Today, will computers replace scientists? The University of Houston's Math Department presents this series about the machines that make our civilization run, and the people whose ingenuity created them.
Newton's Laws of Motion embody some of humanity's greatest insights. Each law describes a universal relation between the motion of an object and forces acting on it. These laws condense the results of millennia of human thought. It was therefore a surprise when computer scientists announced that an algorithm named Eureqa rediscovered Newton's Second Law on its own. After only a few hours of computation, a machine concluded that force equals mass times acceleration.
Newton's First and Second Laws of Motion in Latin
as they appeared in the original 1687 Principia.
Here is how it did it: The scientists let the computer observe a swinging pendulum. The algorithm then attempted to guess the law that governed the pendulum's motion. The first guesses were off the mark. But, some were less wrong than others. In a process similar to natural selection, the program slightly adjusted and combined the best guesses to create the next generation of potential laws. Those guesses that did not agree with the observations or were too complicated had no descendants — only the best ones reproduced. After refining its guesses over many thousands of generations, the program arrived at the same answer as Newton. For humans this would be a mind numbing approach that would take generations to complete. It took only a few hours on a modern computer.
Double Pendulum long exposure, tracked with LED light at its end.
Does this mean that we no longer need scientists? It has become easy to collect vast amounts of observations. Perhaps we can simply let clever algorithms sift through these mountains of data. Computers can tease out relations and rules that would take humans centuries to uncover. Links between certain diseases and genes are already being made this way. Will computers discover new physical laws? Or novel mathematical truths?
We are not there yet. Humans still form an essential part in any process of discovery. Machines are great at uncovering correlations. For instance, they can tell us that a variant of a gene increases the likelihood of diabetes. However, computers are not good at finding the mechanisms that are behind such links: Does the gene cause diabetes, or does it simply occur more frequently in people who are predisposed to it?
Moreover, many statements are true, but only few are interesting or useful. People evaluate the output of machines. And people synthesize these facts into the theories and models that describe the world around us.
It is easy to be dazzled by the powers of our machines, and take for granted the powers of the humans that operate them. It was humans who decided what problem Eureqa should tackle. And it was humans who steered and fine tuned the algorithm, evaluated its output and decided which parts were interesting and insightful. This is the type of intelligence that computers do not yet possess, and may never achieve. It is still only humans who can fully appreciate them.
This is Krešimir Josić the University of Houston, where we are interested in the way inventive minds work.
You can read more about Eureqa at http://www.wired.com/wiredscience/2009/12/download-robot-scientist/, and about the end of theory at http://www.wired.com/science/discoveries/magazine/16-07/pb_theory.
The search algorithm that is used by Eureqa is a variant of the genetic algorithm described at https://en.wikipedia.org/wiki/Genetic_algorithm
Eureqa is free to download and to use. For a video that explains what they have done, visit https://www.youtube.com/watch?v=MSo6eeDsFlE.
Researchers at Google also argue that data can start to speak for itself https://research.google.com/pubs/archive/35179.pdf.
Both images are from Wikipedia.