Artificial intelligence user's guide

The availability of these tools and their ease of use is undeniably having a considerable impact on all higher education and research establishments. Inalco is strongly concerned, as these tools offer generation functionalities ("genAI") for languages, the capabilities of which are, however, extremely variable depending on the language.

In recent years, AI tools have been developed and gradually made available to the general public. The online launch of the ChatGPT tool by OpenAI at the end of 2022 has met with considerable success, well beyond the computer science community and even technophiles, particularly among the younger generations.

Definitions

IA

Algorithms falling within the scope of Artificial Intelligence (AI) aim to reproduce human cognitive abilities through calculations achievable on an electronic device (computers).

Generative AI

Generative AI models use algorithms to determine which words are the most likely in a process of artificial generation of data (texts, images, videos, etc.). Generation is initialized by a "prompt" (a command written in natural language). The spectrum of use cases is wide.

Large Language Models

Large Language Models (LLMs) are computer models built by adjusting a very large number of parameters calculated by artificial "learning", from very large volumes of text. They model language through the search for regularity in textual data. Parameters and statistical calculations enable them, among other things, to "generate" apparently coherent texts, without it being possible to claim that they "understand" texts as a human does. These models can be developed for a single language or for several languages simultaneously, depending on the textual data on which they are parameterized. The performance of these models varies greatly depending on the language: languages with a high presence on the Internet (English, French, Spanish) give better results than languages with a low digital endowment.

AI uses

As these tools can be used for many different use cases, it is useful to distinguish between them in order to better identify the methodologies to be adopted according to the use case, and to determine the risks and frame practices.

In the context of work carried out at Inalco, we can list the following uses:

• information retrieval,
• identification of issues,
• structuring ideas (e.g. in the form of an outline),
• checking and suggesting improvements to a text,
• automatic text translation,
• digitization or transcriptions (images for writing or audio files for speech).

Generalist AIs offer to perform these tasks, the quality of their output remaining extremely variable depending on several parameters (task, language, input data, etc.). There may also be specialized AI-based tools for each use case; we are not listing them at present, as the tools available are numerous and evolving very rapidly.

It is important to note that these AI tools are gradually being integrated into existing applications for which the use of AI is not always explicitly mentioned (search engines, proofreaders in text editors, summaries for document reading, transcriptions of videoconferences, etc.).).

Good practice recommendation

When using AI tools, it is imperative to check what has been produced by the tool with a critical eye and a great deal of vigilance. There have been many cases of "hallucinations" (when AI makes assertions that are not based on any verifiable source), particularly in specialist fields and in the generation of bibliographic references. When AI is used in a piece of work, it is recommended that this be explicitly stated, and that it be indicated which parts of the work made use of AI tools (data analysis, information retrieval, translation, writing enhancement, etc.).

If the use of AI is envisaged, it is strongly recommended that this be done in an assistance and suggestion approach, and that it not be used for the generation of the work itself. In fact, an AI can provide an insight into a production as part of an academic project. The author will be able to take note of the suggestion and take it into account if it seems relevant.

We invite those using AI to be particularly vigilant about the potentially biased or incomplete nature of the generations. These are effectively designed to provide answers from sources selected by the AI manufacturers, which are often partial and incomplete, and only very rarely admit to incompetence or limited knowledge, particularly in areas that require considerable expertise. Furthermore, their use in the context of research work presents the risk of restricting this work to the avenues suggested by the AI, so it is strongly recommended that they be used sparingly when initiating research work.

Several studies have shown that certain uses of AI can be unfavorable to the acquisition of knowledge and skills, so we recommend to everyone a reasoned use of this technology, in order to make AI a useful tool for learning, complementing without replacing traditional learning methods. Furthermore, the use of AI in the context of activities at Inalco must, as far as possible, be consensual. You can contact the AI Mission (link below) if you notice any inappropriate incitement to use AI. Finally, it is common knowledge that AIs are extremely energy-intensive tools, so their use and the call to their functionalities must be reasoned and, as far as possible, set against the energy consumption they induce.

Prohibitions

For Inalco and in a pedagogical context (courses and exams), we ask students to respect the following three prohibitions:

• not to deposit personal or confidential data on an online AI tool,
• not to appropriate an AI-generated text as a personal production,
• not to use machine translation as part of language coursework, except when such use is explicitly authorized by the teacher.

These uses of AI are currently explicitly forbidden in the internal regulations (article 71) and in the regulations governing knowledge control procedures (article 2.2) and subject to sanctions.

Page updated March 26, 2026.