5 areas of science where AI is already helping to make big discoveries

1. Story

Historians are already commissioning AI to study manuscripts. He copes with this task faster, besides, he does not see a problem in poor readability: the author's strange handwriting, yellowed paper or faded ink do not interfere with his work. At the same time, he is able to recognize not only words and sentences, but also the structure of the text - he observes the division into paragraphs, chapters and paragraphs.

An example of such cooperation between historians and AI is the Russian project "Digital Peter». The neural network is trained in the handwriting of Peter I and in a matter of minutes deciphers any handwritten texts of the emperor. Another similar tool is the Austrian platform Transcribus. It can recognize different languages ​​and handwriting, but first it requires calibration: the neural network studies a couple of pages of text, goes through several revisions, and then starts working accurately and quickly.

AI powers allow analyze large amounts of information: not only texts, but also various diagrams and drawings. Scientists can instruct the neural network, say, to find all the translations and expositions of one text in different books.

AI is also able to fill in the gaps in old documents and determine the time and place of their origin. Such platforms include Ithaca. For example, she clarified the date of creation of some ancient Greek decrees. Previously, it was believed that they were written in 446 BC. e., AI saw patterns pointing to 421 BC. e.

2. Medicine

AI in medicine accelerates the work of both doctors and scientists. He is the first to help with diagnosis: quickly studies screenings, searches for the necessary markers and gives an answer, which is then interpreted by specialists. Artificial intelligence in Moscow clinics use from 2020 to analyze x-rays, CT and MRI results.

It is likely that algorithms will soon be able to detect rare diseases as well. Similar mechanisms are already being studied. For example, researchers at Harvard Medical School created SISH tool that classifies different types of malignant tumors. As part of the experiment, AI studied about 22,000 images and quickly distributed them into more than 50 categories.

Scientists in artificial intelligence laboratories facilitates work on the development of drugs and vaccines. It calculates different combinations of active substances and reports the estimated percentage of their effectiveness. As a result, you do not need to spend years testing failed options in advance. It is already being actively used. For 2021 only to the US Department of Health came over 100 drug approval applications developed with AI.

One of the medical assistants in creating drugs is the AlphaFold neural network, built structure of more than 200 million proteins. Thanks to her work, scientists at the University of Oxford identified the structure of a key protein in the malaria parasite, which would help strengthen the vaccine against the disease. Previous studies using X-ray crystallography did not allow this.

AI also use to modernize gene therapy. In the future, he provide and faster comfortable study of the human genome. Scientists suggest that within a decade, research in this area will generate up to 40 exabytes (quintillion bytes) of data: for a person to process such a volume is an impossible task.

Digital technology experts, such as Tech Whisperer Limited founder Jasprit Bindra, also believe in the bright future of AI in medicine. At the educational marathon “Knowledge. The first" of the Russian society "Knowledge" he suggestedthat AI has a chance to revolutionize medicine, as penicillin once did, and become an indispensable assistant in the implementation of UN health programs. Also, according to Bindra, the fifth version of the GPT neural network language model, which will be released at the end of 2023, will cope with the interpretation of analyzes and the selection of treatment faster than doctors.

3. Physics

AI in physics has long been used to analyze big data. And he has a lot to be proud of. In 2012, machine learning models helped the staff of the European Center for Nuclear Research CERN open Higgs boson. The task of the AI ​​was to analyze the endless stream of signals from the Large Hadron Collider, look for signs of this elementary particle and mark them.

In the future, AI can simplify the solution of quantum problems. The proof of this is the work of researchers from New York: they created and trained an algorithm that shortened calculations of the Hubbard model from 100,000 equations to four. The accuracy of the calculations was not affected by this.

Another possible task of AI in the future is the search for new physical laws. To make this a reality, we need an algorithm that can determine state variables. And scientists at Columbia University have this happened. Their AI was able to independently guess what drives the pendulum and lava lamp, as well as why the fireplace is on fire. Of the inputs, the instrument had only video recordings. The variables proposed by artificial intelligence did not always coincide with those that physicists themselves were accustomed to. Scientists have come to the conclusion that AI has a chance to show people the previously unknown driving forces of nature and push them to new conclusions that are likely to change both science and our understanding of the world.

4. Astronomy

Galaxies, planets, stars and other space objects are huge in reality, but on large-scale photographs from a telescope they look like crumbs. It takes a lot of time to find them on your own. AI helps scientists to cope much faster. For example, the platform can analyze images from space Morpheustrained on frames from the Hubble telescope. AI detective skill will especially useful in the search for exoplanets, that is, celestial bodies that are outside the solar system.

Scientists at the Smithsonian Astrophysical Observatory are also using AI to hunt for short-term cosmic events like supernovae and monitor changes in the weather on the Sun. For the last task, the neural network has to collect 1.5 terabytes of information per day.

Scientists also use AI to create images of non-existent galaxies. It looks frighteningly realistic. NASA in 2021 laid out on his website a collage of 225 images, among which only one was taken by a telescope. It is almost impossible to find the original among fakes. But scientists need fake pictures and models not just to play pranks on non-professional space lovers. With their help, the neural network learns and tests hypotheses: they check how a space object similar to a projection will behave in different conditions.

5. Ecology

For environmentalists, artificial intelligence is primarily useful for its ability to collect and analyze data. For example, in 2022, UNEP (United Nations Environment Programme) launched an AI-powered digital platform WESP. Its algorithms collect information from different sensors around the world, analyze and visualize. And all this in real time. In particular, the instrument monitors the change in the mass of glaciers and the concentration of carbon dioxide in the atmosphere. In addition, WESP provides forecasts.

There are other AI tools operating within the UNEP ecosystem. Platform IMEO monitors methane emissions, and GEMS - for air pollution.

Artificial intelligence is able to simplify and control ecosystems. So, this year's machine learning program will help scientists from England monitor the plankton community around the clock. So they will check how these creatures are affected by environmental changes.