More Smart Machines

I just did a blog entry showing scepticism about a university announcing it had created a "robot scientist" able to do chemical research. Then I ran across this headline:

Computer Program Self-Discovers Laws of Physics

Wow! First machines conquer chemistry, now physics. Mathematics and medicine can't be far behind!

So I dug a little deeper...

Here's the claim:

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.

That sounds wonderful, but how real is it? The Wired article reporting this "result" does show a little scepticism by reminding readers of previous claims:

Lipson's program, co-designed with Cornell computational biologist Michael Schmidt and described in a paper published Thursday in Science, may represent a breakthrough in the old, unfulfilled quest to use artificial intelligence to discover mathematical theorems and scientific laws:

• Half a century ago, IBM's Herbert Gelernter authored a program that purportedly rediscovered Euclid's geometry theorems, but critics said it relied too much on programmer-supplied rules.


• In the 1970s, Douglas Lenat's Automated Mathematician automatically generated mathematical theorems, but they proved largely useless.


• Stanford University's Dendral project, was started in 1965 and used for two decades to extrapolate possible structures for organic molecules from chemical measurements gathered by NASA spacecraft. But it was ultimately unable to assess the likelihood of the various answers that it generated.


• The $100,000 Leibniz Prize, established in the 1980s, was promised to the first program to discover a theorem that "profoundly affects" math. It was never claimed.

But now artificial intelligence experts say Lipson and Schmidt may have fulfilled the field's elusive promise.

Here is a video put out by the lab to demonstrate the underlying research behind this claim of "discovering laws of physics":



I'm not an expert, but it seems to me the claim is a bit more than the underlying reality. I can see a very clever program that can fit equations to motion. And, although they don't show the evolution of the equations over time, I believe they are able to simplify the equation. This is impressive. But the claim that the software algorithm has "discovered" physical laws is not convincing to me. At 4:20 into the video the researcher highlights key physical laws represented by terms created during the equation fitting. But that isn't the same as "discovering laws of physics". The underlying algorithm has no awareness of any "discovery". It simply has fit an equation to data. The interpretation of "law" is something the human did in looking at the equations produced. I do recognize that terms in the quation represent laws, but I don't know enough to say if the full equation of the double harmonic system is in fact a "law" or just a nice mathematical formulation of the motion. I do know that real behaviour is messy because reality is messy. The best scientific laws are idealizations that remove clutter so that we understand "the essence" of the idea. The machine is clearly not doing that.

Here is a TED Conference talk by Hod Lipson (click to see video), one of the co-authors of the above claim. Here he is giving a presentation of other research in which he claims to have evolved machines that can learn to "walk", machines that can use self-awareness to organize its walking, and machines that without any direction can self-replicate. These claims sound absolutely marvelous. But if you watch the video, and you are like me, you will be underwhelmed. The claim isn't wrong, but it sure leads to an impression that the actual robot behaviour does not match what common sense expects. Sure, the behaviour can be interpreted as demonstrating the principle, but the "accomplishment" is underwhelming. Maybe I'm being too harsh. If I had been at Kitty Hawk in 1903with the Wright brothers and seen the first "flight of an airplane" I would have been underimpressed. But to my mind, there is a long way from this to realizing the claim. This researcher has a long way, just as real airplances were a long way from that first "flight" of 120 feet in 12 seconds, at a speed of only 6.8 mph.

Bottom line: these researchers are overselling his research to get more research funding.