In the category of scientific computing, Wired magazine reports that researchers have programmed a computer using a form of artificial intelligence (it looks like genetic programming to me) to figure out how flatworms are able to regenerate their body parts in the correct size and shape.
"For the first time ever a computer has managed to develop a new scientific theory using only its artificial intelligence, and with no help from human beings.
“Computer scientists and biologists from Tufts University programmed the computer so that it was able to develop a theory independently when it was faced with a scientific problem. The problem they chose was one that has been puzzling biologists for 120 years. The genes of sliced-up flatworms are capable of regenerating in order to form new organisms – this is a long-documented phenomenon, but scientists have been mystified for years over exactly what happens to the cells to make this possible.”
It was a brute-force approach, requiring the program to make successive iterations until its results matched what actual flatworms do, but at least it reached a successful answer.
Very cool! But to nitpick, what it sounds like the computer did was develop a mathematical model for the problem, which is not the same as forming a theory at least as the process is typically understood. Which is especially true if the model has a probabilistic basis.
Thank you for pointing out yet another research paper that I will need to read!
I wouldn’t expect a computer program to produce a formal theory in the form that humans would submit for peer-review, but the linked article suggests that the computer did more than just model the problem; it also modeled the circumstances that would result in the observed data. I take that to be the theory the computer generated.
It will depend upon what they mean by model, and what the computer actually did develop. Artificial Intelligence is an area where researchers have been making over blown claims for decades. I am skeptical that a computer could generate a deterministic model of any behavior that wasn’t already sufficiently understood that it could simply iterate through solutions. If the behavior isn’t sufficiently understood (which means the theory already exists), then I suspect that the computer could only arrive at a stochastic model, which is insufficient for any theory.
The main reason I doubt this article’s claim is that when a human develop’s a scientific theory, they start with attempting to describe/understand a possible series of mechanisms that would produce the observed behavior. Then testing those concepts against the data. But the core concepts are usually a result of inspiration/intuition not a rational (at least as we currently understand it) process. For a computer to replicate that process they would have to start with a nearly infinite number of possible mechanisms and narrow them down by testing against known data. Then with the subset that best fits the known data they would have to refine parameters. But the real problem with that approach is that the computer doesn’t actually comprehend what it is looking at so, it is very easy for coincidental conditions to produce what appears to be a ‘valid’ theory based upon the ‘real data’ it has available to it. It takes a human to review the raw material produced by the computer algorithm to decide validity. And if any of that applies to what these reseachers did, then their AI didn’t develop a theory.
Of course, as with all things, I could be mistaken and they may have made a real leap. Which is why I look forward to getting a copy of their paper to read.
Thank you for taking the time to look up the journal articles! BTW, I like your postings on these articles, I rarely have time to review media much any more,
