A CREEPY artificial intelligence has left scientists puzzled after it discovered physics that even professionals still don’t understand.
Physics is one of the more rigid disciplines in science with complex equations and exact measurements for secrets to be revealed.
Professor of engineering at Columbia University Hod Lipson said it is a task that has no well-laid path to follow.
“It’s an art, there is no systematic way,” he told Motherboard.
“It’s almost like, how do you discover the alphabet? It just happens organically.”
At his Creative Machines Lab, Lipson and his colleagues aim to better understand how this process of discovery happens and how it can be improved using machine learning.
The team developed an algorithm capable of studying physical phenomena by “watching” videos featuring the flicker of a flame or the swing of a double pendulum.
The algorithm was able to predict the correct number of variables within known systems and even make predictions for unknown ones.
The team published their findings last week in a study titled Automated discovery of fundamental variables hidden in experimental data in the Nature Computational Science journal.
Lipson said this work stands apart from other attempts to study similar data because it is the first not to provide the algorithm with any information on the number or variables in a system.
This means the system isn’t restricted to looking for variables through a human-only gaze, which Lipson said could be crucial for finding hidden physics within the systems.
“It’s not that people are toiling away day and night to look for these variables and this can expedite the process,” Lipson said.
“It’s more that we are probably overlooking a lot of stuff, but so much is hinging on those variables that we thought if we could throw some AI power at this, maybe we’ll discover things that are super useful and will change the way we think.”
The team, including the paper’s first author and assistant professor of engineering at Duke University Boyuan Chen, gave the algorithm videos showing dynamic motion.
The videos also included not-yet understood motions such as lava lamps and inflatable air dancers.
The AI attempted to model the phenomenon after studying the videos to create a list of increasingly smaller variables.
Then, it would give the minimum number of variables needed by the system to capture the motion accurately.
The AI was successful in discovering the correct number of variables but it currently lacks the language needed to describe what the variables are.
This will hinder it from entering science labs anytime soon, but Chen believes it’s not a huge problem for the time being.
“What we have right now is like a general framework,” said Chen.
“One thing that will be very interesting is to collaborate with experts who have data and an intuition about what that data is doing. What we want to do is to help them to discover what they do not know yet about the data.”
Lipson believes the algorithm could study systems beyond physics, such as disease evolution or climate change in the future.
The team hopes the algorithm will help communicate its findings more easily to humans as it could potentially be a great advance in scientific discovery.
“Humans have been doing this for 300 years, and it seems to me like we have kind of reached the end of what we can do manually,” said Lipson.
“We need something to help us go on to the next level.”