Artificial intelligence has deeply penetrated many scientific fields – proteins for drug development are created in a matter of seconds , thousands of new galaxies are found in one fell swoop , droughts and floods are predicted long ago. Although AI cannot replace a scientist, technology can definitely change our understanding of phenomena that are obvious at first glance.
Not only living organisms are capable of recognizing images
Vision is one of the main channels of information perception for people and animals. The human eye can recognize thousands of shades and textures, representatives of the cat family are able to see in the dark, and even a deep-sea squid navigates the ocean floor in the darkness with the help of its eyes.
In addition to the ability to “see” this or that object, the ability to recognize it is very important. Dogs, for example, can distinguish people and other animals from members of their own species, even on television screens. Until 2012, no one in the scientific community believed that a computer could be taught to do the same thing.
That year, at one of the key conferences on the topic of AI, NeurIPS, a group of scientists presented a neural network that learned to recognize pictures well. It was named after the first author, Alexander Krizhevsky, “Alex Network,” that is, AlexNet.
The researchers trained a large deep convolutional neural network to classify 1.3 million high-resolution images. The pictures were taken from the giant ImageNet database of described images. Largely thanks to this work, the topic of computer vision began to develop so actively that in 2024 we will be able to unlock phones with our faces.
By the way, ten years later the same thing happened with the understanding of texts. ChatGPT was born, bringing complex models closer to end users, and the world became interested in the nature of AI technologies.
Logic is not only subject to humans
For a long time, mathematical problems were solved by computer systems using human-specified conditions—akin to a calculator. Nobody believed that complex logical conclusions were available to them. For example, solving geometric problems or searching for proofs of theorems. Asterisk problems were an integral part of the search for young mathematical talent. For this purpose, an international mathematical Olympiad was organized, during which schoolchildren competed for years to find answers to the most ornate problems. No other system could boast the same results.
In January 2024, a team of Google researchers introduced AlphaGeometry. A real breakthrough. This artificial intelligence system solves complex geometric problems at a level approaching that of a human Olympic gold medalist. You can read a publication about the program in the journal Nature .
In a comparison test, out of 30 geometry Olympiad problems, AlphaGeometry solved 25 within the standard Olympiad time. For comparison, the previous modern system solved only ten such geometry problems, and the average gold medalist solved 25.9 problems.
Climate does not always affect the size of plant leaves
There is a rule in botany: in a humid environment like a jungle, a plant’s leaves will be much larger than those of the same plant in a dry climate, like a desert. In other words, temperature and the amount of precipitation necessarily affect the size of the leaves.
It turned out that this rule only works within genera, but not within species. Maples in general throughout the planet will actually have more leaves at the equator and fewer at the poles. But in Norway maples, size rather depends on the exchange of genes with other populations. Australian researchers found this out a year ago: using computer vision, they selected and analyzed several thousand specimens from the National Herbarium of New South Wales.
This discovery is useful not only for agriculture – it can form a new perspective on the evolution and adaptation of plants, and preserve rare species even taking into account climate change.
The extinction of species is not followed by their growth
We know about global warming, and cataclysms such as meteorite falls and volcanic eruptions periodically occur on Earth – as a result of all this, over tens of thousands of years, biodiversity can decrease by 95%. At least five mass extinctions are known – and a sixth, man-made, is happening right now due to hunting, deforestation and environmental pollution.
Biologists believed that after each such extinction a boom occurs – just as after dinosaurs the planet was populated by mammals, so in other cases empty ecological niches are quickly filled by new species.
In 2020, this important evolutionary theory had to be revised: using machine learning, British and American scientists proved that there is no connection between extinction and reproduction. To do this, the program analyzed more than a million descriptions of fossils of 170 thousand species. The article did not become particularly cited in its subject area, but was still liked by fans of popular science news.
Fingerprints aren’t that unique
Using fingerprints, people log into banking applications, open doors, and identify criminals. And everyone is convinced that they cannot coincide – even if you compare, say, the index fingers from both hands of one person. And that it is impossible to say for sure whether a set of prints belongs to the same person.
Most likely, this approach will have to be reconsidered: there is no evidence that all prints are truly unique. On the contrary, in January, the opposite data appeared – “intrahuman” prints are extremely similar, especially for paired fingers.
Forensic scientists did not notice this because they compared the length of the lines and the places where they branched, and the AI found a new marker – the curvature and inclination of the curls. In other words, visually the prints may be completely different, but in fact have the same “owner”.
Scientists from Columbia University found this out. They trained the AI model on half a million artificial and 60 thousand real fingerprints – several per person. In most cases, the system was able to correctly match different fingerprints of the same people.
From a technology point of view, the AI model used is quite simple and unremarkable. But the results obtained thanks to it can help increase the efficiency of forensic medical examination by almost two orders of magnitude.
And this, in turn, will simplify the search for connections between crimes and speed up the exclusion of innocent people from judicial actions. Owners of gadgets will only need to scan one finger and then open the device with any of them.
What’s next?
These are just a few examples of how AI helps scientists refute axioms – and ultimately improve life and move towards new scientific discoveries.
How else can researchers use technology to their advantage? First, to find patterns in big data. Several thousand analytical articles alone are published every day – a person cannot comprehend so much information. Secondly, to formulate specific hypotheses – “this vaccine will work on a person like this”, “this cell will react to this irritant like this”.
AI models are not yet ready for autonomous operation; the systems are developing, but so far they have many prejudices borrowed from people. However, the situation will probably change in the coming years, and sometime in the 2050s, scientists may receive a Nobel Prize for answering the question about the origin of the universe, obtained together with AI.
