Breakthroughs in science made by AI in 2024: archaeological finds, conversations with sperm whales and protein folding

Artificial intelligence or machine learning has become a central topic in many news stories this year. And deservedly so, although there is a certain amount of advertising in all this. As the weeks ahead wrap up the year, CNN journalists came up with a warm-up article highlighting four unexpected breakthroughs in science in 2024, made possible by the use of artificial intelligence.

Appearance of a charred scroll that was read by AI. Image source: scrollprize.org

The first story told is about an archaeological discovery. Machine learning has helped read a number of fragments of a baked and petrified Roman scroll from Herculaneum, buried during the eruption of Vesuvius in 79 AD. Several such scrolls have been found, in every sense representing a brand, and not a document. There is no way to physically unfold them and read them. X-ray tomography and machine learning helped, which revealed patterns in the chemical structure of the sample and helped not only separate the layers of the scroll, but also track traces of burnt ink and begin to read the text.

The second story is that thanks to AI, the phonetic basis of the “language” of animals is becoming clearer. In the future, machine learning may help people learn to understand the “speech” of animals. Scientists started with sperm whales, which are known for their rather complex and structured system of sound communication. The AI ​​analyzed nearly 9,000 recorded messages from 60 sperm whales in the Caribbean.

The neural network detected 18 types of rhythm, five types of tempo, three types of duration variations, and two types of extra clicks added at the end of a code in a group of shorter codes. This is more than enough for communication of any complexity, but there is no decoding of all this yet. But thanks to the “words” found and examples of “phrase construction”, scientists intend to start talking to whales and see their reaction. The same practice can be applied to any animal.

The third story continues the theme of archeology. In the Peruvian Nazca desert, there are huge drawings on the ground – geoglyphs. They cannot be seen from ground level. Geoglyphs were discovered by accident from airplanes in the air. Usually these are stylized images of animals, people or ornaments. Until recently, 430 images were opened. The AI, trained on these examples, was able to discover another 303 previously unknown patterns, proposing 47,000 candidates that it considered necessary to highlight in the desert topography. As you can see, there are a lot of errors, which indicates the imperfection of the algorithm. But the almost doubling of archaeological sites found is definitely a stroke of luck.

The fourth story, perhaps, can leave the most significant mark in the history of mankind. This is a story about predicting protein folding. Proteins consist of two dozen amino acids, but there are hundreds of millions of combinations and resulting spatial forms of proteins. In order for a protein to react with a living cell of the human body, it must take on one or another 3D shape – it’s like picking up a key to a lock, one will open it, and the other is just a piece of unnecessary iron. Artificial intelligence in the form of the AlphaFold package from Google DeepMind has learned to analyze all proteins known to scientists and can predict new ones that have not yet been discovered. These are medicines and much more to improve people’s health.

And finally, the clearest recognition of AI in earthly science was the awarding of the Nobel Prizes in physics and chemistry this year to scientists who developed the corresponding neutron networks for discoveries in these sciences. The new instrument was in place and at the right time, predetermining the development of science for many years to come.