NEWS Are there "smart" robots? Not today. But now researchers at Umeå University will try to change that in a new collaboration between Computing Science and Umeå School of Business, Economics and Statistics. – We will develop techniques so that robots learn what we sometimes call "common sense" with the help of us humans, says Thomas Hellström, Professor at the Department of Computing Science.
Suna Bensch, Associate Professor in Computing Science, Filip Edström, PhD student in Statistics and Professor Xavier de Luna, Umeå School of Business, Economics and Statistics, and Thomas Hellström, Professor in Computing Science, are now collaborating to make robots understand cause and effect.
ImageVictoria SkeidsvollThomas Hellström is in charge of the new ROCC project, "Robots with Causal Capabilities", funded by the Swedish Research Council. The aim is to develop the robots' ability to see causal relationships and at the same time figure out what actually doesn't matter.
"This is essential for problem-solving, decision-making, and prediction, as well as creativity. In this respect, we humans are completely superior to today's AI systems," says Thomas Hellström.
He is a professor in computing science with a focus on robotics and leads the Intelligent Robotics research group at Umeå University, a team that has both coordinated and participated in a number of national and international projects over the past few years.
So, what do we mean by causal abilities? A simple example might be a supermarket and the amount of ice cream sold in a day, versus the store's electricity consumption at the same time.
"If it's hot, both ice cream sales and electricity consumption increase, so they are correlated, but there is no causal relationship that makes ice cream sales increase if electricity consumption goes up. We, humans, understand this – but unfortunately not today's computers and robots," says Thomas Hellström.
Another example is a service robot working in the dining room of a home for the elderly. The robot observes how Jan, after eating lunch, often sits down next to Julia on the couch for a chat. After a while, the kitchen door opens and the coffee cart rolls into the room. Julia claps her hands and smiles delightedly, while Jan moves to the dining table to drink coffee there.
"An intelligent robot must understand that it was the coffee that made Julia smile, and not Jan leaving the couch, i.e. it must understand the difference between observed correlations and real causal relationships," says Hellström.
Together with Suna Bensch, Associate Professor at the Department of Computer Science, Xavier de Luna, Professor at Umeå School of Business, Economics and Statistics (USBE), and Filip Edström, PhD student at the Statistics department at USBE, Thomas Hellström will now be teaching the robots how to do this.
"In this project, robots will talk to humans and ask questions about how the world works. Does leaving the lights on at night increase ice cream sales? Or is the increase due to something else? Hellström says.
Combining observations with questions and answers, the robot can gradually learn to understand cause and effect, a capability that can be used in many different ways.
"The robot can better make plans to achieve set goals because it knows more about what happens when it performs different types of actions. It can also learn to better understand us, humans, because it understands the purpose and goal of the things we do," says Thomas Hellström.
One research question is how the robot decides which questions to ask. Another is what statistical methods it should use to draw the right conclusions from these answers.
In the ROCC project, the researchers will also investigate how people experience and best interact with robots that can understand and make use of causal relationships.
- Yes, just as we can get annoyed with children who incessantly ask "why?" we will probably get annoyed with robots who do the same!
The project will require a combination of expertise in robotics, linguistics, and statistics.
"The transdisciplinary environment within TAIGA, Umeå University's new AI initiative, provides valuable support for this type of project, which would not be possible within individual research areas, concludes Thomas Hellström.
