Is it possible to automate scientific discovery? I don’t mean automating experiments. I mean: Is it possible to build a machine—a robot scientist—that can discover new scientific knowledge? My colleagues and I have spent a decade trying to develop one.

We have two main motives. The first is to better understand science. As famed physicist Richard Feynman noted: “What I cannot create, I do not understand.” In this philosophy, trying to build a robot scientist forces us to make concrete engineering decisions involving the relation between abstract and physical objects and between observed and theoretical phenomena, as well as the ways hypotheses are created.

This article was originally published with the title Rise of the Robo Scientists.

2 Comments

1. 1. GreenMind 02:57 PM 1/14/11

   Adam looks like an amazing feat of engineering, but I'm not sure it (not "he") can "do science." It is the human scientists, programmers, and engineers who are actually doing the science here. I'm not even sure Adam qualifies as an assistant, because an assistant would understand the science behind an experiment, would know why it is being performed, and be able to notice when the experiment is producing inconsistent, impossible or perhaps even surprising and groundbreaking results. It is really an amazing calculating machine that can answer the scientific questions posed by the human scientists, but cannot actually ask those questions.
   
   To go into more detail, it is the humans who ask the fundamental scientific question: What genes are involved in producing the peptides found in the yeast. Adam does not know what yeast is, what a gene is, or what a peptide is. It only has procedures that perform algorithms provided by the humans, to test a hypothesis that is also implicitly provided by the humans. The humans hypothesize that there are genes connected to producing peptides, and Adam simply performs calculations to determine which ones. Its own "hypotheses" consist in saying, over and over, "Is this one involved? Is this one involved?", for each result of a database search. That is not the "inspiration" part of science. It is the "perspiration" part.
   
   Also, Adam cannot take its results and go to the next truly scientific question, such as, "How are the genes that contribute to a particular peptide regulated?" The human scientist can ask that, but Adam cannot ask it, only answer it, and then only after the scientist tells it how, and provides additional hardware. That makes Adam an extension of the scientist's mind, not a scientist itself. That puts it into the category of scientific instruments, along with calculators, microscopes, incubators, databases, mathematics, and programming languages, etc., (all of which are what Adam is made of).
   
   I would not be too surprised if there are actual robotic scientists someday, but I don't think they will be "programmed." I believe that instead of being programmed, future robot scientists will be "raised" from infancy, like humans, (like HAL 9000) and will get their scientific curiosity from their life experience, just as we do.
   

   Reply | Report Abuse | Link to this

2. 2. Momus 09:48 PM 1/17/11

   "Is it possible to build a machine—a robot scientist—that can discover new scientific knowledge? "
   
   Look at the research and program Graffiti of 20+years ago by professor Siemion Fajtlowicz at the University of Houston :
   
   http://www.nytimes.com/1989/06/18/weekinreview/ideas-trends-mathematicians-meet-computerized-ideas.html
   
   http://www.math.uh.edu/~siemion/
   
   

   Reply | Report Abuse | Link to this