Artificial intelligence has become an integral part of many people’s lives. You have very likely already used it, whether to draft an email, plan a holiday or search the web. In many cases, it is now hard to do without.
But beyond these “mainstream” uses, AI and the technologies associated with it have also taken hold in many technical, medical and scientific fields. Exploring these niche applications helps us better understand the real impact — present and future — of this technology on our lives, often in an indirect but decisive way.
Medicine
In the medical field, AI was integrated very quickly, particularly in radiology. AI systems are now able to detect suspicious anomalies in medical images, and also to predict the risk of developing certain diseases, such as cancer, based on clinical history and patient records.
These technologies are already delivering concrete results. At Saint Göran Hospital in Stockholm, the introduction of an AI that analyses mammograms led to:
- an 11.3% increase in the breast cancer detection rate,
- as well as a 21.9% reduction in false positives.
(Source: RTS)
Another striking example comes from McGill University, where researchers developed an AI model able to detect the spread of metastatic brain cancers from MRI scans. This model can identify the presence of cancer cells in the tissues around the brain with 85% accuracy, without surgical intervention.
Closer to home, research is also under way at the University of Geneva. An AI has been developed there to anticipate the risk of metastases and recurrences for certain cancers. This approach relies in particular on the analysis of gene expressions, with announced accuracy of around 80%. Source
A Swiss national initiative also aims to integrate AI at scale into cancer treatment, notably through the development of large language models designed to assist oncologists in analysing vast volumes of patient data. Source
Molecule simulation
You may never have heard of it, but one of the most spectacular advances made possible by AI concerns molecular biology, and more specifically protein folding.
Proteins are long chains of amino acids, comparable to strings of beads. The way these chains fold in space directly determines their biological function. For example, a protein shaped like a “lock” can bind to a specific molecule — the “key” — in order to trigger a chemical reaction.
The problem is that:
- a misfolded protein can no longer fulfil its function,
- and worse still, some misfolded proteins can aggregate and become toxic to the body.
This is precisely the mechanism involved in serious diseases such as Alzheimer’s, Parkinson’s or cystic fibrosis. Understanding protein folding therefore opens the way to new treatments.
Yet mathematically predicting how a protein folds is a problem of extreme complexity. A single protein can adopt an astronomical number of possible configurations, making any brute-force approach totally unrealistic. Until recently, scientists had to rely on long and costly experimental methods.
Thanks to deep learning, trained on immense databases of known protein structures, it has become possible to create models capable of identifying folding rules invisible to the human eye.
The most emblematic example is AlphaFold, a project developed by DeepMind (Google), which has profoundly transformed structural biology by making it possible to predict the structure of millions of proteins with remarkable accuracy.
Engineering – CAD
In engineering and computer-aided design (CAD), AI is also disrupting traditional methods through generative design (generative design).
Historically, an engineer drew each part by hand, relying on experience and intuition. Today, the process is reversed: the engineer describes the problem, and AI proposes the solutions.
In practice, the engineer defines:
- the maximum weight,
- the mechanical constraints,
- the attachment points,
- the permitted materials.
The algorithm then explores thousands of possible geometries and keeps only those that respect the laws of physics. Material is added where it is needed for strength, and removed elsewhere.
The results are often surprising: the generated parts have organic shapes, reminiscent of bones or natural structures. Above all, they are lighter, stronger and more efficient than their manually designed equivalents.
NASA, for example, uses generative design in its Evolved Structures project to create components for telescopes and space probes. These parts make it possible to cut weight by nearly two thirds while withstanding extreme constraints, and development time has gone from several months to a few days.
The automotive industry is also adopting these methods. Toyota, for example, has used generative design to rethink its seat frames. The result: a thinner, lighter structure that reduces the vehicle’s total weight while improving passenger space and comfort.
(https://www.autodesk.com/fr/design-make/articles/generative-design-seat-frame).
