I haven't read the article yet, but the picture made me smile. Ah, pity the
poor young folk who didn't grow up on robots like this in the movies!
----- Original Message -----
From: "Mr. Worf" <hellomahog...@gmail.com>
To: scifinoir2@yahoogroups.com
Sent: Thursday, July 22, 2010 5:32:33 PM
Subject: [scifinoir2] The Future of Robot Scientists
The Future of Robot Scientists
• By Brandon Keim Email Author
• July 22, 2010 |
• 2:10 pm |
• Categories: Tech
•
Future science historians will mark the beginning of the 21st century as a time
when robots took their place beside human scientists.
Programmers have turned computers from extraordinarily powerful but
fundamentally dumb tools, into tools with smarts. Artificially intelligent
programs make sense of data so complex that it defies human analysis. They even
come up with hypotheses, the testable questions that drive science, on their
own.
At the University of Cambridge, Ross King’s program “Adam” designs and runs
genetics experiments . At Cornell, Hod Lipson’s Eureqa finds equations to fit
data, attaining Newton’s insights in a single afternoon . University of Chicago
mathematical biologist Andrey Rzhetsky designs programs less glamorous but
equally powerful, able to analyze millions of papers at once.
In the future, the human scientist’s job may be “to do the programming, and
make sure the robot has enough reagents,” said Rzhetsky, only partly
tongue-in-cheek.
Wired.com talked to Rzhetsky about the intersection of artificial intelligence
and science.
Wired.com: Why do scientists need artificially intelligent computer assistance?
Andrey Rzhetsky: During Newton’s time, a scientist could read everything that
was published, at least in English. That’s just not an option anymore. We can’t
deal with all this information.
Wired.com: How have you used AI in your own work?
Rzhetsky: In our paper on brain malformations in mice and humans , the program
analyzed 368,000 full-text articles and 8,000,000 article abstracts in the
PubMed database. That’s something no human curator, or even a group of human
curators, could ever do. In a program, it’s possible.
We made available a huge knowledge base and a tool for prioritizing genes and
making hypotheses about associations between genes and phenotypes. A bunch of
the predictions we made were followed up by our experimentally talented
collaborators, and seem very reasonable.
The problem is how to design a process to discover a good hypothesis, because
it’s expensive to test all possible hypotheses. That’s where literature
analysis and computational modeling can help. It prioritizes.
Wired.com: So much published research isn’t replicated. Isn’t there a
garbage-in, garbage-out problem?
Rzhetsky: That’s always a possibility, but good statistical analysis doesn’t
throw away data. Even with good data, you get a lot of noise. Even noisy data
with false positives can be useful.
Think about it as intelligence data. Obviously, when it’s collected, there are
lots of false positives. But when it’s collected from multiple sources,
compared and examined, it becomes more certain.
Wired.com: Cornell’s Hod Lipson designed a program that discovers equations to
explain relationships between data. Researchers then have to figure out what
the equations mean. It’s like interpreting an oracle’s pronouncements . Is that
the role of the human in all this?
Rzhetsky: It’s an interesting question. I talk to electrical engineers who use
genetic algorithms to design circuits, and the circuits end up being completely
alien to humans. They’re very robust, but designed in such a way that it’s not
obvious how to understand them. That’s similar to what Lipson discovers:
non-human logic. In Lipson’s analysis, he wants to make it transparent and
understandable to humans. I’m not sure that’s necessary.
Wired.com : Some scientists say that being able to crunch huge datasets makes
hypotheses obsolete — why worry about testing when you can find connections.
You don’t like that idea, though. Why not?
Rzhetsky In the movie Memento , a man has only a short-term memory. Every 15
minutes has to reconstruct causal relationships. He observes people talking to
him, and doesn’t know who’s a friend and who’s a foe. That’s my metaphor for
abandoning hypothesis and context.
There are a lot of approaches claiming you can reverse-engineer the world from
the flow of data. With an infinite dataset, the statement probably gets close
to truth. But I don’t think it’s true for individual datasets. Prior hypotheses
and contextual knowledge need to be used.
Wired.com : So is the role of human scientists to come up with hypotheses?
Rzhetsky : The tools can come up with hypotheses, too.
Wired.com: One of the great human abilities is to come up with insights that
combine knowledge and speculation across disciplines. How could a program ever
have those insights?
Rzhetsky: One kind of creativity is combining old symbols in a new way. The
best thinkers digest the experience of previous thinkers, and come up with
their own syntheses. I would claim this is still in the space of symbolic
reasoning and symbolic hypothesis generation.
Wired.com: But wouldn’t this require far more general artificial intelligence
than the narrow, task-specific types we have now?
Rzhetsky: Possibly. But you can think about the human brain as a collection of
specialized tools. There’s a tool for discerning vertical symmetrical patterns
in noisy backgrounds in order to find predators, a tool to recognize faces, a
tool to classify experiences as pleasant or unpleasant, and so on. I don’t see
why a tool that does several specialized tasks well can’t be upgraded to
something more comprehensive.
Photo whiskey kitten /Flickr
See Also:
• Robot Makes Scientific Discovery All by Itself
• Computer Program Self-Discovers Laws of Physics
• Download Your Own Robot Scientist
Citation: “Machine Science.” By James Evans and Andrey Rzhetsky. Science, Vol.
323 No. 5990, July 23, 2010.
Brandon Keim’s Twitter stream and reportorial outtakes ; Wired Science on
Twitter . Brandon is currently working on a book about ecological tipping
points .
Read More
http://www.wired.com/wiredscience/2010/07/robot-scientist/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+wired%2Findex+%28Wired%3A+Index+3+%28Top+Stories+2%29%29#ixzz0uRudrihG
--
Celebrating 10 years of bringing diversity to perversity!
Mahogany at: http://groups.yahoo.com/group/mahogany_pleasures_of_darkness/
