Outstanding scientific discoveries thanks to AI in 2024

Decoding the ancient blackened scroll

The Herculaneum Scrolls are a collection of about 800 Greek scrolls that were carbonized during the volcanic eruption that buried the ancient Roman town of Pompeii in 79 AD. Because they were blackened and brittle, they would crumble if tried to open them and suddenly Any remaining traces of writing are nearly impossible to read. To date, they have not been opened, but thanks to AI, scientists can now decode part of the content.

Using AI and high-resolution X-rays, Youssef Nader, a PhD student in Germany, Luke Farritor, a SpaceX intern in the US, and Julian Schilliger, a Swiss robotics student, decoded more than 2,000 symbols. from the scrolls in 2023. They became an unprecedented treasure trove of information about ancient Rome and Greece.

Scientists have launched the Vesuvius Challenge, a competition to speed up the decoding process, hoping that 90% of the four scrolls’ contents will be cracked by the end of 2024. The biggest challenge is to open them. in the virtual world, differentiate black ink from carbonized paper to make Greek and Roman letters easier to read.

“AI is helping us enhance our ability to read inkblots. The inkblot is still there. It’s just buried and disguised in complexity, and AI helps distill and condense,” Brent Seales, professor of machine science computer at the University of Kentucky, said.

Decoding the language of sperm whales

Sperm whales make clicking sounds thanks to the spermaceti organs on their heads. Researchers know that these clicking sounds vary in tempo, tone, and length. However, what they want to convey with that sound is still a mystery to humans.

Now, machine learning has helped scientists analyze nearly 9,000 clicking sound sequences, called codas, representing the voices of about 60 sperm whales in the Caribbean Sea. This work could allow humans to communicate with marine animals in the future.

Scientists examined the timing and frequency of codas in individual sounds, choruses and even dialogues of sperm whales. When visualized with AI, previously unseen coda patterns emerged and were described by the team of scientists as phonetic analogs of human communication.

In total, the program detected 18 types of melodies (sequences of spaces between clicks), 5 types of tempo (duration of the entire coda), 3 types of rubato (variations in duration) and 2 types of tones. point – click added to the end of a coda in a group of short codas.

The above elements can combine to form a “huge treasure of phrases”. This method can also be applied to the sounds of many other animals, according to Brenda McCowan, a professor at the University of California Davis School of Veterinary Medicine.

Search for giant ancient carvings

AI is also helping to speed up the search for giant carvings in the Nazca desert, Peru. Usually only seen from above, they depict geometric designs, human-like characters, and even strange figures like knife-wielding killer whales.

The research team of Masato Sakai, professor of archeology at Yamagata University, Japan, trained an object detection AI model using high-resolution images of 430 known Nazca symbols as of 2020.

During September 2022 – February 2023, the team tested the model’s accuracy in the Nazca desert, surveying potential locations through walking trips and drones. In the end, they authenticated 303 drawings, almost doubling the number of known drawings in just a few months.

Although not yet completely accurate, AI has the potential to make a major contribution to archaeology, especially in remote and harsh terrain like deserts, according to Amina Jambajantsan, a data scientist from the Institute’s archeology department. Max Planck archeology in Jena, Germany.

Learn the basic components of life

AI is also helping researchers understand life at the smallest scale: the chains of molecules that make up proteins – the basic components of life. Proteins are made up of only about 20 amino acids, but they can combine in almost limitless ways, folding themselves into complex patterns in three dimensions. Not only do they help form hair, skin, and tissue cells, they also read, copy, repair DNA and transport oxygen in the blood.

For decades, decoding these 3D structures has been a challenging and time-consuming task. But in 2018, a revolutionary AI tool emerged. The new version of the AlphaFold Protein Structure Database, developed by Demis Hassabis and John Jumper at Google DeepMind London, predicts the structures of nearly all 200 million known proteins from amino acid sequences.

Trained on amino acid sequences and protein structures, this database functions like a Google search engine. It provides access to predicted protein models, helping to accelerate advances in basic biology and several related fields, including medicine.

AlphaFold is just one of many AI tools being deployed in biomedicine. Machine learning is helping to promote the compilation of “maps” of each type of cell in the human body and the discovery of molecules from which to create new drugs to serve humans.