Researchers at a British university say they might have invented themselves out of a job.
A new robotic system they developed can, for the first time, independently design and carry out a genetics experiment, then interpret the results.
No difference was found between the lab bench results generated by the robot scientist and those gathered by graduate students doing similar work, the researchers report in the Jan. 15 issue of the journal Nature.
The system remains in its infancy, but they hope it will someday conduct lab-intensive work, freeing the researchers from drudgery.
“The sort of grunt research can be done this way, and more creative stuff humans will have more time to do,” said study author Stephen Oliver of the University of Manchester.
Other researchers described the robot as a “harbinger of the future,” but said more sophisticated reasoning software had to be developed.
Once that happens, labs would adopt such advanced artificial intelligence systems “pretty rapidly and pervasively,” said Larry Hunter, a computational biology expert at the University of Colorado School of Medicine, who was not involved in the experiment.
The robotic system was designed to determine the function of baker’s yeast genes. About 30 percent of the yeast’s 6,000 genes are unknown, but scientists believe they may be shared in the human genome and might someday be medically important.
To determine functions of the genes in question, the experiments used “knockout” varieties in which a specific gene is removed. By determining how the yeast sample grows, the function of the missing gene can be determined.
In the automated experiments, the researchers first developed a mathematical model showing how various genes, proteins, enzymes, and growth mediums interact.
Armed with that knowledge, the robot independently generated hypotheses about the missing genes, then used equipment to grow yeast strains. Later the robot evaluated the growth of each strain against its original hypothesis. The process was repeated over and over as the system developed new hypotheses based on the accumulating data.
“It’s like if you have a machine which is broken, the system can automatically reason to find all the possible ways it can be broken,” said Ross King of the University of Wales-Aberystwyth. “Some philosophers have thought this is impossible for computers, because that’s the imaginative leap.”
The robot scientist uses a type of reasoning called abduction. King said it is the kind of reasoning police use to reconcile clues when investigating a crime.
“If this person committed the crime, all the clues make sense,” King said.
Hunter said the new work marks the first time that experimental design, computer control of instruments, and analysis of the resulting data have been “hooked together in a closed loop.”
“It is now possible to design artificial intelligence systems that are able to reason well enough to be effective partners in scientific research,” Hunter said.
Oliver said the next step is to see whether the robot can make a completely novel discovery rather than simply match the graduate students’ results.
Sidebar: Computer analysis becoming the latest tool for literary research From eSchool News staff and wire service reports
In the mid-1970s, Floyd Horowitz embarked on a long, one-man literary journey: to discover early, uncredited stories by Henry James, stories that had never appeared in book form.
Thirty years later, thanks to tireless research and the emerging field of statistical literary analysis, the former English professor declares his project a success.
“I vowed to continue with this as long as I kept finding interesting material. And I kept finding it,” says Horowitz, editor of The Uncollected Henry James, a new anthology of 24 previously unpublished stories by the author of such classics as “Daisy Miller” and “The Turn of the Screw.”
Now retired, Horowitz recalls reading James’ “The Story of a Year,” published in 1865 and believed to be the author’s first signed work of fiction. Convinced that the story was beyond the abilities of a novice, Horowitz spent three decades looking for previous works through such 19th-century periodicals as the Newport Mercury and Arthur’s Home Magazine.
Because young authors at the time often published anonymously or under pen names, Horowitz did not simply look for James’ byline. Instead, he sought common themes, phrases, and pen names, including “Mademoiselle Caprice” and “O. Chickweed.” Horowitz then assembled a computer database that compared text he believed was written by James to James’ other works and to material from other contemporary authors.
“I came to the conclusion that these early pieces provided a clear window into … James’ known fiction,” Horowitz, who taught English and computer science at the University of Kansas and at Hunter College, writes in the book’s foreword.
Horowitz is among a growing number of scholars who rely on statistical research, which joins the traditionally alien worlds of literature and computer analysis. Computers have been used to examine authorship of countless ancient texts, including the New Testament gospels and Greek and Roman documents.
“There have been ups and downs, but over the years more and more people have accepted computer analysis,” says Bernard Frischer, a professor of classics at the University of California at Los Angeles who used computers to determine the date of some writings by the Roman poet Horace.
The rise of statistical literary analysis, including a method known as “stylometry,” dates to the 1960s when statisticians Frederick Mosteller and David Wallace resolved the authorship of 12 of the Federalist Papers. With scholars unsure whether the essays in question should be credited to James Madison or Alexander Hamilton, Mosteller and Wallace compared the word usage of each writer and concluded that Madison was the author of all 12, a finding most historians agree with.
But Frischer acknowledges that such studies often do not provide definitive answers and that doubts remain over reliability. A recent case involved a 17th-century text, “A Funeral Elegy,” that Vassar College professor Donald Foster identified in 1995 as a poem by William Shakespeare.
Numerous experts accepted the findings of Foster, who had deciphered that journalist Joe Klein was the “Anonymous” author of “Primary Colors,” and the poem was added to prominent anthologies published by Longman and W.W. Norton.
But a few years later, French scholar Gilles D. Monsarrat released a study that contended the author was actually a contemporary of Shakespeare’s, playwright John Ford. Foster agreed, and acknowledged that he had failed to include Ford in his database as a possible alternative to Shakespeare.
“We’re still at the birth of this field, and we still have a long way to go,” says Richard Abrams, a friend of Foster’s and a professor of English at the University of Southern Maine.
“But as more texts become available online and we have more information to draw conclusions from, the results will become more stable.”
Some observers are already skeptical of the James book. Daniel M. Fogel, founding editor of the Henry James Review, a publication that comes out three times a year, says that “the texts themselves are not compellingly Jamesian.” He also believes that Horowitz’s statistical arguments are not “sufficiently elaborated.”
The Uncollected Henry James is being published by Carroll & Graf, which years ago released a popular compilation, The Great Short Novels of Henry James. Company co-founder Herman Graf, a longtime James reader, says he was impressed by Horowitz’s research and became “excited” about the book.
But Graf acknowledges that he did not read all the stories and that some academics declined to offer blurbs, citing concerns about authenticity. Still, he believed the book worthwhile, if only to start a debate.
“This is not a science, and you can never be sure, but I thought it would be of interest to people who love James and would want to decide whether these stories were really his,” Graf says.
Carroll & Graf Publishers