AI and new digital platforms

To go beyond mere fantasy and be able to materialize the promises linked to artificial intelligence, it is first of all essential to understanding what are the levers for integrating new data processing techniques within companies.

The animation of new communities, bringing together talents from different backgrounds, is the cornerstone of the development of vibrant and dynamic digital platforms. They allow the sharing of data, algorithms, and visualizations, with the aim of exposing services (web services) whose added value is increasing at the rate of industrialized and daily deployments.

These initiatives bring together business managers, the information system, and data architects (data-analyst and -scientist) around a common goal: to share a data catalog, build intelligent algorithmic modules, and finally allow a quick production launch of new products or services.

The establishment of a data mapping makes it possible to bring the various actors together around a common base and to facilitate access to qualified, processed, and reliable information. A second significant advantage of this approach is that it reduces duplication and repetition of tasks performed when validating the quality of these data. Data scientists have an important role to play in this upstream phase, which is often perceived as ungrateful and with low added value.

Although experts point to the generally time-consuming work of cleaning, pre-treatment, format change, and qualification, it is nevertheless a key step in providing a stable and solid database for quality work.

Finally, the sharing of raw data, but also of processed and qualified data and processing algorithms, facilitates the creation of a benevolent and supportive environment for sharing, a real catalyst in creating a proactive and involved community.

A robust and efficient way to build the algorithmic core of a product is to proceed in a progressive and iterative way by integrating the functional elements module by module. In our agile environments, it is essential to integrate a functional separation upstream: acquisition, pre-processing, core Machine/Deep-Learning, restitution module, etc. This segmentation also facilitates the version management of product codes (back and front).

Design workshops must focus on a shared deliverable and the same language: API end-points. The use of connectors on these web services regardless of the technology is chosen, makes it possible to establish a single point of passage for information, a concept that is essential to guarantee security and traceability.

This incremental prototype construction carried out between business actors, IT managers and Data scientists makes it possible to engage the community but above all to provide viable, functional products adapted to multiple needs and constraints.

The deployment of these Machine and Deep-Learning experiments in production remains a complex operation with many obstacles. For fear of uncontrolled side effects, operational activities today remain resistant to the operational use of these algorithmic black boxes: whether for monitoring an electrical network, managing bank fraud, or predictive maintenance of industrial equipment.

Nevertheless, these new models can coexist with the old world to gain both robustness and quality. The digital platform is proving to be an essential tool for aggregating but also for dynamically monitoring and weighing traditional methods on the one hand, and Machine Learning techniques on the other.

Keeping the Data scientist as close as possible to operations and the field makes it possible to envisage a secure production start-up combined with a continuous improvement of the quantitative approach.

The implementation of digital platforms remains the ideal opportunity to bring together data communities, highlighting the benefits of better sharing of data sets, facilitation during the creation phase, and, above all, enlightened production!