Robots and computers could soon be having meaningful conversations and even arguments with humans.
A new research project will develop systems that allow men and machines to debate decisions. It opens up the possibility of human operators discussing action plans with robots and, if necessary, ordering them to break rules.
For their part, the computers would be able to argue in favour of decisions or inform their operators that certain tasks are impossible. Early versions of the software could be available in just three years. Lead researcher Dr Wamberto Vasconcelos, from the University of Aberdeen, said the aim is to increase human trust in intelligent technology.
"Autonomous systems such as robots are an integral part of modern industry, used to carry out tasks without continuous human guidance," he said. "Employed across a variety of sectors, these systems can quickly process huge amounts of information when deciding how to act. However, in doing so, they can make mistakes which are not obvious to them or to a human.
"Evidence shows there may be mistrust when there are no provisions to help a human to understand why an autonomous system has decided to perform a specific task at a particular time and in a certain way. What we are creating is a new generation of autonomous systems which are able to carry out a two-way communication with humans."
Talking computers with the ability to converse with humans have long been a mainstay of science fiction. Examples include Hal, the deadpan-voiced computer in the film 2001:A Space Odyssey, which goes mad and sets out to murder the crew of a spaceship.
The system Dr Vasconcelos is developing will communicate with words on a computer screen rather than speech. Potential applications could include unmanned robot missions to planets or the deep sea, defence systems and exploring hostile environments such as nuclear installations.
A typical dialogue might involve a human operator asking a computer why it decided on a particular decision, what alternatives there might be and why these were not followed. "It gives the human operator and opportunity to challenge or overrule the robot's decision," said Dr Vasconcelos. "You can authorise the computer system to break or bend the rules if necessary, for instance to make better use of resources or in the interests of safety.
"Ultimately, this conversation is to ensure that the system is one the human is comfortable with. But the dialogue will be a two-way thing. The supervisor might not like a particular solution but the computer might say: sorry, this is all I can do."
Conversing with robots would actually make humans more accountable, since failures could not conveniently be blamed on computer error, Dr Vasconcelos added. "With power also comes responsibility. All these dialogues are going to be recorded, so there is a name to blame if something goes wrong. It's a good side-effect."