Bottom-up learning of feedback in a categorization task

Abstract

We designed a laboratory study to investigate the influence of social interaction on category learning. The objective in the present study is to examine what kind of teaching behavior can improve an agent's learning of categories. In a computerbased study participants learned four categories for sixteen objects which appear on a computer screen. The objects' categories determine what kind of manipulation is to be done on the objects. Five tutors and twenty participants were recruited to participate. For the study the tutors were placed in front of a computer in one room whereas the learners were in another room. The learners' task was to manipulate the objects appropriately through the instructions they received from the tutor on their screens via six symbols. These six symbols were the only way for the tutor to communicate with the learner. We call this a bottom-up learning as the it relies entirely on the perception of the tutors' symbols without any prior knowledge of their meaning. The focus in the present study is not on the ability by the learner to acquire knowledge of the categories but on the types of instructions that the tutor gave during the trials and the effects of the feedback given to the learner. Therefore, the feedback given by the tutors via the symbols was classified and quantified.

Studying the co-construction of interaction protocols in collaborative tasks with humans

“…[11]). Two studies which do consider asymmetric communication are the studies conducted by de Ruiter et al [12] and Griffiths et al [13].…”

“…The setup of the study conducted by Griffiths et al [13] is more directly related to our setup. It is based on the alien world game setup by Morlino et al, in which in a square world shown on a computer screen, positions (left or right) and movements (shake horizontally or shake vertically) of 16 objects have to be explored via a mouse to maximize a score [14].…”

Improving interactive reinforcement learning: What makes a good teacher?

“…However, Taylor et al (2011) gave some interesting insights about a humanagent interaction approach. Also, Griffiths et al (2012) studied different teacher behaviors to improve the apprenticeship in learner-agents. Although their experiments are based on human-human interaction, they have used tutors that have mastered a given task without any classification about the level of expertise.…”

“…As part of this study, we evaluated the involved interaction parameters namely probability of feedback (L), consistency of feedback (C), and whether the learner follows the received advice or not in order to mimic actual human-human behavior where the learner occasionally does not follow the advice (Griffiths et al, 2012). We called this parameter learner obedience O ∈ [0, 1], 0 being an agent that never follows the advice and thus corresponds to a pure RL learner.…”

“…Nonetheless, to the best of our knowledge, there is no study so far about the implications of utilizing artificial teachers with different characteristics and different internal representations of the knowledge based on their previous experience. Moreover, the effects when the learner ignores some of the advice has also not been studied in artificial agent-agent interaction, although some insights are given in Griffiths' work using human-human interaction with a computational interface (Griffiths et al, 2012).…”