• Keine Ergebnisse gefunden

6   Conclusion

2.1.1   The concept of situation in HRI

2.1.1.3   HRI as a social situation

In this section it is shown how the situation concept translates to HRI and an approach to describe HRI situations is developed for further analysis. In previous work on this topic, Fischer (2000) has analyzed experimental data in terms of how the subjects conceptualized an HRI situation. She has reported that beliefs about the communication partner showed in linguistic behavior. Moreover, the situation influenced the language the actors used.

„the speakers have been found to constantly define and, if necessary, redefine what they understand the situation to consist in, depending on their current hypotheses about their communication partners and on their own emotional state.“ (Fischer, 2000, p.7)

Thus, Fischer (2000) again points out the person-situation interaction process and its existence in HRI. Users seem to form hypotheses of their robotic interaction partner and attribute occur-rences to the robot as an entity depending on the situation. Hence, the robot seems to be perceived as a social actor acting in a social situation. Goffman (1961) defines the term social situation as follows:

“By the term social situation I shall refer to the full spatial environment anywhere within which an entering person becomes a member of the gathering that is (or does then become) present. Situations begin when mutual monitoring occurs and lapse when the next to last person has left” (Goffman, 1961, p.144)

According to Goffman’s (1961) view, a social situation exists when at least two people share the same space and monitor each other, i.e., they interact. This is also true for HRI situations, even though the actors are not two people but a person and the robot. HRI also often follows the same episode structure as social encounters in HHI (Argyle, Furnham, & Graham, 1981) as the following examples from the home tour scenario show (in parenthesis: utterances of the users):

1. greeting (“Hello Biron”)

2. establishing the relationship, clarifying roles (“I will show you some objects today”) 3. the task (“This is a cup”)

4. re-establishing the relationship (“Now let’s turn to the next task”) 5. parting (“Good bye, Biron.”)

Wagner and Arkin (2008) define social situations in HRI as “the environmental factors, outside of the individuals themselves, which influence interactive behavior“ (p.278). Furthermore, the authors (2008) “currently know of no direct consideration of the theoretical aspects of social situations as applied to interactive robots“, even though most researchers focus on certain situations that occur within the limits of the scenarios. Wagner and Arkin (2008) introduce interdependence theory that represents social situations computationally as an outcome matrix.

In the context of a human-robot clean-up situation, they vary the number of victims and hazards in order to calculate the characteristics of the HRI situation and to predict reward and cost outcomes. However, they do not identify general characteristics of situations. In social psychology several methods for analyzing such characteristics in social situations exist (Argyle, Furnham, & Graham, 1981). Among others these are the dimensional, componential, and the environmental approach and the study of particular kinds of social behavior. The dimensional approach assumes that situations can be described or compared along certain dimensions, whereas the componential approach is based on the belief that social situations are discrete, not continuous entities. This approach seeks to determine the components in each situation type and to understand their relationship. It further assumes that some elements are common to all situations but they take different forms. Hence, situations can be compared at the level of single elements. However, this approach does not result in a comprehensive list of situational elements to explain their interaction. The main concern of the approach is to describe how various components of situations vary across situations. The environmental approach concentrates mainly on the physical aspects of situations and seeks to explain how certain elements of the environment are perceived or related to behavior. Finally, the study of particular kinds of social behavior is a more limited approach. It exclusively seeks to investigate the influence of salient situational variables on a particular kind of behavior.

Argyle, Furnham, and Graham (1981) propose a functional view of situation: people enter situations to achieve goals that satisfy their needs. The authors introduce nine features for the analysis of social situations that will be explained in the following. These features can be viewed as a mainly componential approach which the author finds useful to describe situations and to determine differences between them. In the following, the components are introduced and, thereafter, they are related to HRI.

Goals and goal structure

As mentioned before, Argyle, Furnham, and Graham (1981) view situations as occasions for attaining goals. Most social behavior is goal directed. Knowing the goals of the interactors provides the basis for understanding their behavior. A person can pursue more than one goal at a time. The interrelation of the goals is then called goal structure.

