Tracing what AI does
This major stake we should not miss

Turning budget constraints into strengths

Artificial intelligence models, particularly generative models, consume a lot of energy. Do you think this is sustainable in the long term?

No. The main environmental challenge is model training, as this activity accounts for the highest electricity consumption and therefore most of the pollution generated by the digital sector.

In Europe, we don't have the same budgets as the United States or China. This forces us to think differently. For example, we are exploring avenues such as smaller, more specialised models and local data processing through edge computing. It is this more efficient approach to artificial intelligence that also allows us to use less data and less energy. What was initially a constraint (a smaller budget) has become a strength.

The use of renewable energy is also an important way to reduce the carbon footprint of artificial intelligence. However, these sources are not always stable or predictable, and it is necessary to be able to compensate for variations. The availability of electricity has become a strategic issue for investors. In this context, France has a real advantage with its nuclear power plants and long-term energy policy. This is very important, especially for companies like Mistral, which need to train their own models in Europe and have high energy requirements.

Are innovation and ecological transition compatible?

In your opinion, can AI be used as a tool to accelerate the ongoing ecological transition? Can technological innovation and climate imperatives be reconciled?

I think so, provided it is used properly. In supply chains, for example, AI makes it possible to quickly simulate disruptions, adapt logistics routes, anticipate price increases or customs constraints and adapt as best as possible, almost in real time. For example, we have a customer in the maritime transport sector, and in their case, if there is a blockage on a route in the Red Sea, AI can be used to optimise delivery routes in specific ways, thereby avoiding a disruption in the chain. Another specific example: the current uncertainty surrounding US customs tariffs is forcing companies to recalculate their production and distribution areas in real time. Here again, AI is decisive and helps to make the most optimal decisions.

I believe that we can reconcile technological innovation and climate imperatives, provided that we do not replicate the same models as elsewhere. In Europe, we will not win the race for “raw” power. However, we can invent other paths that are more sober, more efficient and more frugal. Artificial intelligence does not need to be massive to be useful. Its proper use is far more important than its size. Especially since, globally, we have more or less used all freely available data to train AI models. Technological innovation in artificial intelligence will instead come through efficiency.

Towards sensory and autonomous intelligence

In the future, AI will no longer be limited to generating text or images. It will interact with the physical world. What does this mean in specific terms?

If we take the mimicry between artificial intelligence and humans even further, the next step for these tools will be to learn as we do: that is, using all of our senses. This will require robots and AI to be coupled with sensors (sight, sound, smell, touch). By acquiring context, artificial intelligence will become even more relevant and its responses will be closer to our expectations.

Some AIs could develop their own language. Is this a plausible scenario

If we think about it, our languages are too complex for them. Too many silent letters, too many syntax rules... If they want to go faster, straight to the point, they could very well exchange data in an optimised format by adopting their own language to communicate with each other. For example, I want to obtain a presentation of my company's financial situation with proposals for future prospects. I will choose to entrust the various tasks to several AI tools, aiming for the best trained for each of the different tasks. The first AI will analyse the figures; the second will be better at drawing up a strategic plan and the third will be excellent at producing a nice presentation of all this with graphics and text. But how will these three AI tools “talk” to each other to exchange information while performing the requested task? It is at this stage that a language ex nihilo can be born.

So you fear a lack of transparency in AI chains?

Yes. The risk is losing traceability of decisions, since no human will be able to “read” the AI's thought process that led to the final result. How can you control something you don't understand? And who guarantees that the original intention is being respected?

No break planned in R&D

If you work in digital development on a daily basis, all it takes is a two-week absence for the landscape to change. This accelerated pace is forcing companies to be cautious. Why invest in an innovation that will be obsolete in a few months, or at least no longer the innovation at the heart of all discussions?

Artificial intelligence may reach a plateau in its technological evolution. If resources – data, energy, servers – are no longer available, we will have to adapt and make better use of existing technologies.

But I don't believe there will be a global slowdown. As long as there are challenges, there will be solutions. And the next big breakthrough will certainly come from a different type of architecture, one that is more economical and more rational.

In any case, I don't believe there will be a pause.

What is SAP Labs France ?

SAP is Europe's leading enterprise software publisher, founded in Germany in 1972. Although little known to the general public, SAP is involved in 76% of global transactions, working behind the scenes. SAP Labs France in Sophia Antipolis is one of the group's twenty R&D centres. Its engineers develop solutions that are used worldwide. Artificial intelligence is integrated into SAP software to automate, improve reliability and speed up processes such as human resources management, supply chains and business expenses.

A European leader with diverse roots

Born in Ukraine, educated in England, and based in Sophia Antipolis since the late 1990s, Helen Kushakovska embodies a unique career path in the tech world. She grew up in the Soviet Union, where women already held positions of responsibility. “Where I came from, it was normal to see women running factories, driving tractors or teaching in laboratories.”

When she arrived in France, she discovered a more patriarchal environment. “People asked me why I wanted to work when I had a husband and children. As a Ukrainian woman with a degree, I was shocked by this question!” She then encountered deeply ingrained stereotypes. “I always had to justify my skills and my legitimacy,” although she was relatively spared from clichés due to the international environment at SAP Labs France. Today, she campaigns for better representation of women in technical professions.