6
$\begingroup$

I have been reading about a machine that observed a double pendulum and created equations that both described its motion and associated conservation laws. The authors claims:

We have developed a technique for extracting the laws of nature from experimental data by identifying invariant and conservation equations. We demonstrated this approach by automatically searching motion-tracking data captured from various physical systems, ranging from simple harmonic oscillators to chaotic double-pendula. Without any prior knowledge about physics, kinematics or geometry, the algorithm discovered Hamiltonians, Lagrangians, and other laws of geometric and momentum conservation.

This appears remarkable, not least because it took Humans thousands of years to discover these same facts. Are such machines being used in more difficult areas such as quantum gravity or string theory?

$\endgroup$
  • $\begingroup$ It is a specific request for information on the use of AI and genetic algorithms as tools in modern theoretical physics. If you can come up with a list of such projects I will be happy. $\endgroup$ – user56903 Dec 20 '14 at 16:47
  • 1
    $\begingroup$ I feel like there are some really nasty snakes under the grass with this. Surely, they must have told the machine to search for certain types of equations... $\endgroup$ – Danu Dec 20 '14 at 16:52
  • 1
    $\begingroup$ @Danu it could have just tracked the data and done mathematical fits. Then used that to make laws $\endgroup$ – Jim Dec 20 '14 at 16:56
  • 1
    $\begingroup$ But this seems mostly observation based. Quantum gravity, string theory, etc. These are all less observation based. You can make observations after a while, but thought experiments, imagination, and trial and error are kind of needed to get to some of those areas. I doubt this thing could do that $\endgroup$ – Jim Dec 20 '14 at 16:59
  • 1
    $\begingroup$ @Jim Fits, based on what? Surely, one must tell it what kind of variables to look for. This is a major issue for any more complicated theories $\endgroup$ – Danu Dec 20 '14 at 17:00
2
$\begingroup$

One example I can give of using Algorithmic (Statistical) Inference in contemporary Physics research is solving the inverse problem in material science. That is, what should be the chemical composition and structure of a material if it is to have given macroscopic properties (within some tolerance range). These properties could be spectral, band-gap, etc. and the output could be chemical formula and unit cell configuration.

To learn more, please see the DOE Office of Science project http://www.centerforinversedesign.org/

$\endgroup$