Trash-Talking Robot

Trash Talk: Disparaging, taunting or boastful comments especially between opponents trying to intimidate each other (Merriam-Webster). OK, but what if your trash-talking opponent is a robot?

Welcome back. Looking ahead to our future with robots, researchers with Carnegie Mellon University conducted a study of the effect of linguistic nuances and social behavior on human-robot interactions.

Rather than add to the large body of work on cooperative interactions (e.g., robots assisting humans), they examined the case of robots and humans having different or even conflicting objectives. This would occur, for example, if a sales robot is programmed to convince a human customer to buy a specific item, but the customer thinks other items would be better.

The question addressed by the study was how would an opponent’s comments about one’s game playing ability impact a human if the opponent were a robot?

Pepper, the humanoid robot that voiced encouraging and discouraging comments during a strategic game with humans (www.softbankrobotics.com/emea/en/pepper).

Let the Games Begin
The researchers enlisted 40 participants to compete against a humanoid robot in a strategic game (Stackelberg Security Game). The participants were not initially advised of the true purpose of the study. Each participant played two practice rounds of the game without the robot, then 35 rounds with the robot.

During the 35 rounds, the robot played with optimal strategy and offered expressive verbal commentary either encouraging or discouraging its human opponent. Whether encouraging or discouraging, the comments had nothing to do with the participant’s actual performance.

A selected group of the participants also played another 35 rounds, during which the robot’s comments were opposite to its comments during the first 35 games with those participants; encouraging comments became discouraging and vice versa. 

Examples of the robot’s encouraging and discouraging
comments when playing a strategic game with human
opponents (from arxiv.org/pdf/1910.11459.pdf).

The researchers collected a variety of data on the participants’ game strategy and their perceptions of the task, their performance and the robot. These included a pre-task questionnaire of demographic information, familiarity with robots and technology, and emotional state; records of actions taken during the game; a post-task questionnaire; a post-game verbal semi-structured interview (video recorded); and, for some participants, video of the participant playing the game against the robot.

Did the Robot’s Comments Matter?
The study found that, overall, the robot’s comments strongly influenced the participants’ feelings in the two-opponent, competitive interaction. This occurred regardless of the participants’ level of technical sophistication.

Discouraging comments caused the humans to play the game less rationally and perceive the robot more negatively--as being less optimistic, cheerful, cooperative and cute. In contrast, encouraging comments caused the humans to play the game more rationally and perceive the robot in a more positive manner.

In all, 30% of the participants labeled the robot’s goal as “distraction,” but they excused the robot, blaming its behavior on how it was programmed.

Wrap Up
The researchers suggest their findings may be useful for robot designers. The robot’s ability to prompt responses could have implications for automated learning, mental health treatment and even the use of robots as companions.

Game developers can also use the results to develop more interactive opponents and increase the sense of engagement.

Placing importance on the humanoid aspect of the robot in their study, the researchers posit that nonverbal modes of expression in competitive settings should be investigated in future work.

Won’t it be swell when robots as well as people start giving us a hard time? Thanks for stopping by.

P.S.
Study of effect of robot’s comments on human game opponents in Proc. of 28th IEEE Int’l. Conf. on Robot Human Interactive Communication: arxiv.org/pdf/1910.11459.pdf
Article on study on EurekAlert! website: www.eurekalert.org/pub_releases/2019-11/cmu-tth111819.php