background

Artificial intelligence

Beyond knowing about AI, where do we go from here? Conversation with Yves Caseau

Share on
Custom objects
Content page

Yves Caseau

Since 2014, investments in artificial intelligence have increased considerably. Venture capital funds have invested massively. But will the variety of topics involved and the turbulence encountered translate into a real revolution, or simply evolution? Will all the algorithms developed become the property of big tech companies, the leaders of the new digital economy, or could they be developed and used by all companies, regardless of their size? Yves Caseau IT Director of Michelin Group and Member of the French Academy of Technologies is the co-author of a report on artificial intelligence and gives us his opinion on the subject.

Can you take us through the diversity of issues involved in AI?

Merck, a vaccines manufacturer, who have very complex processes, are looking to improve the performance of their production facilities with machine learning, moving beyond the traditional optimisation methods, which focus on each stage of synthesising medicines individually, to instead look at the entire chain as whole. Continental, the American hotel chain, traditionally approached customer segmentation as a way to analyse churn rates and appetites for their services. Now, new technologies are helping them create ultra-refined segmentations, with tens of thousands of segments covering their millions of customers. Finally, in medical imaging, doctors have been wondering how to use 2D echocardiogram images to come up with data that could only be obtained using 3D models, which are much more difficult to obtain.

What are the common threads amongst all these approaches at these companies?

All the projects have a systemic, cyclical vision. The companies are using simple systems, sometimes simple correlation analyses, but with millions of parameters, considerable databases created over time, which are analysed continuously. Analyses develop as and when new data is acquired. So to do such remarkable things, this dynamic approach is vital. Rule number two is to share the data collected with the company’s entire ecosystem, including its customers, partners and suppliers. It’s this collection of all possible parameters, both internal and external, associated with the systemic cycle, that really accelerates performance.

Is the French ecosystem favourable to the development of AI?

In France, upstream systems are efficient, but the average company’s ability to develop large-scale distributed systems remains lacking. The desire to put AI-enriched services into the hands of companies and their customers also remains weak, limited by too much caution. Moving forward, it’s this iteration, anchored in concrete applications, that will drive performance. Finally, the GDPR and CNIL add additional constraints and more weight to the brakes.

How can we map artificial intelligence approaches when our toolbox is so full of solutions?

We need to segment things according to the amount of data and types of questions being asked - distinguishing a specific question to respond to a given issue or question with entirely open exploratory approaches. Depending on the issue and approach, we need to play with the hybridisation of methods.

How can we begin an AI approach?

Any AI strategy starts with data collection, on an ongoing basis, alongside open source algorithm investigations, identifying the software building blocks and acquiring internal skills. It’s vital to introduce analysis protocols to climb the AI experience curve by building ‘training sets’. There’s no way of importing these from outside, you simply have to get to the heart of the issue to really analyse a situation. AI must be built on concrete data. Getting help from the experts, start-ups and universities can be useful, but the key to any approach lies in internal leadership.

How can artificial intelligence respond to the issues facing Bouygues Energies & Services?

For Bouygues Energies & Services, one of the key benefits of AI is being able to extract industry skills from data collections, reproduce analysis capacities and anticipate team experiences, all the while modelling and revisiting processes. In this way, ‘trouble-shooting’ approaches can respond to the issues facing Bouygues Energies & Services’ customers. AI as a facilitator of dialogue with customers, like chatbots, even though that can be tricky, is another area to look into. Finally, recommendation engines, adapted to your industries, could help improve the operational performance of your teams.