Robot Scientist Conducts, Interprets Lab Tests

Reuters via Yahoo ^ | Wed, Jan 15 2004 | Patricia Reany

[RightWingAtheist]

LONDON (Reuters) - It doesn't look anything like R2-D2 of "Star Wars" fame but British researchers said on Wednesday they have created an intelligent robot capable of doing experiments and interpreting the results.

The robot scientist can formulate theories, do research and could be useful in discovering new drug targets. It works as well as a graduate student but is unlikely to put anyone out of a job.

Instead, its creators say it could free scientists from routine laboratory tasks and allow them to concentrate on more important aspects of their research.

"As with many other developments in the lab, it will hopefully give people more time to do the creative part of the work," Professor Stephen Oliver, of the University of Manchester, told Reuters.

Oliver and a team of computer scientists, microbiologists, molecular biologists and other researchers from the University of Wales, Aberystwyth, the Robert Gordon University in Aberdeen, Scotland and Imperial College London have been developing the robot scientist for three years.

It sits on top of a desk and is attached to a computer.

In the latest issue of the science journal Nature, the researchers told how the robot scientist gained top marks when asked to determine functions of genes in yeast.

"It was given background knowledge about the biochemistry but it had no knowledge of the genetics and had to deduce the genetic relationships," Oliver said.

Now that it has passed its first scientific hurdle and proved that it can uncover something that was already known, Oliver and his colleagues want to see if it can discover something new.

"Although the problems we set for the robot were relatively simple, we have shown that it could be used to help solve real-world problems," Professor Ross King, a member of the team from the University of Wales, said in a statement.

The scientists hope the robot will speed up the search to uncover the function of other genes in yeast.

"We also think there are applications in the pharmaceutical industry for (drug) target discovery and verification," Oliver added.

He said there is nothing novel in the robot's hardware: its uniqueness lies in the way it has been put together and the programs used to control it.

The researchers have no plans to mass produce the robot but said they would sell the instructions to make it.

[RightWingAtheist]

Some intriguing stuff here. I'm fascinated by the cognitive approach to the study of science ("recursive science" if you'd like) and one of the questions frequently posed by these scholars is: can we get a machine to "do" science? Not merely work as a tool, but be able to independently experiment and theorize, and perform all the other mental operations which constitute science. The question I am asking here, is do you think this program is really doing science, or is it just going through programmed motions (sort of the AI equivalent of Koko the Gorilla-we think it's coming up with concepts independently, but it really isn't) ? My own feeling is that some sort of advanced modeling program, which allow to conceptualize knowledge in an independent manner.

[Doctor Stochastic]

The question I am asking here, is do you think this program is really doing science, or is it just going through programmed motions...

Is it possible to tell the difference? A sequence of (non-nested) several hundred IF...ELSE..ENDIF statements (in whatever language) has more paths through the program than the number of atoms in the universe. A small Turing machine (23 internal states, 2 symbols) can compute any "computable" function (which is not very restrictive as every attempted description of computable has led to the same set of functions.)

If a set of meta-instructions can be programmed, then it should't be too hard to have a machine do things that seem hard. For example, a program can generate other programs and call them (or call itself) so the complexity of the program can self-increase.

[PatrickHenry]

This is way out of my experience, but this "robot" doesn't strike me as being all that different, in principle, from a computer which is programmed to play chess. It's given a bunch of rules, and it runs the data through the rules in a brute force way, possibly bypassing some obvious dead-ends via a few good algorithms. I think it shows promise of being a handy lab assistant, but I doubt that it's going to do original work.

[anymouse]

This just in, Tony Blair announces amnesty for androids working illegally in the UK. Socialists confused on which side of this issue they should be on, but demand that robotic workers must be unionized immediately. ;)

[sixmil]

I'm not sure about what you said regarding unnested if-else statements and Turing machines, but suffice it to say that, assuming the processing power is there, a computer is capable of doing anything a progammer can train it to do. That is, the only limiting factor is a programmer being able to understand the processes a scientist goes through and codifying them so that a computer can replicate them. The only way computers think for themselves is through organic computing, but in reality, the computer is still given initial instructions and simply employs the power of brute force like a chess computer. The more you look into things like this the more you see that many things we do are instinctual, and we have trouble explaining the exact methods we use to come to conclusions. So really it is never the computer thinking, but the programmer doing all the thinking up front, then the computer takes over with its vastly superior endurance, speed, and precision.

[Doctor Stochastic]

... So really it is never the computer thinking, but the programmer doing all the thinking up front, ...

There's no indication that this is a limitation on the power of a computer. That's the queston, can a programmer program a computer to "think"? Of course, there are people who define "think" as "whatever a computer cannot do." On the other hand, computers can produce answers that are unexpected by the programmers (else why run the program in the first place?)

One can also program a computer to accept randomly generated (QM type) input (or at least input that is not pre-computable by the computer.) This doesn't help compute more functions though.

[sixmil]

On the other hand, computers can produce answers that are unexpected by the programmers (else why run the program in the first place?)

Yes, but those are usually classified as bugs.

[Yeti]

It works as well as a graduate student but is unlikely to put anyone out of a job.

hehe: "but?"