“Artificial intelligence is at the heart of the digital transformation of businesses.” It’s true. But once we have said that, we have said everything, and said nothing.
What AI (ML, computer vision, generative, trust, etc.)? For what: automate, create new products, optimize, augment humans? And with what data (while studies underline, one after the other, that companies are still struggling with their “data” assets)?
Data, carbon footprint and law
In this issue, all the articles revolve, in one way or another, around the AI planet… and data management. Since one does not go without the other. “Without data, no AI”, one could say.
This issue also reminds us that AI is not just a technical problem. It has become a legal issue (with the European AI Act).
And tomorrow, it will also become an ecological problem.
Will AIs consume less energy, water (data center cooling) and rare metals (for GPUs and CPUs) than they save by optimizing processes and resource management?
The Olympic swimming pool and the garden cover
Time will tell. But in any case, it will first be necessary – it is our opinion – to put an end to the “beliefs” and the confusions. And there are many of them.
One of the most important to defuse is the belief that AI – generative or not – can magically solve problems on its own.
Little anecdote on this point. In 2021, Bercy launched a project to automate the detection of undeclared swimming pools, by massively analyzing satellite images (from Google), using computer vision algorithms. The thing works wonderfully, and since then, more than 140,000 swimming pools have been identified and regularized.
The project gave rise to exciting feedback at Big Data & IA Paris. Its managers insisted: there was no question of automating the chain from end to end, even if this was perfectly technically possible (detection of a swimming pool, information cross-checked by the land registry, sending of the regularization letter, recovery).
And great good has come to them. Because no algorithm is infallible.
Those at Bercy are much more efficient (and faster) than humans in knowing whether a blue spot seen from the sky is an above-ground swimming pool (not to be declared) or a solid swimming pool (to be declared). And yet. One of my close friends had the unpleasant surprise of receiving, for his little house in the Pyrenees, a letter concerning an Olympic-sized swimming pool. He had, in fact, covered his fallow land with tarpaulins – blue in color – to prevent weeds from proliferating.
The FISC obviously did not follow up.
Does this anecdote seem trivial to you? Maybe. You wouldn’t have let an AI manage everything either? So much the better ! But BCG X recently confirmed to MagIT that this type of belief (and others) is very much alive. And that we had to fight them to avoid a backlash, disillusionment, or an AI winter.
Summary this semester
This issue of Applications & Data is part of this logic. You will find testimonials from Decathlon, URSSAF, Etam (three great digital transformations), an application of AI in battery recycling, advice (and a concrete dive into ACT AI) as well as a big interview on how to limit the impact of your AI (generative, in particular).
Very good reading to you.
Download for free
Applications & Data 24
