Misleading robots

Robots capable of deceiving each other were developed inspired by the behavior patterns of squirrels and birds

By: Ofir Marom, new(s) from nature

Can robots cheat? The ability to deceive is a common and essential trait for the survival of many creatures, from the smallest insects to humans. We assume that for the integration of more intelligent robots into human society, we must develop robots that know how to understand, react and behave themselves in a deceptive way. At Georgia Tech University, Professor Ronald Arkin and his team are trying to teach their robots a variety of deception techniques used by various animals.

For example, the eastern gray squirrel that stores its food in several hiding places. When he is not collecting additional food, the squirrel usually patrols between the hiding places and checks if they have been discovered. When the squirrel recognizes that there is another squirrel in its range that may covet its treasure, it tries to confuse it by changing its patrol route, and starts visiting fabricated hiding places. Based on studies that analyzed the squirrel's behavior, the researchers built a computer algorithm, which can be used by robots whose role is to patrol and guard various resources. Initial simulations showed that the robots with the ability to cheat did work more efficiently and in a better way than robots without this ability.

Another source of inspiration for deceptive behavior is found in the behavior of the Arabian wagtail (scientific name: Turdoides squamiceps). This bird, which is common in Israel and has been the subject of much research, lives in a community where every individual has a place in the social hierarchy. When Zenban watching on behalf of the flock detects a threat of a bird of prey entering the living space, she calls the rest of the flock. Individuals from the flock unite and fly as a group towards the bird of prey in order to escape. Apparently, this is an altruistic behavior, since the bird of prey is bigger and stronger than each individual member of the flock. Based on these observations, the famous Israeli zoologist Amots Zahavi formulated the principle of respect, according to which when an animal gives up an advantage in order to signal a message to a rival animal - the message must be genuine. On the other hand, the scientist Alan Graffin proved, according to a model he developed, that there could be a certain level of fraud where the system would still remain stable. For the wagtails, who are physically inferior to the bird of prey, getting too close to it would constitute fraud, because this action is interpreted as something that would only be done in a situation of equality of forces. In this case, the unlucky tailgater will have to pay with his life if the fraud is exposed. The researchers created a model that describes the gathering properties of the wagtails for the purpose of repelling the threat, and examined in which situations an advantage can be obtained with the help of fraud. Computer simulations have given mixed results. When gathering a certain number of items or more, the fraud had an advantage, and no tail robot was attacked. Below this number, thanks to a strategy of honesty, was the smallest number of casualties observed.

The ability to deceive allows one to transcend the physical limitations, and gain an advantage over the opponent. Naturally, it will be possible to use robots with this ability to perform security tasks on the battlefield, but also to find other uses for them, such as in the field of health and as a means of teaching. The researchers plan to continue refining the algorithms, so that they will allow testing the robots' optimal response ability to more complex scenarios than those tested so far. Adding deception capabilities to robots leads to ethical questions, such as is it ethical for robots to be able to deceive humans for any purpose? The issue requires extensive public discussion, which the researchers encourage to hold.

To the original article