To Win the Evolutionary Race, Robots Learn to Deceive

By Eliza Strickland | August 19, 2009 1:22 pm

lie-botsSurvival of the fittest is a brutal game, as a group of robots in a Swiss lab have just demonstrated. When the robots competed for points and only the most successful passed on their computer code (which is analogous to our genetic code), they soon evolved into greedy deception-bots that tried to horde all the points for themselves.

The robots — soccer ball-sized assemblages of wheels, sensors and flashing light signals, coordinated by a digital neural network — were placed by their designers in an arena, with paper discs signifying “food” and “poison” at opposite ends. Finding and staying beside the food earned the robots points…. After each iteration of the trial, researchers picked the most successful robots, copied their digital brains and used them to program a new robot generation, with a dash of random change thrown in for mutation [Wired.com].

The first generations of robots rolled around the arena while flashing their lights at random intervals. But as they were programmed to stay by the food to collect more points, ensuing generations learned to follow the lights of their fellow robots to find food. But because space is limited around the food, the bots bumped and jostled each other after spotting the blue light. By the 50th generation, some eventually learned to not flash their blue light as much when they were near the food so as to not draw the attention of other robots, according to the researchers. After a few hundred generations, the majority of the robots never flashed light when they were near the food [Technology Review]. A few robots even evolved to be repelled by the blue lights, on the principle that any robot flashing its light must be far from food.

The study, which will be published in the Proceedings of the National Academy of Sciences, suggests that similar evolutionary processes are at work in nature. When animals move, forage or generally go about their lives, they provide inadvertent cues that can signal information to other individuals. If that creates a conflict of interest, natural selection will favour individuals that can suppress or tweak that information, be it through stealth, camouflage, jamming or flat-out lies [Not Exactly Rocket Science].

Related Content:
DISCOVER: Robots Evolve and Learn How to Lie
80beats: I, for One, Welcome Our New Robot Scientist Overlords
80beats: Helicopters Learn to Fly Themselves by Studying an Expert Pilot
80beats: Autonomous, Snooping Robots Almost Ready for the Front Line
80beats: Rat Neurons Build a “Biological Brain” for a Robot

Image: PNAS

CATEGORIZED UNDER: Technology
MORE ABOUT: evolution, robots
  • http://blogs.discovermagazine.com/80beats/ Eliza Strickland

    Can’t resist adding these links to the original Lie Bot, from the magnificent comic Achewood. If the Swiss robots keep evolving in this nefarious direction who knows what kind of havoc they’ll wreak?

    http://achewood.com/index.php?date=10122001
    http://achewood.com/index.php?date=10242001
    http://achewood.com/index.php?date=10302001

  • http://clubneko.net robot makes music

    I must say, this is the last place I thought I’d be seeing links to web comics. I guess Discovery really gets the nerd factor of blogging and hired the appropriate people. Good show!

  • Alex B

    wow i think this says smthn abt human nature. maybe this is why some of us are greedy? i could definitely see how it would help individuals.

  • http://roboticmagazine.com ketanco

    The fact that the algorithm enables robots to develop certain behaviours is very useful in many cases. “Earning points”, “assigning costs” etc… are actually very commonly used in many different applications and this is nothing new. The only thing changing is, as the visual processing, motion mechanisms, hardware etc… evolve, the same principle enables robots to perform different tasks. Playing chess for instance, which has since long been achieved by AI, has the same principle. It assigns costs to moves and then determines the best move. Another example looks very different, and much more complicated, but it is still the same basic principle. Recently the scientists at carnegie mellon university enabled the famous robot ASIMO to walk through moving set of obstacles. Again, the robot uses cost assignment / earning points principle, and based on the assigned costs, it determines the best route for that moment and moves accordignly. To see that article visit: http://www.roboticmagazine.com/androids/019-improved-navigation-for-androids.html . This principle will have much more practical applications, as the robotic hardware and software capabilities evolve. Even for humans it is the underlying principle for many different ways of thinking. In this example the robot knows nothing about “lying” to each other, or in chess the AI knows nothing about a chess game but they are simply trying to earn more points, which is something machines are very good at doing.

  • Chris the Pragmatist

    ketanco I agree with you and it’s very true that people learn in the same manner. Programmed Machines and robots are basically using a rudimentary Cost/Benefit process, the same way people and animals do. As people we learn things in a number of ways but one of the most basic ways we learn things is by remembering an action and the reaction to it. Touch a hot stove with your hand and it burns. We store this information in our computer (brain) and when we see a hot stove again, we don’t touch it. The robots are doing the same thing. The difference being they are modifying their behavior to better suit their want or programmed need. They’ve ‘learned’ how to keep competition away from their food, thus making things better for themselves. Great article and it opens up innumerable possibilities for computers and android research!!!

  • erik

    The next natural step in research is the far more interesting dynamics of group level interaction, change it from individual score to cumulative group score (keeping the variation on the individual level) and throw in a few hundred more generations and watch what happens. I’d definitely look forward to that article.

  • Dave in Calif

    When they start looking like the evil robot in terminator 3, I want one..:-)

NEW ON DISCOVER
OPEN
CITIZEN SCIENCE
ADVERTISEMENT

Discover's Newsletter

Sign up to get the latest science news delivered weekly right to your inbox!

80beats

80beats is DISCOVER's news aggregator, weaving together the choicest tidbits from the best articles covering the day's most compelling topics.
ADVERTISEMENT

See More

ADVERTISEMENT
Collapse bottom bar
+

Login to your Account

X
E-mail address:
Password:
Remember me
Forgot your password?
No problem. Click here to have it e-mailed to you.

Not Registered Yet?

Register now for FREE. Registration only takes a few minutes to complete. Register now »