Artificial intelligence is undeniably opening up new horizons. But the implementation of these technologies at the heart of professional environments must take into account a number of issues, linked to regulations, security and the preservation of data confidentiality. Jérôme Girs, Senior Consultant at NSI Luxembourg (CEGEKA Group), comments: "The implementation of tools based on artificial intelligence needs to be properly supervised. Because these solutions cannot function without data, questions automatically arise about the use, location and confidentiality of that data. Furthermore, an artificial intelligence solution can be more or less demanding in terms of IT resources. Depending on the platform and the models chosen, the costs associated with using it can vary considerably."
The implementation of tools based on artificial intelligence needs to be properly supervised.
Opportunities and constraints
An AI solution, such as a chatbot, that supports human resources by providing answers to questions asked by employees will be required to process personal data, which is protected under the General Data Protection Regulation (GDPR). In the context of a legal company, AI can facilitate the analysis or drafting of contracts on the basis of an existing corpus. This data, however, is not intended to leave the company. "There's nothing trivial about these considerations. In view of its needs and constraints, any organisation wishing to base its development on AI will have to ask itself a number of questions. The answers provided will determine the nature of the platform to be implemented to support the solution, but also the policies for using AI in companies", continues Jérôme Girs. While CoPilot or ChatGPT, in their versions accessible to the general public, have demonstrated their effectiveness from many points of view, their use in companies must be strictly supervised, at the risk of exposing the organisation to regulatory sanctions.
The crucial choice of platform
Fortunately, these artificial intelligence models, like many others, can now be easily deployed in dedicated environments. "If required, it is possible to deploy models on a private infrastructure, locally. However, you need to be aware that artificial intelligence requires significant computing power to operate effectively. In this case, GPUs will have to be installed locally. But these are expensive and their availability is limited", adds Jérôme Girs.
The alternative, then, is to implement a shared platform, but located in Luxembourg or Europe, hosted in environments offering sufficient guarantees of security and confidentiality. With this in mind, in partnership with IBM, we are implementing environments from which our customers can deploy artificial intelligence models according to their needs," adds NSI Luxembourg's senior consultant. We also help them to structure their data, which they can host either on their own premises or via the platform, depending on the constraints they have identified".
Opting for the right models
As we said earlier, data is the fuel needed for artificial intelligence solutions to function. The models, whether they are called Llama (META's model), Mistral or DeepSeek, are the engine of the machine. "In addition to locating the data, the choice of model is also a decisive factor in implementing the solution. The IBM platform (Watson X), in this case, provides access to a vast variety of models that can meet a wide range of needs. The capacity of the models, their size and their level of instruction are all parameters that influence the performance of the solution, but also its GPU requirements and cost", adds Jérôme Girs.
Meeting the governance challenge
IBM's platform guarantees that the models accessible via its platform do not expose users to intellectual property issues.
"The other major challenge lies in the governance aspects that need to be put in place around the models, to ensure that use complies with regulations (IA-Act, RGPD, DORA, etc.) and to ensure the validity of the models over time", continues the expert. "Moreover, IBM's platform offers standard contractual protection of intellectual property, which is a definite advantage over other solutions in the sector."
AI solutions, whether predictive or generative, are formidable levers for innovation and improving competitiveness, as long as they are properly implemented.
Unleashing potential
Given all these possibilities, the NSI Luxembourg teams are there to help organisations ask themselves the right questions and make the right choices, based on their needs and constraints. AI solutions, whether predictive or generative, are formidable levers for innovation and improving competitiveness, as long as they are properly implemented," explains Jérôme Girs. They are currently very effective for extracting data from documents, generating summaries, classifying information or e-mails, producing code or even translating text. Our role, alongside organisations, is to enable them to exploit all this potential, through the implementation of technical solutions, but also by training teams to ensure optimal management and use of the tools deployed."