While the impact of artificial intelligence (AI) In the working world or in terms of misinformation spilled a lot of ink, industry leaders, including Elon Musk, are calling for decelerate. An ultra-publicized artificial intelligence played a major role in the opening of these debates: ChatGPT. But did she really create a break?
There is no real technical novelty in ChatGPT. His ancestor, Eliza, dates from 1966. The coding technique, as sophisticated as it is, is part of a continuity of the conceptions of programming language. Moreover, in a sociological approach, remember that we are a digital humanity, a society artifactual (since we evolve through technical progress), in which all innovation is part of a continuity.
Can we then really speak of breakthrough innovation for ChatGPT?
And if it was finally necessary to seize this media tidal wave to understand the fulgurance of its deployment, and even more, to understand where we are, we, users, in our representations, between fantasies and realities, of artificial intelligence?
These conversational intelligence software bring us back to two major trends related to the technicization of society. First of all that of the servitization, that is, a standardization of personalized assistance on a daily basis. This trend arrived with the transition from an economy based on mass logic to another based on individual logic.
The second major lever is the plateformisation : created within digital empires (among others GAFAM), this software is designed to be infused throughout the ecosystem. This is the case for example of ChatGPT, brought to invest Bing, Outlook, and other Microsoft services.
And for good reason, the unsightly "GPT" in its French version, refers to "general purpose technologies". It is therefore by its nature a tool designed to be easily transposed and used in different fields. Different from a search engine, it answers complex questions and seeks to understand the intention.
How to explain the so rapid appropriation of ChatGPT by users?
The media impact is due to its users. As pointed out by the anthropologist Lionel Obadia, it is difficult to speak of invention. Nevertheless, the remarkably rapid diffusion and appropriation make ChatGPT a remarkable innovation. It can be used by anyone, initiated or not, for free online, and therefore leads to as many new uses, workarounds, so many potential businesses.
I asked chatGPT to write a song for Yannick Noah and he did it perfectly 👌 pic.twitter.com/q2AF1WDK7C— Maud Estfashion 🇺🇦 #womanlifefreedom (@elnabernina) March 2, 2023
Three forces explain the rapid integration and social appropriation of the software.
First, ChatGPT is very accessible: which allows it to be scaled up, in other words a very rapid deployment. The accessibility of the tool and its generalized appropriation massiveize the use, but also the interest for what seems "new", even "entertaining".
Then ChatGPT looks like us. The process of machine learning makes it a tool for continuous self-enrichment, like humans. The dialogue allows him for example to improve his answers with a natural language. So the more it is used, the better it performs. In other words, we "educate" each other.
Finally, ChatGPT is a field of possibilities: it suggests new scenarios. We, as users, are the condition for the successful deployment of these digital innovations. Done for ChatGPT which now has more than 100 million users. We are beginning to become aware of the “after” and the emergence of other innovations resulting from the model of this software.
"It was with a certain thrill that I took up my red pen to try to understand the reasoning elaborated by a machine. it was "not so bad" [...] So it was with a mixture of fascination and dread that I finished the exercise, sincerely wondering what we were going to do for the world to come, for these generations who will not have to know the effort."
Professions of cognition doomed to disappear?
What really differs with previous innovation-related evolutions is that ChatGPT touches the cognition professions, rather spared until now. It is also interesting to remember that one of the most (technologically) complex software, Tesla's Autopilot, was 85% made by artificial intelligence. The dynamic of hybridization between professions and artificial intelligence is increasingly strong. However, it would be fantasized to speak of a "great replacement".
The autonomy of conversational intelligence is not total. For example, ChatGPT developed by Open AI was built from 175 billion parameters, it relies on a gigantic dataset. Gigantic certainly, but selected, therefore fallible.
Recently, exchanges and readings on the art of coding allowed me to shed even more light on the limits of the myth of the autonomy of artificial intelligence. This is particularly through the dimension of labeling in neural networks. In this process, it is label data, that is to associate a label (a predefined value) with a content (image, text, shape, etc.). The goal is to bring a learning model to machines. This practice recalls the need for configuration and supervision in the interpretation of data, a reality to (re) discover in the documentary mini-series The click workers by Antonio Casilli. Open AI has different contrats with workers in Kenya for this moderation work.
We therefore see the limits of a fantasy around a perfectly autonomous artificial intelligence.
Although artificial intelligence does not understand what it is saying, these human implications in the supervision of machine learning show us that the data that feeds this intelligence are not neutral. They reproduce and amplify the biases of those who supervise it, carrying a potential batch of stereotypes, misinformation, content sucked from invisible sources, censorship, or even conspiracy (Meta experienced this with Blender bot 3).
The central question is therefore: who sets the parameters? The law professor to Lawrence Less emphasizes it in his clairvoyant "Code is law" : the person who sets is the person who has the possibility of orienting a system of thought. A tool like ChatGPT is mainly designed by and for a North American target. Its rapid deployment, reinforced by its playful dimension, leads to the standardization of its use. A standardization not without effect, since it reinforces the impression of a politically neutral object among users.