adam, artificial intelligence adam
Rowland/Aberystwyth University/PA Wire
A robot called Adam that can think up
scientific theories and test them with
almost no human help

Will the Future Bring a Population of Deceptive, Angry Robots?

August 19, 2009 05:30 PM
by Haley A. Lovett
Swiss scientists have found that deception can have an evolutionary advantage among competing robots. Now researchers wonder if anger is the key to making more humanlike robots.

Robots Deceive to Get "Food"

According to Ed Yong of the blog Not Exactly Rocket Science, 10 robots were placed in a controlled environment in a Swiss laboratory. The robots were programmed to compete for “food” and were also able to talk to one another through flashing blue lights. Food was made scarce—only eight of the 10 robots could be at the food source at any one time—so not all robots would be "healthy" at the end of each experiment.

“Each [robot] can produce a blue light that others can detect with cameras and that can give away the position of the food because of the flashing robots congregating nearby,” Yong explains. “In short, the blue light carries information, and after a few generations, the robots quickly evolved the ability to conceal that information and deceive one another.”

The robots that got the most food were allowed to “reproduce,” which, according to John Timmer of Ars Technica, meant that their “33 digital ‘genes’” were “mutated, recombined, and randomly assorted” to produce robot offspring. Even though the robots had no reaction to the blue light at the beginning of the experiment, “by generation nine, the robots had evolved so that they were attracted to light, a behavior that wound up predominating in future generations,” Timmer writes.

In this scenario, the “healthiest” robots are those that “evolve to avoid tipping their competitors off to the site of a food source,” according to Timmer. The Swiss researchers suggest that the experiment provides insight into some of the social cues we see in other species, and how these cues evolve.

Opinion & Analysis: Is an angry robot a more human robot?

Practicing deception is not uncommon in the animal world, and can readily be observed in humans. Traditionally, traits such as deception—along with emotional traits such as anger—have separated humans from robots and other sources of artificial intelligence. 

In a recent Scientific American podcast, Christie Nicholson examines the ongoing debate about how, and if, artificial intelligence could ever be accepted as part of the human “ingroup.” Nicholson notes that some researchers think that in some experiments, the only way to distinguish “incredibly humanlike robots” from humans may be through an “angry attack.” In other words, insult the robot and see if it responds. The artificial intelligence “required to grasp human emotion may forever be the telling distinguishing feature,” Nicholson writes. She goes on to ask “if anger turns out to be an effective measurement for ‘human,’” then would a society of robots be a cantankerous one?

Related Topic: Robots taking on more complex roles

Artificial intelligence technology has improved a great deal over the last decade, including within the military. The U.S. Air Force recently predicted that killer robots with the ability to make decisions such as when to strike a target could be less than 50 years away.

With the unveiling of ADAM, a robot that can analyze data and then form and test multiple hypotheses, scientists are getting closer to using robots to help execute and record experiments. One of the scientists working with ADAM notes that it is unlikely that humans will be taken out of the discovery process altogether, however: Robots are still unable to look at the “big picture” of a series of experiments.

NEXT: Learn more about advances in robot intelligence

Most Recent Beyond The Headlines