Researchers Teach Robots Deception by Mimicking Squirrels
Georgia Tech researchers Ronald Arkin and Alan Wagner pose with robots that have been trained to deceive each other by studying the behavior of squirrels trying to protect their food stash. The work funded by the Office of Naval Research could eventually be used by the US military. (Source: Georgia Tech)
In my coverage of robots, it always interests me how researchers try to mimic the natural world to create robots not just for movements, but also for artificial intelligence. Squirrels are a clever choice for this type of deceptive behavior.
True! It's kind of like art imitates life imitates art...so here, robots imitate nature imitate robots...? Actually, I suppose since it's humans teaching robots how to think, perhaps we project our robotic tendencies onto the machines themselves...
Thanks for this fun article, Elizabeth! Biomimetics has long been an intriguing area of research. Quorum Sensing in bacteria is one of the latest discoveries that may well revolutionize how we deal with infectious disease. The folks at Georgia Tech are doing great work. A major difficulty arises when the robots playing hide and seek are not limited by the relatively few senses used by a squirrel. Add multispectral vision, penetrating radar, and residual trace chemical detection and it will be very difficult to hide from the seeker. And then you add a swarm of tiny seekers that just flood the area and check out every hiding place in parallel, and the game is no longer fun. =]
I think we all like to think that we do things in some sort of logical fashion that we can then project onto robots. The problem is occasionally we have one of those brain duds and what we do doesn't have any logic to it at all. We also have the fact of making a mistake that ends up defying logic and working out really well. Robots, in the end, will always do exactly what they were programmed to do. Whether it be a learning program or simple logic, it will only do what it is told to do.
Interesting article, Liz. It's amazing to see the strategies that are being used to teach machines to learn and think. A couple of years ago, we ran a story about robots using humor to make themselves more acceptable to humans. To do that, the robots had to understand humor and have a vague understanding of what's "appropriate," which, of course, is a big part of humor. I can't imagine what robots will be capable of in 50 years.
Thanks, Charles. Humor in robots! Now that would be something. Sometimes you can't even find that in human beings. :) Joking aside, I am with you. I hope I am still around in about 50 years to see how robots have evolved. But I bet it won't even take that long before they are sophisticated beyond most people's imaginations.
Thanks, Charles, I will check that out. I am quite interested to see how this would be done, considering how difficult humor is even for humans who speak the same language, not to mention it is the most difficult thing to translate in other languages. Appreciate the info!
It's amazing how much effort in the area of artificial intelligence is trying to create what the world already does. The difficulty with this is the fact that we occasionally do something that is not predictable and how can an algorythm ever predict something that just comes out of no where.
Yes, jmiller, this is true...and I hope this is what will always separate humans from robots! But robot design is getting so sophisticated that who knows, maybe sometime there will be artificial intelligence even for that! Personally, I hope not, but the mind boggles...
I wonder if they'll still call it artificial intelligence when the computer is programmed to have a blonde moment...no offense to any blondes. It is scary to think about how much artificial intelligence has come in the last few years. In college I remember they used to test aretificial intelligence by some test and computers improved so much that they had to change the test.
Thanks, Elizabeth--this is very funny, as well as an intelligent use of biomimicry in research design. Squirrels are very sly creatures. It's also interesting that the biomimicry here is now aimed at behavior, not movements, as you point out. That's a sign that design is making progress past some simple basic problems, and moving into a different phase.
Indeed, Ann. Movements are one thing that engineers have really made significant progress on in terms of robots...behavior really is the next wave now. It's interesting when they use animal behaviors, I think, because humans can never really fully understand animals, if you think about it. They can observe them and get a good idea of the how and why of behaviors, but we don't really know what's going on in an animal's mind. To transfer that to robots is quite a challenge in its own right, I think! Will be interesting to see how this all develops. It seems like there is something new every day!
The movements phase took a long time. Much initial research was simply fundamental work trying to see all the different ways you could design a robot that moves. Later, that went 180 degrees and design of movements was targeted to specific uses, such as much of Boston Robotics' work. That seems to be about the time that more military funding was becoming available.
That's an interesting bit of history. It seems now that a lot of the behvaioral type of robot innovation seems to be coming out of research institutions and universities vs. the military. In fact, I think also it's becoming the same for research into robotic movement, too.
Ah, really...I wasn't sure about that. I guess it's not always so transparent who is paying for the research. I also have covered a few things in Europe and sometimes it's funded there by European commissions or consortiums rather than the military.
You're right, it's not always obvious where funding is coming from. And I've noticed the same difference in sources of funding between Europe and the US for robotics research. European governments and consortia are also more likely to provide funding for research in other areas, such as bioplastics or biofuels.
Why chase squirrels when you can simply pick up a book on WWII military intelligence? For example, before D-Day, the RAF was tasked with knocking out the German radars on the landing sites. So as not to give away the sites, the RAF targeted 3 other sites for every sortie on the designated beaches. These robots are basically carrying out a similar tactic. What the robots can't do is change tactics once they know that the other side is "on" to them - their behaviour, although irrational, is predictable.
Researchers said the hiding robot "was successful at deceiving the other robot in 75 percent of tests" but can they deceive a second time? Or do we end up with dueling algorithms where the second robot is learning from its mistake .. and they battle to a standoff?
It won't be too much longer and hardware design, as we used to know it, will be remembered alongside the slide rule and the Karnaugh map. You will need to move beyond those familiar bits and bytes into the new world of software centric design.
Focus on Fundamentals consists of 45-minute on-line classes that cover a host of technologies. You learn without leaving the comfort of your desk. All classes are taught by subject-matter experts and all are archived. So if you can't attend live, attend at your convenience.