The Digital Edito
Chat-GPT Revolution or threat? [04 April 23]
Since a few days, Chat-GPT is back in the spotlight, and is once again causing a buzz. Many voices in the French digital industry are raising their voices to highlight the French and/or European “delay” in this field, claiming that we are missing the new technological revolution. At the same time, many voices of the American digital giants wish to put down the development of these technologies… What to think?
I would be careful not to make assertions and certainties, however I would like to put forward some elements to add to the reflection on the subject.
Are France and Europe behind? Yes and no. Indeed, there are many French start-ups, including Lexistems, whose co-founder Marie Granier gave me an interview. Contrary to the buzz and the means used by Chat-GPT, we have here solutions intended for companies and organizations with a real added value… Does not being on the front of the media scene mean that we do not exist and that we do not bring solutions with high added value?
Are we at the dawn of a technological revolution, again yes and no… Yes because the algorithms of these generative AIs have been improved, and the allocated computing power keeps increasing, and more and more powerful chips (hardware) dedicated to these uses are developed.
No, because these are incremental improvements, and there are still a number of questions that raise multiple concerns.
These generative AI technologies are based on an architectural brick: the cloud. The problem of mastering this brick, both technologically and legally, persists. In particular, issues such as information security and the use of personal data remain a critical issue without a real and serious answer. Those who tell you that we must devote ourselves to AI without having resolved our strategic autonomy on the cloud have a selective blindness!
Then, come for me problems more specific to the use case of Chat-GPT. The one of direct, voluntary biases and the indirect ones linked to its learning base : the web. The opacity that surrounds the functioning of chat-GPT and the management of the inevitable biases pose a real problem:
- Reinforcement of fake news: AI generates content based on biased content that feeds the web, which is used to produce new content… Self-reinforcing bias…
- Chat-GPT unlike for example Lexistems for professional use, does not source the origin of its contents. It will be able to give you a nice dissertation without supporting it with citations and/or reference works (at least at the moment).
- Last but not least, it is ideologically oriented like other tools on the internet, which poses a real problem of risk of indoctrination or mass propaganda. Here, we are talking about Chat-GPT, but it would be wrong to think that we don’t have the same problem with Chinese AI…
For the moment, these generative AIs are not yet at the stage of their own consciousness, the fears linked to this imaginary world are not yet current. However, it is worrying to note that the creators of these AIs master less and less the real functioning of their creation.
These elements only confirm Jacques Ellul’s observation, we are in a technical system which is self-sufficient and self-increasing, the place of Man being reduced to the smallest part of an interchangeable part in this technical world…