Groundbreaking AI platform from Microsoft accelerates scientific discoveries

The Microsoft company presented at the Build 2025 conference a new AI platform called Microsoft Discovery, designed to accelerate scientific research and discoveries. The platform will allow scientists to work in collaboration with a team of specialized AI agents, who will assist them in all stages of the scientific process - from data analysis and hypothesis formulation to experimental plan generation, modeling, and result analysis.

According to Microsoft, the use of the latest advancements in artificial intelligence and supercomputers will significantly increase the speed, scalability, and accuracy of scientific research. This is not an isolated case - in 2025, Google has already introduced its "AI co-author" to assist scientists, and companies like Anthropic, OpenAI, FutureHouse, and Lila Sciences also claim the potential of AI to accelerate scientific discoveries, especially in medicine.

However, many researchers still view the use of modern AI systems in science with skepticism, mainly due to their unreliability. Creating an effective "AI scientist" faces challenges related to the unpredictable number of influencing factors. While AI can be useful for searching through vast amounts of data and narrowing down the range of potential solutions, its ability for unconventional thinking and groundbreaking discoveries remains questionable.

The results of applying AI in science so far have been contradictory. For example, in 2023, Google reported the synthesis of around 40 new materials with the help of its AI GNoME, but an independent analysis showed that none of them were truly new. Similarly, companies like Exscientia and BenevolentAI, using AI in drug development, have faced resounding failures during clinical trials.

Microsoft hopes that its Discovery platform will achieve better results than its predecessors. The success of Discovery will largely depend on the platform's ability to account for the complexity of the scientific process and overcome the limitations of current AI algorithms.