Tons of breakthroughs in robotics

Seriously, I am always astonished by the fields of robotics and artificial intelligence.  Here are a few really cool things that robots have been doing recently:

First, a robot has been designed so that it can essentially think for itself.  It is able to use reasoning and past experiences to make educated guesses as to how it should act in novel situations.  It implements a technology called Self-Organising Incremental Neural Network (SOINN).  I’m not surprised at all that this feat is accomplished by a neural network rather than traditional AI (GOFAI).  GOFAI is notoriously fragile and can only take “complete” inputs to give and output.  For example, if a robot using GOFAI is given the task of walking around a room but has not been instructed in how to navigate obstacles, once it reaches the first obstacle it’ll essentially be paralyzed and unable to cope with the situation.  Neural networks, especially ones implementing back-propogation (essentially, feedback from an outside source, such as a lab technician), are able to “learn” what the correct output for a given input is through trial and error rather than hard-coding it.  This new SOINN robot is able to self-organize the data it has collected from previous experiences in order to come up with a solution for novel problems.  It is able to ask for help if it’s completely stumped, and it remembers how to complete that task and is able to use that experience on a future task.

http://www.wired.com/wiredscience/2011/08/robot-thinking-autonomy/

Second, robots in a test environment have “evolved” altruism.  Robots in a Swiss lab were tasked with finding “food”, and those that successfully found food had their computerized “genes” mixed up with the other successful foragers, creating the next generation of robots.  The robots were also given a choice of sharing the points awarded for finding food, thus enabling less successful robots to survive and pass on their computerized genes as well.  In different generations and iterations of the experiment, there were costs and benefits for sharing thrown into the equation.  What the researchers found was that the robots acted in ways that were predicted by Hamilton’s equations.  Hamilton attempted to outline why altruism makes sense evolutionarily, but it’s difficult to measure the development and evolution of altruism in living creatures because we’d have to measure hundreds and hundreds of generations.  With these robots, scientists were able to really speed up “evolution” and take note of how altruism evolves amongst (synthetic) organisms.

http://www.wired.com/wiredscience/2011/05/robot-altruism/

Third, (and most terrifying) robots have been taught how to use deception.  Researchers at Georgia Institute of Technology have provided a robot with the capacity to deceive, as well as an algorithm to help it determine when deception should be used.  The researchers tested robot deception by having two robots play hide and seek.  The black robot was supposed to hide in one of three locations, and the red one was supposed to find the black one based on color-coded markers knocked over by the black robot leading to its hiding place.  The black robot would trick the red one by walking towards one hiding place (and passing the color markers on the way), then changing directions and hiding somewhere else.  The black robot tricked the red one 15 times out of 20, the other 5 times were due to the black robot’s difficulty knocking over the correct markers.  There are, of course, lots of ethical implications related to teaching a robot/killing machine how to deceive…But the Georgia Institute of Technology researchers feel that it’s an important field to continue research in, and one that has many ethical benefits as well as negative repercussions.

http://www.wired.com/wiredscience/2010/09/robots-taught-how-to-deceive/

Related Photos