Anthropologist Rodrigo Ochigame studies how AI is changing scientific research practices. From astrophysics to mathematics to climate science, we're seeing the adoption of new AI models raise questions about what constitutes reliable scientific evidence.
In 2019, the world witnessed its first ever image of a black hole: an eerie shadow surrounded by a fiery orange halo. In 2023, a team of scientists published a “sharper” image based on the same data, now using AI techniques. The data collected from his eight observatories on Earth is so noisy and limited that scientists have to develop computer algorithms and AI models that make many assumptions to generate images. . Depending on the algorithm and model you choose, your results may vary widely.
Anthropologist Rodrigo Ochikame studies how AI is changing scientific practice in a variety of fields, including black hole imaging. In addition to interviewing scientists, Ochikame created another black hole image based on the same data. By experimenting with different algorithms and models, Ochikame investigated what the image of the black hole would look like if scientists had made different choices. More importantly, the adoption of new AI models raises questions about what should and should not count as reliable scientific evidence.
Besides black hole imaging, Ochigame is researching other scientific fields that use AI in new ways. For example, explore the work of mathematicians who use AI to discover and prove new theorems, and environmental scientists who use AI to predict the effects of climate change on complex ecosystems. All these cases he has one thing in common. That means scientists can't build traditional AI models that simply reproduce patterns found in existing datasets. “I am interested in areas where traditional 'ground truth' data is not available,” they explain. “No reliable image of a black hole exists to date, and climate models need to account for potential extreme scenarios that have never occurred before.”
They're looking over their shoulders at the scientists who are using AI in their research and asking important questions. are they okay with that?
“Sometimes people are hesitant to accept anthropologists into their midst. But luckily, I am welcomed by the people I study. One reason is that I I try to contribute to their discussions and sometimes even build computational models myself. are questions that scientists themselves consider unanswered. They too want to find answers. And scientists are very open-minded and open to hearing anthropological insights into their field. ”
What are the limits to the use of AI when conducting scientific research?
“It's important not to treat the results of an AI model as conclusive evidence without first questioning how the model works and where the data comes from.” For example, “More Vivid.” The black hole image was generated by a machine learning model trained using images from the simulation. Only after the model is trained in simulation is it applied to real observed data. This is an interesting phenomenon. Scientists increasingly use simulations as training data, as if they were the so-called “ground truth.” I'm not opposed to this type of research, but I think the results should be interpreted with caution. ”
Is science currently relying too much on the use of AI systems?
“I think so.” There are many examples where the application of AI is not so important. Time and time again, we see scientific claims that are not justified by the data or algorithms used. Scientists often have incentives to expand their claims. There are so many applications of AI that I think are completely unwarranted, so I prefer to study more complex cases. What interests me most are cases where it is difficult to formulate an opinion.
Previously, you were thinking about developing alternatives to the computational models that are now considered fundamental in many scientific fields, such as computer science.
This was my PhD research that questioned the universality of the most commonly used formal models of computing and AI, including mathematical logic, Turing machines, game theory, and neural networks. . We found that before these models took off, researchers around the world were questioning some of their most basic assumptions. And some researchers have developed their own alternative models. For example, Brazilian mathematicians developed logical systems that tolerate partial contradictions, and Indian scientists developed non-binary models of computation. It's not necessarily the goal of my work for people to adopt these unconventional models and implement them right away. But sometimes it's nice to see people get inspired and incorporate those ideas into their work in unexpected ways. ”
Some of Rodrigo Ochigame's alternative images of black holes are currently on display at the Boerhave Rijksmuseum in Leiden as part of the exhibition “Toward a Black Hole”.
Header image credit: First M87 black hole image from Event Horizon Telescope, 2019 (left). M87 black hole image based on the PRIMO machine learning algorithm by Lia Medeiros et al., 2023 (center). Black hole accretion simulation by Hotaka Shiokawa (right).
Text: Jan Joost Aten
