⇧ [VIDÉO] You might also like this partner content (after ad)
What if whole swathes of the universe were opaque to us simply because we didn’t observe them in the right way? Engineers decided to “query” a machine learning system for certain physical phenomena, and came up with amazing results, with the researchers unable to understand the “mathematical language” their creation used.
” I’ve always wondered, if we encountered an intelligent alien race, would they have discovered the same physical laws as us, or could they describe the universe in a different way? asks Hod Lipson, one of the scientists who conducted this research, in a statement from Columbia University. It is somewhat with this idea in mind that a team of engineers designed a program based on automatic learning intended for the observation of physical phenomena. The research results were published in the journal Nature Computational Science last July 25.
Indeed, as scientists remind us, the observation and understanding of variables have always preceded the great physical theories. “ For millennia people knew about objects moving fast or slow, but it was not until the notion of speed and acceleration was formally quantified that Newton was able to come up with his famous law of motion F=MA “, notes for example Hod Lipson. It is therefore quite plausible to suppose that physical phenomena can still remain inaccessible to us simply because we have not yet understood their operating rules. ” What other laws are we missing just because we don’t have the variables? sums up Qiang Du, who co-directed the work.
In order to carry out their experiment, the scientists therefore first “fed” their program with raw videos of already well-identified phenomena. For example, they offered him a video of a swinging double pendulum, which is known to have exactly four “state variables” — the angle and angular velocity of each of the two arms. The algorithm they used is specially designed to observe physical phenomena via this type of video, and to “ find the minimum set of fundamental variables that fully describe the observed dynamics “.
It took him a few hours of analysis to produce an answer to the question “by how many variables can this phenomenon be described”: 4.7. ” We felt this answer was pretty close, especially since the AI only had access to raw video footage, with no knowledge of physics or geometry. But we wanted to know what the variables actually were, not just how many “, describes Hod Lipson. This is where things got complicated for researchers.
Data that does not correspond to any variable
Indeed, the four variables identified by the program did not seem to correspond to anything known. Only two of them could vaguely match the angle of the arms. ” We tried to correlate the other variables with everything we could think of: angular and linear velocities, kinetic and potential energy, and various combinations of known quantities. But nothing seemed to fit perfectly says Boyuan Chen. However, the team was convinced that the program had found a valid set of four variables, because it had proven that its predictions were correct. So they came to a conclusion: they just couldn’t understand the ” mathematical language used by their creation.
They then continued the experiment by validating a number of physical systems they knew, then feeding the AI (artificial intelligence) with videos whose exact “answers” they did not know. A wind dancer in front of a used car lot, for which the program found 8 variables, a lava lamp, which also produced 8 variables, and a chimney fire, which returned 24 variables. It therefore remains to know what exactly these variables correspond to. Could they be clues to new principles of physics?
“ Maybe some phenomena seem cryptically complex because we try to understand them using the wrong set of variables. In the experiments, the number of variables was the same each time the AI restarted, but the specific variables were different each time. So yes, there are other ways to describe the universe and it’s entirely possible that our choices won’t be perfect. “say the scientists.
Source: Nature Computational Science