Why Robots Should Be Social: Enhancing Machine Learning through Social Human-Robot Interaction

Abstract

Social learning is a powerful method for cultural propagation of knowledge and skills relying on a complex interplay of learning strategies, social ecology and the human propensity for both learning and tutoring. Social learning has the potential to be an equally potent learning strategy for artificial systems and robots in specific. However, given the complexity and unstructured nature of social learning, implementing social machine learning proves to be a challenging problem. We study one particular aspect of social machine learning: that of offering social cues during the learning interaction. Specifically, we study whether people are sensitive to social cues offered by a learning robot, in a similar way to children's social bids for tutoring. We use a child-like social robot and a task in which the robot has to learn the meaning of words. For this a simple turn-based interaction is used, based on language games. Two conditions are tested: one in which the robot uses social means to invite a human teacher to provide information based on what the robot requires to fill gaps in its knowledge (i.e. expression of a learning preference); the other in which the robot does not provide social cues to communicate a learning preference. We observe that conveying a learning preference through the use of social cues results in better and faster learning by the robot. People also seem to form a "mental model" of the robot, tailoring the tutoring to the robot's performance as opposed to using simply random teaching. In addition, the social learning shows a clear gender effect with female participants being responsive to the robot's bids, while male teachers appear to be less receptive. This work shows how additional social cues in social machine learning can result in people offering better quality learning input to artificial systems, resulting in improved learning performance.

Fig 1. Schematic illustration of the language game flow.

Both teacher and learner examine a shared world-view consisting of images. The teacher chooses one image as the topic and communicates an associated linguistic description to the learner. The learner tries to guess which image the teacher has in mind and receives feedback on its guess from the teacher. Based on this feedback the learner modifies its word-meaning associations.

Fig 2. The LightHead robot.

A semi-transparent mask mounted on a 6 DoF robotic arm, in which an animated character is projected.

Fig 3. Overview of the experimental set-up, showing the LightHead robot, the touchscreen used to play an interactive learning game and the participant.

The individual in this image has given written informed consent (as outlined in PLOS consent form) to publish these case details.

Fig 4. Difference in robot learning performance between social and non-social group.

Error bars indicate 95% confidence interval, ‘*’ indicates a significant difference with p = 0.0328, two sample t-test.

Fig 5. Distribution of participants’ category choices in social and non-social condition, compared to the dataset distribution.

Error bars indicate 95% confidence interval.

Fig 6. Social-responsiveness.

The responsiveness to the robot’s social cues plotted against the robot learning performance. Straight lines depict a 33% baseline of following the robot’s preference by chance (blue) and the mean value of social responsiveness (green).

Fig 7. Mean trends of robot learning performance; data is split based on gender and social condition.

Error bars indicate 95% confidence interval.

Fig 8. Mean trends of robot learning performance at the end of interacting with the robot.

Numbers are split based on gender and social condition, error bars indicate 95% confidence interval.

Fig 9. Q2: Participants’ rating of robot behaviour in terms of naturalness.

Error bars indicate 95% confidence interval.

Fig 10. Q4: Participants’ rating of robot behaviour in terms of who was in control.

Error bars indicate 95% confidence interval.

Fig 11. Q8: Participants’ rating of robot behaviour in terms of how smart the robot is.

Error bars indicate 95% confidence interval.