One is a robot lab assistant that could replace graduate students, if it could ever learn to survive on a diet of junk food and alcohol while being paid far too little to afford either. The other is a robot ‘baby’ designed to mimic human motion and the process of human learning. Of the two, it promises the greatest potential for making a robot more than a manufacturing accessory, but only if its creepiness is reduced by several orders of magnitude.
Reuters carries a story today about a robot built by two teams of university based researchers in England that can carry out experiments on yeast.
The robot – named Adam – is a step above normal lab-automation gear in that it is not only to carry out the steps of one experiment, but is also able to monitor the result and use it to determine which experiment to carry out next.
The robot, a project from the Computational Biology research group at the Aberystwyth University, appears to be a serious advance on previous generations of lab automation.
It carries specific instructions on how to carry out specific experiments and uses abductive logic programming to generate hypotheses that fit the data and choose which in a range of other experiments would support or disprove the hypothesis.
It follows experimental protocols, gathers data, reaches conclusions and ends up with usable proofs and evidence of specific chemical processes. That’s valuable, especially when principal researchers or time-limited graduate assistants don’t have to carry out the dozens or hundreds of experiments.
It’s a big step into the physical world beyond software that models physical processes and extrapolates conclusions from incomplete (or virtual) results. It is not a replacement for even inexperienced graduate students, however. It’s a machine that automates some of the drudge work of a series of not-too-exciting lab experiments.
It’s not a scientist, however. It doesn’t identify anomalies and apply innovative solutions to explain them; it doesn’t identify new areas that need to be investigated. It doesn’t decide when a conclusion is justified and explain it in a way that its colleagues can understand and repeat.
To their credit Ross King, who led the project, and his colleagues don’t claim it is a scientist. Media outlets who picked up the story do, however. Reuters calls Adam a robot scientist that can think for itself.
Scientific American, thankfully, only claims that ”AI Lab-Bots Take on Reams of Data,” which doesn’t stretch the facts too far. (We don’t know what to think about the Sydney Morning Herald, which claims the ”four-brained sci-bot Adam takes path of yeast resistance,” which doesn’t really exceed the facts, but only because it appears to make no sense.
What most of the press outlets share is the assumption that robots will replace humans once they can accomplish some part of a task humans currently have to handle themselves.
That assumption leads not only to misunderstandings about what a robot does, but also to poor investment decisions and the failure of businesses that hold too closely to the replacement theory, according to William Townsend, president and CEO of Barrett Technologies, Inc. Barrett’s main product is the highly manipulable, haptic-sensitive WAM robot hand, which has been adapted for both surgical and industrial applications.
“In the ‘90s there were a couple of attempts to replace surgeons with a robot,” Townsend said, during a conversation completely unrelated to either Adam or lab automation. “They thought you could take the person out and put the robot in, but it didn’t work on several levels. Robots have no intuition, so if something happens that’s not planned, or is unusual, it’s stuck. Surgeons are usually pretty good at knowing what to do.”
Robots, like any other type of automation or service machinery, are designed to take on tasks humans are unable or unwilling to do, or that are so repetitive that a piece of equipment that can do the same thing over and over, without variation, usually very quickly, makes the job go faster.
That ability has made countless specific functions in manufacturing and other areas far safer, more efficient and higher quality. They have replaced the work of humans who were doing that mindless work, but have not replaced the knowledge, intuition or originality of the humans involved in lab or industrial processes. And they won’t, until AI goes beyond pattern recognition to the point that it is able to use non-linear logic to connect pieces of information that seem unconnected, or some up with solutions that don’t follow precisely the pattern of the previous solution.
That will happen eventually. But it hasn’t yet. And we’re serving no one, least of all the public, by using the “robot-replaces-human” meme to describe the capabilities of the latest and greatest thing to come out of robot labs or design studios.
And the robot baby? It’s learning to walk by watching humans walk and learning to talk by listening to humans talk.
If it does well that will make a great contribution to robot programming, and it’s already more useful than either the robot fashion models or robot teacher, neither of which can get their prospective audiences to emulate the way they look or learn or behave, which is the primary responsibility of both models and teachers.
But it’s creepy enough to scare real babies who, like dogs, are excellent judges of what to trust and what not to, no matter how much they pretend to be a fashion model or teacher or scientist.