Computer program learns physics

From Wired:

"In just over a day, a powerful computer program accomplished a feat that took physicists centuries to complete: extrapolating the laws of motion from a pendulum's swings.

Developed by Cornell researchers, the program deduced the natural laws without a shred of knowledge about physics or geometry.

...Condensing rules from raw data has long been considered the province of human intuition, not machine intelligence. It could foreshadow an age in which scientists and programs work as equals to decipher datasets too complex for human analysis."


The software has a long way to come, but apparently it's showing a lot of promise.
What do you guys think? If computers can do physics better than humans, does that make human physicists obsolete?

Picture of Computer program learns physics
sort by: active | newest | oldest
1-10 of 16Next »
codyburker7 years ago
um people's brains are just big calculators, running enormously complex equations to control us. The computers are better at running raw numbers because they're set up to do so, but when was the last time that a computer successfully analyzed a human face? If we would ever be able to get the hardware and software down, it would be more that possible for computers to crunch numbers and walk around and problem solve but we won't ever be able to. I heard an article on the radio that was analyzing on how cats drink water, and they lick the surface and the water actually shoots into their mouths. The scientists said that " it is like the cats are running all of the complex equations." so yeah, its possible but kinda pointless in the end.
It's funny to hear people talking about how machines made by other people must be smarter than the people who made them. I personally think it's impossible to build something that's smarter than the builder. On purpose, anyway.
You think so, but do you have any evidence to back up your opinion? If not, some research might be in order, then come back and tell us what you've learned. Here is one piece of evidence against your opinion. Deep Blue, the IBM supercomputer which eventually beat the world chess grandmaster, was trivially much "smarter" (in the sense of being a better chess player) than the programmers who constructed it.
This is a discussion I have with my friends quite often when we are in philosophical mode. The difficult thing about the question at hand is: how do you determine which is smarter? The scientists and programmers may certainly be more adept at creating machines, but the machines that they create can "learn". If a machine, in the process of this learning, actually learned HOW to learn, it could theoretically surpass us since we have a fairly limited knowledge of how our brains learn and how to optimize that learning. At least, that is my side of the discussion. Additionally, if you were to create a program that seems to "think" by giving it the criteria for thinking and stimulating it to do so, how do you prove that the machines thoughts are not just as real as yours? Tough to answer but fun to discuss, for sure.
Computers suffers from severe condition of autism. Sure they can make a billion multiplications per second, but none of them can tie their shoe laces... Still far to go, to match the versatility of the human mind!
puffyfluff8 years ago
My thoughts are that if this program were taught to break down sentences and extract the meaning, then programmed with a dictionary, and plugged into the internet, then it wouldn't just be learning the laws of motion, it would be interpreting everything we know today and making its own conclusions.
(If greatly improved)
fozzy138 years ago
I definatly think we should use a similar computer program to see if it can come up with any arrangement for a working magnetic motor.. : ) Please don't rip on me for being open to the possibility of a magnetic motor..
kelseymh8 years ago
I recommend reading the original preprint. The software is fairly narrow, searching correlations among the input variables for socalled "invariants" or conserved quantities in the time series. It doesn't work analytically, but rather returns a list of about ten possible invariant expressions, sorted by the number of polynomial terms. It's still up to a human to select from among those expressions. One advance is that it does handle "noisy" (i.e., realistic) data much better than past attempts.
Goodhart8 years ago
I'm with Randofo here: until they come out with a Positronic processor (with apologies to Isaac Asimov), I have my doubts too.
1-10 of 16Next »