Using AI, historians track how astronomy ideas spread in the 16th century
Novel analysis of 76,000 textbook pages shows how widespread astronomy education in Europe was
This is a random selection of digits extracted from scanned textbooks and used to train an AI as part of a project to see how widespread astronomy education was in the 16th century. The task was challenging, because different printing presses produced different-looking numerals.
O. Eberle et al/Science Advances 2024
Historians working with an artificial intelligence assistant have begun tracking the spread of astronomical thinking across Europe in the early 1500s.
The analysis contributes to challenging the “lone genius” idea of scientific revolutions. Instead, it shows that knowledge about the positions of the stars was widespread and used in a variety of disciplines, researchers report October 23 in Science Advances.
“We can see here the first formation of a proto-international scientific community,” says computational historian Matteo Valleriani of the Max Planck Institute for the History of Science in Berlin.
Valleriani and colleagues used AI to examine a digitized collection of 359 astronomy textbooks published from 1472, less than 20 years after the first printing of the Gutenberg Bible, to 1650 (SN: 5/31/05).
These textbooks were used to teach introductory classes on geocentric astronomy — the view of the cosmos that places Earth at the center and moves outward in sequential spheres. Knowledge of the positions of the stars was thought to be important for studying everything from medicine to Greek and Latin poetry, so intro astronomy classes were mandatory for all students. Among other things, students learned to use the position of the sun in the constellations of the zodiac to figure out the date of an event that happened in antiquity, before standardized calendars were common.
Studying these past texts can give historians an idea of the background knowledge most educated people had about the universe and how that understanding changed over time.
The dataset included 76,000 pages of text, images and numerical tables, many with different fonts, formats and layouts. A historian might be able to analyze a handful of books in a single career. But Valleriani and colleagues wanted to study all of them.
“What we wanted to know, in general, is what the students were learning in astronomy over these 180 years and all over Europe,” Valleriani says. “This was humanly impossible.”
The team used machine learning to identify 10,000 separate numerical tables in the textbooks. Next, they trained an AI model to recognize individual numbers in the tables. “This was extremely hard, because the tables are not formatted in the same way,” says physicist and machine learning expert Klaus-Robert Müller of the Technical University of Berlin. “Everything is quite a mess.”
Once the AI had extracted all the numbers, it compared the different tables one by one and highlighted similarities and differences. For example, some textbooks were basically reprints of an earlier edition, and their tables were almost identical. Others introduced new ideas or new ways to use astronomical data.
The AI couldn’t tell the researchers what those similarities and differences meant (SN: 8/2/24). But it could give them a place to look for trends or moments of change.
“It’s moving from AI being used as a tool, to help do something I preconceived, to using AI as a team member, suggesting new solutions that I couldn’t see,” Valleriani says.
A common story about astronomy in this time period is that individual heroes of science, like Copernicus, Galileo and Kepler, shook the world by showing that Earth is not the center of the universe.
But historians of science have been moving away from the idea that science is driven by such lone geniuses making big discoveries (SN: 3/5/16). Those discoveries had social, political and cultural contexts, and they had to be disseminated into the wider culture somehow.
“When you deal with the scientific revolution, the triumph of the Copernican worldview, we know the big names,” says computational scientist Jürgen Renn of the Max Planck Institute of Geoanthropology in Jena, Germany, who was not involved in the new work. “But in Europe, this was a broad movement. There were many participants.”
One of the team’s major findings is that textbooks printed in Wittenberg, Germany, in the 1530s were widely imitated elsewhere in Europe. Similar books that were sold in cities with bigger markets, like Paris and Venice, created a new, homogeneous approach to astronomy.
Valleriani finds this ironic. Wittenberg is best known for being the city where Martin Luther kick-started the Protestant Reformation, which split a new branch of Christianity off from the Catholic church.
“It sounds paradoxical,” Valleriani says. “While Wittenberg and the Protestant Reformation was dividing Europe … and creating the background against which wars came out, at the same time, Wittenberg was able to develop a scientific approach at the educational level that was in truth taken over everywhere.”
There are limitations to this kind of research, the team points out. Historical data are always incomplete, and historians have to choose a subset of that data to focus on. AI can’t account for that sort of selection bias. Human historians must always be part of the process, the researchers stress.
This work “shows how historians can in the future deal with artificial intelligence methods, and cleverly use them without this utopian or dystopian illusion that they can do the work for you,” Renn says. “They’re just a fantastic new tool that helps us understand history as a broad stream of human actions and human thinking, and not just a string of singular events.”
