The ability to understand and use causal relationships is fundamental to a range of activities that we associate with human intelligence, such as problem-solving, decision-making, prediction, and creativity. Humans are far superior to today's robots and other AI systems in their ability to detect and exploit causal relationships, in part because machines lack the ability to distinguish between correlations and true causal relationships.
A simple example is data showing daily sales and electricity consumption in a supermarket. On hot days, both sales and electricity consumption increase, and are thus correlated. However, there is, of course, no causal relationship that would mean that ice cream sales would increase if electricity consumption increased, or vice versa. We humans easily understand this, but it is much more difficult for a machine.
The ROCC project aims to develop techniques so that robots can learn, and use causality. What makes the research unique is that it combines traditional statistical methods with human-robot interaction. First, the robot observes what is happening in the world, and then it asks questions of the people it is interacting with. These questions are designed to clarify correlations and causality, thus avoiding the problem described in the example above.
The result is a robot capable of better understanding cause and effect. This capability could be used in a number of different ways. The robots can make better plans to achieve set goals because they know more about what happens when they perform different types of actions. The robot can also better understand humans because it understands the purpose and goal of the things humans do. The research in the project consists of algorithm development of statistical methods, implementations on physical robots, and user studies to determine how we humans best interact with robots that understand causal relationships.