News Analysis: Why Researchers are Creating New Algorithms for Robot-Human Interaction

Credit: Geralt via Pixabay

April 12, 2018      

Many predictions of the future have humans working with robotic doctors, nurses, teachers, and co-workers. Even your boss may be a robot. But this only happens if the robot-human interaction and communication improve.

Researchers that suggest the current state of interaction between humans and collaborative robots is unsatisfactory, with new algorithms needed.

Current artificial intelligence, and, by extension, how robots interact with humans, is based on unsuitable algorithms, claim some researchers. Collaborating with humans puts robots in dynamic and unknown situations that they not programmed for.

Physical interaction with humans requires robots to adapt their behavior in socially acceptable ways by understanding and interpreting human communication, both verbal and non-verbal. Scientists said this requires new AI algorithms.

Suboptimal robot-human interaction can have life-or-death consequences. What if the robot doctor doesn’t look beyond what you tell it your symptoms are and takes you at your word — something human doctors learn quickly not to do? Or a robotic police officer misinterprets your non-verbal cues, arresting you and letting a real criminal get away?

New skills required for robot-human interaction

In his paper, “Artificial cognition for social human-robot interaction: An implementation,” Séverin Lemaignan of Plymouth University said several individual and collaborative cognitive skills are needed for a robot to “successfully share space and tasks with a human.” These skills include:

  • Geometric reasoning and situation assessment based on perspective-taking and affordance analysis
  • Acquisition and representation of knowledge models for multiple agents (humans and robots, with their specificities)
  • Situated, natural, and multi-model dialogue
  • Human-aware task planning
  • Human-robot joint task achievement.

No ‘natural’ communication

Another researcher said communications with robots is not natural – humans will fall into a trap if we trust that “if our instructions are clear enough, it will all work out.”

AI researcher Nikolaos Mavridis said the changing design of robots – from computer boxes to more humanoid shapes – prompts people to communicate “naturally,” assuming that the robot will understand the context of a question or remark, as people often do. However, robots are typically unable to define and use such contexts – at least for now, Mavridis said.

Ten points of attention

In designing effective human-robot interaction (HRI), Mavridis identified 10 types of communicative interaction:

  • Breaking the “simple commands only” barrier
  • Multiple speech acts
  • Mixed initiative dialogue
  • Situated language and the symbol grounding problem
  • Effective interaction
  • Motor correlates and non-verbal communication
  • Purposeful speech and planning
  • Multi-level learning
  • Utilization of online resources and services
  • Miscellaneous social abilities

Human-robot armed forces

Effective robot-human interaction is also crucial in military situations, leading researchers to create new algorithms for interaction between robots and soldiers. The U.S. Army Research Laboratory recently concluded that new algorithms are needed to enable robots to properly execute orders given by human instructors. The study results were presented in February 2018 at an AI conference in New Orleans.

Garrett Warnell

Dr. Garrett Warnell speaking at a recent AI conference. Credit: U.S. Army

The algorithm, known as Deep TAMER, “provides robots with the ability to learn how to perform tasks by viewing video streams in a short amount of time with a human trainer.”

Dr. Garrett Warnell, research project leader, said the army of the future will include soldiers and autonomous robots working together. While both groups can be trained in advance, teams will likely be asked to perform tasks such as search and rescue or surveillance in environments they haven’t seen before. Humans can quickly generalize their training, but current AI agents cannot, researchers said.

“If we want these teams to be successful, we need new ways for humans to be able to quickly teach their autonomous teammates how to behave in new environments,” Warnell said at the conference. “Deep TAMER, which requires only critique feedback from the human, shows that this type of real-time instruction can be successful in certain, more-realistic scenarios.”

Can a robot explain how it’s feeling?

Humans often ask each other, “What are you thinking about?” The reply often comes back, “Nothing,” as a way to avoid having to explain true feelings. Something similar may happen in future robot-human interaction.

AI levels are becoming so advanced that they may eventually surpass humans in many areas, claimed the researchers. Smart robots won’t need human instructions; they’ll know better and act accordingly.

To avoid counterproductive situations, robots will have to explain their thoughts and actions in a way that’s understandable to humans. Researchers said AI will need to understand the human psyche, including individual differences, collective contexts, and verbal and non-verbal communications.

At the same time, humans will need to accept that AI-powered robots are as complex as humans, if not more so, and try to understand them as well. Using new algorithms, AI will help robots achieve this. Will humans then learn to live and work with their artificial companions?