How will AI change astronomy?

Author: Yi Zhenping

**Source:**National Astronomical Observatory of Chinese Academy of Sciences

Things are mixed, congenitally born. Lonely! Independence does not change, Zhou Xun does not perish. – “Tao Te Ching”

Picture 1. Staring at the universe (generated by Unbounded AI tool)

Since ancient times, human beings have had endless fantasies about the vast starry sky, and are constantly on the journey of exploration. In recent years, with the rapid development of science and technology, human beings have accelerated the pace of exploration of the universe. Using advanced equipment and technology, astronomers have revealed amazing phenomena such as gravitational waves, black holes, dark matter and dark energy, which have greatly improved our understanding of the universe.

Since 2011, artificial intelligence (AI) technology has experienced rapid development, successfully bridged the technological gap between science and application, and achieved major breakthroughs in image classification, speech recognition, knowledge question answering, driverless driving, etc. The breakthrough has entered a period of vigorous development. In the field of astronomy, the application of AI is also accelerating year by year. Through the statistics of papers involving AI in astronomy research on arXiv, it is found that the number of papers has increased from 46 in 2013 to 2,328 in 2022, an increase of 50 times in ten years.

Figure 2. The number of AI articles published on arxiv physics in the past ten years (statistical results of searching machine learning, deep learning or neural network keywords in the title)

Astronomy has entered the era of multi-band, large samples and high information content, and the amount of data generated by telescopes and other observation instruments has become a truly “astronomical figure”. AI technology is more and more widely used in astronomy. Its algorithms have evolved from traditional machine learning to deep learning, and its analysis tasks have ranged from simple to complex. With the continuous improvement of neural network structure, the ability of AI model in feature extraction and representation has been significantly enhanced. We have now been able to directly identify very faint objects such as low-face brightness galaxies (shown in Figure 3) and L dwarfs (shown in Figure 4) from photometric images. These objects are some of the faintest galaxies and stars that can be discerned in visible-light images.

Figure 3. Use AI to search for galaxies with sideways low brightness (Source: Xing Yongguang, Yi Zhenping, Shandong University)

Figure 4. Using AI to search for L dwarfs (Source: Cao Zhi, Yi Zhenping, Shandong University)

The birth of ChatGPT marks a major breakthrough in AI and is regarded as a milestone in triggering a new round of AI revolution. ChatGPT is the result of the combination of big model and big data. In the field of astronomy, a large amount of observation data has been accumulated. With the support of algorithms and computing power, astronomy will also enter the era of AI revolution, which is likely to bring subversive changes.

The established large data analysis model can extract richer and more accurate information from astronomical big data. For example, it can automatically detect and classify various celestial bodies directly from images, automatically fuse data from multiple bands, provide reliable celestial body attribute parameters, and automatically generate statistical charts.

Figure 5 shows a framework for image captioning tasks, consisting of a convolutional neural network (CNN) and a recurrent neural network (RNN). The model can identify objects in images and describe the content of the images in words, that is, through the AI model, the computer has the ability to “see pictures and talk”. This kind of task involves two different information modes of image and description text. How to make the AI model make full use of multi-modal information and interact is also a trend of future research.

Figure 5. An image description task framework

With the help of image description and other technologies, astronomical images can be automatically analyzed, and feedback results can be provided in direct and friendly ways such as text, images, and voice. Figure 6 shows a schematic diagram of automatically analyzing an image from the Dark Energy Camera Heritage Sky Survey and generating text feedback.

Figure 6. Use the AI large model to automatically identify celestial bodies in astronomical images, and generate text descriptions and star catalogs. (Source: Bi Mingxuan, Shandong University)

AI will also help accelerate the discovery of new phenomena and new laws. Using AI algorithms, it is expected to efficiently identify trends and correlations hidden in astronomical big data. For example, the use of AI algorithms can quickly identify weak signals related to exoplanets, so as to find intelligent civilizations similar to Earth civilizations. This is the potential existence in the Milky Way calculated according to the Drake equation. Such discoveries will provide us with important clues to unravel the mysteries of our universe.

Figure 7. The Drake equation used to estimate the Earth-like alien civilization in the Milky Way

In addition, AI will also be widely used in the automated observation of telescopes, which can identify celestial objects of interest in the telescope in real time and help determine priorities for further observations. This automation allows astronomers to monitor the sky efficiently and respond quickly to transient events.

We look forward to creating an astronomical AI model that is smarter than ChatGPT in the near future, thereby freeing astronomers from the tedious data processing tasks and enabling them to focus more on scientific research. At that time, the first thing for astronomers after going to work every day may be to drink coffee while listening to the AI assistant report work, and obtain the analysis results of the AI assistant in all directions through sound, images or videos.

Let us look forward to this wonderful era together!

Figure 8. Schematic diagram of the future AI assistant (Source: Internet)

About the Author

Yi Zhenping, associate professor of Shandong University, master supervisor, head of the astronomical big data team of Shandong University. Committed to the interdisciplinary research of astronomy and computer science, applying AI algorithms to improve the automatic and intelligent analysis capabilities of astronomical data.