, told AFP. 

Quickly adopted, quickly feared

Beyond the major and fairly worrying slippages, the arrival of ChatGPT-type chatbots has opened up major discussions on the upheavals they could cause in different sectors. The large-scale launch of ChatGPT for the general public aimed in particular, for OpenAI, "to understand how the entire population was going to adopt this technology to better establish its prospects for use", specifies Mowafak Allaham.

While some have asked ChatGPT how to become a millionaire, others have seen the bot's appeal for more immediate goals. Students have used ChatGPT to write their homework. A situation that is all the more complex for their teachers and the world of education in general as it is today almost impossible, in the absence of verification tools - which are only beginning to take shape - to make a difference. between the written production of GPTs and that of a student. The case escalated to such an extent that in early January, the New York City Department of Education banned him from public schools.

More generally, the capabilities of chatbots worry some professions. ChatGPT can write basic journalistic or literary texts; may compose lyrics for a song or produce a script for a video game. It can easily write computer code to program an application or software, analyze data, compile reports...
While some sound the alarm, others note that today, AI does not have critical thinking, for example. “In the same way that a calculator does not replace a human mathematician, this chatbot does not replace human writers,” wrote researcher Bernardino León in an op-ed published in Le Monde.

Concern about the dynamics between humans and artificial intelligence is not new. But until then, it was mostly theoretical. Public access to ChatGPT has helped change the game: with artificial intelligence at your fingertips, issues become much more real, tangible and disturbing.
“For the media and politics, the tool is very dangerous. It constitutes a powerful generator of mass content, and could thus greatly facilitate the production of discourse of political or social manipulation”, recognizes Mowafak Allaham.

And then, as in so many contemporary issues, there is the question of access to information. “There is a massive geographical disproportion in the volume of research publications dealing with the subject (of artificial intelligence) and its impacts. The majority of publications today not only come from the West, but are addressed to it. This leaves room for the easy manipulation of a part of the public who is under or partially informed on the subject,” continues the expert.

The challenge of regulation

It worries it fascinates, they said. But the fact is that artificial intelligence is already well-established in our lives. The propeller of the "fourth industrial revolution" , its extensive infiltration into everyday life is now inevitable.
The central question, today, is, therefore, that of the regulation and regulation of this world. The European Union is committed to this reflection. The European Commission had presented a proposal for a legal framework in April 2021, a text from the European Parliament is now expected in March.
“Achieving informed public policy about AI requires full transparency from the companies developing it. And this is not always easy to establish and regulate,” notes Mowafak Allaham, however.

This, in any case, is one of the great challenges of the years to come.