Montpellier requires its employees to renounce ChatGPT, the chatbot powered by artificial intelligence. A ban that will be difficult to implement.
In Montpellier, city hall employees are now prohibited from using ChatGPT. The instruction was taken by Manu Reynaud, the deputy mayor in charge of digital, revealed the Midi Libre on March 27. The decision has been made as a precaution and as a precaution “, Details the elected, and it also concerns the external interlocutors of the town hall, in particular in the case of contracts.
This is the first time in France that a town hall has taken such an instruction, but ChatGPT is already banned elsewhere, notably at Sciences Po. The chatbot developed by OpenAI, fascinating in many ways, also fuels various controversies and anxieties. Other establishments and metropolises could well decide to take the same path as Montpellier.
A matter of security
In Montpellier, the ban is mainly motivated by security issues, according to Manu Reynaud. ” We cannot guarantee that information exchanged with the platform will be protected against possible breaches of security, rights or confidentiality. “. This security requirement explains why the instruction applies to external interlocutors of the town hall, in particular if contracts are at stake. We often call on consulting companies such as design officess. So we give them our data “, he reminded BFMTV.
ChatGPT saves the data provided to it, that is to say the prompts — the requests — that Internet users send to it. Thus, if employees ask ChatGPT to classify confidential data, the latter is exposed. Data that ChatGPT may retain as part of OpenAI’s process to improve the efficiency of its AI systems.
It is therefore a very bad idea to provide confidential, sensitive and personal data to ChatGPT, especially if it concerns third parties or your company. There are also risks of leaks, as we saw in March, when an AI bug made visible excerpts of conversations he had with other Internet users.
Another reason explaining the reluctance of the city: ChatGPT can be wrong – it is therefore dangerous to rely on him for questions related to work. The elected official adds that legal questions arise: “ number of markets are calculated on an hourly volume of work. If AI makes it possible to place an order for several weeks in a few minutes, it can lead to litigation. »
How to check that the prohibition is well respected?
However, there remains a problem: the means of control do not yet exist “, admits the chosen one.
Indeed, ChatGPT is available for free from the net, without the need to install any software. Limiting access is therefore difficult. Even though work computers could block the site, there would still be the possibility for employees to access ChatGPT from their personal equipment, through their private connection.
Likewise, there is no software that can say with certainty whether texts have been written by ChatGPT. Some tools have been developed, such as “AI Text Classifier”, but the latter is not infallible and regularly makes errors.
It is difficult to guarantee that the measure will be applied. Above all, the current measure only targets ChatGPT – this is not yet the case for Bard, Google’s AI. No other language model is banned yet.
But Manu Reynaud does not necessarily want to block all use of AI. Le Midi Libre thus specifies that the elected official asked the mayor of the city, Michaël Delafosse, to ” initiate work with experts “. These must “ measure the changes that will occur with the arrival of AI, and how this technology will be able to help guide local action “.
Do you want to know everything about the mobility of tomorrow, from electric cars to pedelecs? Subscribe now to our Watt Else newsletter!