Rules, events and intentions relating to ‘behaving’ artifacts: Exploring young children’s anchors to understanding while learning to program a robot Levy, S.T., Mioduser, D., Talis, V. Tel-Aviv University Paper presented at the 33rd Annual Meeting of the Jean Piaget Society, Play and Development, Chicago, IL, USA (2003). ABSTRACT This paper describes a study aimed at depicting 5-6 year old children’s changing understanding of controlled systems, as they construct them over an extended period. It was found that through extended constructive play, young children could extract the rules underlying a robot’s behavior in a way that can then be used independently of the events within which they are embedded. Ascribing intentions to the robot did not change with experience. However, these intentions later provide a bridge and meaning to the more extensive technological rule-building. With experience, more rules can be used to reason about such devices.

INTRODUCTION This paper describes a study aimed at depicting 5-6 year old children’s understanding of controlled systems, as they construct robot behavior over an extended period. The conference theme of Play and Development is an opportunity to consider children’s play while constructing complex self-regulating systems, and to relate the interactions between such play and learning. We explore these interactions with a toy that stands on the brink of animate and inanimate – a moving and adapting programmable Lego creature. An important aspect of these activities is the excitement that they provoke, a large gateway to learning the underlying causal structures and its overall behavior. Banking on this engagement, we’ve designed a robotic system that young non-reading children may use. While we are not the first to devise such systems (Papert, 1993; Robots for Kids, 2000) ours diverges from prior designs in an important aspect related to the way control knowledge is represented – as time independent non-sequential rules, rather than the usual time-dependent flowchart-like actions and rules (Levin & Mioduser, 1996). Developmentally, scripts (time-dependent) are earliest to emerge (Flavell et al, 1993). However, it is also suggested that in the domain of artifacts, children can operate at more mature levels of understanding (Piaget, 1956; KemlerNelson, 1995). Another important aspect that arises when we consider ‘behaving’ artifacts that seem to have a mind of their own is the animistic view children might hold (Ackermann, 1991; Turkle, 1984). Thus, we explore the children’s attribution of intentions to artifacts.

1

In addition, we investigate the complexity of the rule structures that the children can describe and program, as well as the role of an adult’s interventions in scaffolding this process. Thus, we’ve conducted a study examining the “rule-thinking” of young children and its development in a real-world constructive environment - a computer-controlled robot, traversing a modifiable terrain.

METHOD The sample included 6 children, 3 boys and 3 girls, selected randomly out of 60 children in a public school in the Israeli city of Rishon-LeZion. Their ages spanned 5y4m- 6y0m. A computerized control environment was developed as well as a progression of increasingly-complex tasks. Each child participated individually in 5 videotaped sessions, spaced about one week apart, each lasting 20-45 minutes. During the sessions, the child was asked to describe a given robot behavior and was then invited to program the robot to achieve a specific behavior. This paper focuses on the first: the children’s descriptions of a given behavior. Two types of intervention were employed: ‘jump-starting’, the same question is asked in different forms, encouraging elaboration of the initial version; ‘decomposing’, when differentiation is not made, various questions are aimed at such separation. The children’s descriptions and definitions of the robot’s behavior were analyzed using as a framework the conceptual model of a rule structure. It includes two ways of defining the conditions and actions for a robot’s rule and their mapping. One is technological (input-operation) and the other psychological-intentional (circumstancesbehavior). In addition, we analyze the children’s representation of the flow of events: from an account of sporadic episodes, a pattern of repeating behaviors (a script) or a timeindependent description of interacting rules.

RESULTS

Intentional-psychological versus technological For most tasks, the initial description was psychological-intentional, such as: "he looks all the time for food", or "he does not understand what is white". However, with supporting intervention, all children generated technological descriptions (e.g., "on the white [area] he turns and then goes forward"). This gradual shift can be seen within each task. After the intentional description has been sufficiently defined, the technological description arises, while preserving the original target behaviors. With experience, technological descriptions are more frequent and elaborate.

Rules, episodes and scripts From the start, the children’s descriptions are mainly rule-like structures of general actions or condition-action pairs.

2

Episode-like descriptions are transient. When a child is confronted with confusing robot behavior, and cannot construct a suitable rule, she may focus for a while only on the robot’s actions, following them from moment to moment (episode-type). The conditions are first ignored, and are later incorporated. The use of script-like descriptions develops with practice and gradually leads to the more general, time independent and abstract rule-like descriptions. It becomes a strategy of choice in analyzing the robot’s behaviors. The identification and coding of the actual input/output components of the robot's functioning, results in the formulation of technological rules that are independent of time.

Complexity As a particular task evolves, description moves from focusing on simple behaviors (one condition-action pair) to the consideration of a compound of a number of behaviors (functional chunks), as well as of relevant contextual information. This process does not happen to a child independently of interaction with an adult. Most children provided more complex descriptions when an adult intervened and helped them decompose the situation at hand. Without intervention, descriptions were limited to 2 condition-action pairs. With a decomposing intervention, the children were able to verbalize more advanced rule structures. Eventually, most children reached a complexity level in their descriptions close to the actual complexity of each task.

DISCUSSION This study examined young children's rule-thinking regarding the functioning of a robot in tasks of increasing complexity. In our study children were asked to describe and explain the functioning of technological systems behaving in time and space, while adapting and responding to changing features in the environment. We conclude by describing the anchors and drifts in the children understanding of such self-regulating robots.

What anchors children’s understanding of the robots? One is an a priori knowledge that even if the robot’s behaviors are complex and confusing, they are nevertheless rule-based. Thus they search for time-independent rules from the start and don’t get caught up in the details played out in a particular event. Another is their search for familiar human-like intentions in the moving artifact’s actions. Such intentional descriptions serve to frame their understanding. They then bridge to a more complex understanding of the underlying technological rules of the artifact.

What changes with experience? The number of rules the children can extract or deduce on their own increases with practice in programming the robot. In order to deconstruct the robot’s behavior, the children develop a strategy of temporarily constructing a script in the search for the over-arching rule patterns. Thus, it would seem that the children are well-equipped with important notions about the artificial world: its being rule-based and related to volition. This serves to frame

3

their learning that increases complexity in understanding and strengthens successful strategies.

REFERENCES

Ackermann, E., (1991). The agency model of transactions: Towards an understanding of children's theory of control. In J. Montangero & A. Tryphon (Eds.), Psychologie genetique et sciences cognitives. Geneve: Fondation Archives Jean Piaget. Kemler Nelson, D. (1995). Principle-based inferences in young children's categorization: revisiting the impact of function on the naming of artifacts. Cognitive development, 10, 347-380. Levin, I., and Mioduser, D. (1996). A multiple-constructs framework for teaching control concepts. IEEE Transactions on Education, 39(4), 1996 1-9. Papert, S. (1993). The children's machine - rethinking school in the age of the computer. New York: Basic Books. Piaget, J. (1956). The Child’s Conception of Physical Causality. Littlefield: Adams Co. Robot for Kids (2000). Druin, A., Hendler, J. (Eds.). Morgan Kaufmann Publishers. Turkle, S. (1984). The Second Self: Computers and the human spirit. NY: Simon and Schuster. Acknowledgment: We thank Diana Levi, our graduate student, for her meaningful and insightful contribution to this study.

4