I first mentioned this story back in April, but it deserves a closer look. I really think it's going to end up being the biggest science story of the year.
Mad Observations: With things like the Human Genome Project and the other genome projects that preceded and followed it, we are gathering reams of data, more than we'll be able to fully investigate any time soon. At least, more than we can investigate by hand. But hey, computers are pretty advanced these days...
Mad Reference: "The automation of science." King RD, Rowland J, Oliver SG, Young M, Aubrey W, Byrne E, Liakata M, Markham M, Pir P, Soldatova LN, Sparkes A, Whelan KE, Clare A. Science. 2009 Apr 3; 324(5923): 85-9. I also recommend the excellent write-up on the research in Wired.
Mad Hypotheses: The first hypothesis was chosen by the researchers (who were bordering a bit on mad engineering, so the hypothesis is pretty close to "Let's see if we can do this"). It's along the lines of "It is possible for a properly programmed robot to investigate data, make a hypothesis, and test that hypothesis."
But then we get into the first experiment, and get the cooler hypothesis. The researchers programmed a robot, named Adam, to perform science. Without help, just looking at the data from the Saccharomyces cerevisiae (brewer's yeast) genome project and other genetic databases (plus a model of S. cerevisiae metabolism), Adam hypothesized that certain genes in the yeast genome coded for an enzyme that had a certain function in metabolism. These genes were there in the data, but had not yet been characterized. So Adam set out to characterize those genes, hypothesizing that they would produce an enzyme that would catalyze a certain reaction in yeast metabolism.
Mad Experiment: Unfortunately, I don't have full access to the article, and both the abstract and the Wired article are sketchy on the details here. I'll lay out a couple possibilities, though, for people interested in how a researcher (including a robotic researcher) might figure something like this out.
Put simply (but close enough to give the idea), Adam knew that yeast used an enzyme to turn compound A into compound B, and another to turn B into C, and yet another to turn C into D, etc. He just didn't know for sure what those enzymes were. Let's say he was hypothesizing that the enzyme he was looking at turns A into B.
One way to figure out if he's right would be to create a yeast cell that lacked the genes he was looking at (likely one at a time plus all three); the yeast cell would be exactly like a normal yeast cell, just missing the one gene he was looking at. If he fed normal yeast cells A, they'd grow and produce B, C, D, etc. If he fed his modified yeast cells A, if he was right, they wouldn't produce B, C, D, etc. He could then feed his modified cells B, and they'd then be able to produce C, D, etc. If any step of that didn't work as expected, his hypothesis would be false.
The other possibility would be that he directly characterized the genes, creating copies of the genes he was looking at in a test tube, supplying them with the components necessary to translate those genes into proteins, and seeing what happened when he put A into those test tubes. That sort of research is less reliable, though (if it doesn't work, it could be because you're missing some factor necessary to make the protein, not because the proteins are important in what you're looking at), so I think it's more likely that he used the first approach.
All this time, all the researchers did was supplied him with the chemicals he needed, and emptied out wastes. He did all the rest, designing and performing over a thousand new experiments a day.
They all laughed, but: Since I'm writing about it here, you've probably already figured out that it worked. He was able to identify that three previously uncharacterized genes in the yeast genome code for an enzyme that catalyzes a certain step in yeast metabolism.
The particular discovery made by Adam wasn't particularly Earth-shattering; he found something that would have otherwise have been assumed to be true, but he verified it. The next step is the cool part. Robots like Adam can now dig through the genomes that we've sequenced, making similar hypotheses and performing similar experiments. Now that we know that they work, the interesting part comes when they fail to verify what they're looking at. On top of verifying and adding to the body of science, they would then find something for us to look at more closely.
Mad Engineering Applications: Since this all started with a dose of mad engineering—specifically, making robots to analyze data, make hypotheses, perform experiments, and analyze their results—there isn't much left in this particular area for mad engineers to do. For a while now, this one's pure mad science. Hopefully they'll make us some more robots capable of performing other experiments, but, once they give us our army, it's scientists that will utilize that army. And hopefully more scientists (mad or not) will come up with more ways to apply this research, potentially dramatically increasing the rate of increase of the sum of human knowledge. "What we know" already increases dramatically every year, as does the rate of discovery of new information (so if it doubled last year, it's likely to more than double this year). If robot science catches on, the rate of increase is likely to go way, way up.
Have any ideas for what tasks we should set our army of robot scientists on? Let me know in the comments.