An emerging technological phenomenon of increasing importance in human society, artificial intelligence (AI) has revolutionized a large number of fields, from medicine to the economy, including the army and road traffic. But it also plays a crucial role in science, where it stands out today as an essential tool for scientists.
Artificial intelligence, in all its forms, has spread in many scientific sectors, to become an invaluable resource for researchers. Endowed with enormous calculating and predictive potential, AI is today the main research tool in a large number of laboratories and scientific institutions. From the detection of exoplanets to the diagnosis of complex pathologies, digital brains continue to shine with their prowess.
The search for new particles is the main objective of particle accelerators, and constitutes an extremely active field of theoretical and experimental physics. Physicists involved the first AI models in this particle hunt from the end of the 1980s, through the use of machine-learning algorithms (machine learning) serving as the basis for artificial neural networks.
The millions of collisions within particle accelerators give rise to the recording of a very large amount of data, which must then be carefully studied. This activity of analysis is the most important part of the work of physicists, who strive to identify subtle patterns and recurrences in collision products, with the aim of highlighting new energy signatures.
It was this work that enabled LHC scientists (Cern) to identify the Higgs boson, observing an energy range corresponding to theoretical predictions and simulations.
The signatures of these particles are rarely easily identifiable and are usually found buried in the background noise and energy of the decay products. For example, at the LHC, a Higgs boson is produced every billion proton-proton collisions and, in a billionth of a picosecond, decays into other particles such as photons and muons. As a result, physicists must reconstruct the pattern leading to the Higgs boson, consistently identifying each particle involved in the phenomenon.
This is where neural networks come into play. Within a particle detector, particles such as photons, electrons or hadrons are at the origin of a cascade of sub-particles flowing into an electronic calorimeter. This cascade of subparticles differs very subtly depending on the particle that originated it. machine-learning algorithms have such a capacity for analysis and discrimination that they are able to precisely identify the source of the cascade, and discriminate whether a pair of photons comes from the disintegration of a Higgs boson or not.
While extremely useful, AI has not replaced physicists, who continue to rely heavily on their understanding of theoretical models to identify anomalies in particle spectra. But AI will increasingly prevail according to Paolo Calafiura, analyst at Berkeley National Laboratory (California). In 2024, scientists want to multiply the number of collisions by 10 at the LHC; at that time, AI will become vital.
The advent of social networks has paved the way for the exchange of hundreds of millions of daily communications. Such vast amounts of otherwise easily accessible data have provided scientists with the ability to employ artificial intelligence to study these communications and draw conclusions, according to psychologist Martin Seligman.
At the Pennsylvania Center for Positive Psychology, Seligman and more than 20 psychologists, doctors, and data scientists involved in the World Well-Being Project use neural networks and natural language algorithms to analyze this data and gauge the physical and mental health of populations.
This kind of analysis is usually done through surveys and questionnaires. But the growing amount of data available directly from communications—along with the ease of access and low cost of the process—provides scientists with the means to combine huge amounts of data in a short time. Although such an agglomeration of data often turns out to be chaotic, AI is a powerful discrimination tool to reveal individual or collective behavioral patterns.
In a recent study, Seligman and his colleagues studied 29,000 posts by Facebook users who admitted to an overall state of depression. Using data from 28,000 of them, a neural network identified a match between the words used in the posts and the severity of depression. He was then able to successfully determine the severity of other users' depression based solely on the words used.
In another study, the algorithm correctly predicted the rate of… (continued on next page)
