AI Develops New Theories Like Newton and Einstein
In a groundbreaking development, researchers at Forschungszentrum Jülich have created an artificial intelligence that can formulate physical theories by recognizing patterns in complex data sets, a task traditionally reserved for the likes of Isaac Newton and Albert Einstein. This AI, part of the “Physics of AI” initiative, simplifies complex interactions in data to generate new theories, setting itself apart from conventional approaches by making these theories explainable and grounded in the language of physics.
In a groundbreaking development, researchers at Forschungszentrum Jülich have created an artificial intelligence that can formulate physical theories by recognizing patterns in complex data sets, a task traditionally reserved for the likes of Isaac Newton and Albert Einstein. This AI, part of the “Physics of AI” initiative, simplifies complex interactions in data to generate new theories, setting itself apart from conventional approaches by making these theories explainable and grounded in the language of physics.
Understanding the "Physics of AI"
When it comes to developing a new theory, physicists typically rely on observations of a system to propose how its components interact, eventually leading to the explanation of observed behaviors. This process involves deriving new predictions and putting them to the test. For instance, Isaac Newton's law of gravitation not only explains the force of gravity on Earth but also accurately predicts the movements of celestial bodies.
The AI developed by Forschungszentrum Jülich takes a unique approach to theory formation. Instead of starting with general principles or basic equations, the AI utilizes a neural network to map complex behavior to a simpler system. By simplifying the interactions between system components, the AI then reconstructs the complex system to develop new theories. This perspective, termed the “physics of AI,” views the world through the lens of interactions governed by specific laws—a fundamental principle of physics.
Practical Applications and Computational Challenges
To test the AI's capabilities, researchers employed a dataset of black and white images containing handwritten numbers, a common benchmark in neural network research. The AI analyzed the substructures within the images, such as pixel interactions that contribute to the formation of number edges. This innovative approach not only simplifies complex data but also allows for the interpretation of learned theories in the context of physics.
While the computational effort required for this endeavor is substantial, the AI's ability to efficiently parameterize interactions enables the examination of systems with up to 1,000 components, paving the way for analysis of larger systems in the future. This computational trick enables the exploration of complex systems that would otherwise be impractical to study.
Setting Apart from Traditional AI Models
In contrast to other AIs that learn theories without interpretability, the AI developed at Forschungszentrum Jülich focuses on extracting and formulating theories in the language of system interactions rooted in physics. This approach, known as the “physics of AI,” bridges the gap between complex AI processes and human-understandable theories, enhancing the explainability and transparency of AI-generated insights.
In a study published in Physical Review X, researchers detailed their methodology and findings, showcasing the potential of AI in formulating and interpreting complex theories from data. By merging the realms of physics and artificial intelligence, this innovative approach opens new avenues for scientific discovery and problem-solving.
Reference: “Learning Interacting Theories from Data” by Claudia Merger, Alexandre René, Kirsten Fischer, Peter Bouss, Sandra Nestler, David Dahmen, Carsten Honerkamp and Moritz Helias, 20 November 2023, Physical Review X.
Hey, I'm Fateh Muhammad, a Lahore local with a passion for arts and politics. My journey led me through the halls of the National College of Arts, where I delved into the intricacies of both disciplines. Now calling Lahore home, I'm here to share my insights and perspectives on the dynamic intersection of art and politics. Let's embark on this enlightening journey together! Connect With Me .