Rules

“Rules are shared beliefs which dictate which behaviour is permitted, not permitted or required” (Argyle, Furnham, & Graham, 1981, p.7). Rules regulate behavior by determining what is appropriate so that the goals can be attained. Some rules emerge during the life of a particular group; others are imposed on the members of a group from above. The rules differ

with groups and situations and can be broken. Argyle, Furnham, and Graham (1981) describe three main kinds of functions of rules: (a) universal functions that apply to all situations, (b) universal features of verbal communication, and (c) situation-specific rules, based on the goal structure of the situation.

Roles

With regard to expectations, roles are defined by the rules that apply to a person in a certain position. They provide the person with a fairly clear model for interaction and require special skills. Roles can change within situations and they are interlocked with certain goals, rules, the repertoire of elements, physical environments, and certain difficulties.

Repertoire of elements

Elements provide the steps needed to attain goals. The repertoires of elements are restricted and vary in different situations. No one set of behaviors exists that covers all situations in detail.

Typical sets of elements for certain situations can be defined and are usually more useful.

However, it is necessary to decide on the units to be coded (for example, complete utterances or parts of them, facial expressions or interpersonal distance in certain time intervals, etc.). In general, verbal categories and bodily actions vary strongly between situations, whereas non-verbal elements are used in every situation to a different degree (Argyle, Furnham, & Graham, 1981).

Sequences of behavior

The elements in a situation may have a certain sequence. Some very common sequences are adjacency pairs, social-skill sequences, and repeated cycles of interaction. Also, the episodic structure of situations is sequential, i.e., the main task consists of subtasks with a special order.

Transition probabilities between actions can be identified using ethological methods (for example, A has a high probability of leading to B, C has a high probability of leading to A).

However, this approach has not been very successful in the analysis of human social behavior because human sequences are too long and too complex and earlier parts of the sequence can influence later parts. Moreover, sequences strongly depend on the situation.

Concepts

According to Argyle, Furnham, and Graham (1981), interactors need shared concepts for handling situations. In social situations, the interactor needs categories in order to classify persons, social structure (role, relationship between people), elements of interaction (for example, classification into friendly/hostile), and relevant objects of attention (parts of the physical environment, task-related objects). The actors need to interpret the behavior of the other interactors and to plan their own behavior.

Environmental setting

Social situations have boundaries of action, i.e., physical enclosures in which behavior takes place, such as rooms. All boundaries contain props (for example, furniture, objects). Each prop

has a particular social function and often a special meaning attached to it. Modifiers are phys-ical aspects of the environment or, in other words, the qualities and quantities of conditioners within the boundaries (for example, color, noise, and odor). Spaces refer to the distances between people or objects which usually convey a certain meaning (for example, being close to another person indicates liking). The environmental setting determines which behaviors are appropriate within it. Moreover, the settings substantially influence the perception of the situation and people’s actions.

Language and speech

As Fischer (2000) found for HRI, how people construe the situation influences their linguistic behavior. Also Argyle, Furnham, and Graham (1981) identified language and speech as one factor for the characterization of social situations. They divide between language, which is the underlying system of grammar, and speech, which is the way people actually talk. Generally, linguistic features are associated with every situation. Some situations constrain language much more than others, for example, because of abilities of the opponent or certain rules.

Difficulties and skills

Some situations are difficult because of the stress that they cause (for example, job interview, experiment situation). Difficulties in social situations are a function of the skills of a person. If a person is skilled, the difficulties are smaller.

These components show that all situations vary along the factors. In the following, different levels of specificity are proposed that can be used to describe an HRI situation and relate the factors to the model which is depicted in Figure 2-1. The levels of specificity that will be differentiated here are:

0. characteristics of situations in general (lowest level of specificity), 1. typical characteristics of HRI situations,

2. scenario-specific characteristics, 3. task-specific characteristics,

4. and characteristics of every specific situation (highest level of specificity).

Characteristics of situations in general are that they have a certain timeframe (a beginning and an end), they are dynamic, and they take place one after another. All situations exist physically in the world around us. Also actors are part of the physical situation. However, each actor perceives the situation individually, based on former experience, personality, and interaction goals. These general characteristics apply to all kinds of situations and also to HRI. HRI situations are one specific kind of situation (just as HHI is another one) with specific characteristics.

Characteristics of HRI situations primarily concern characteristics of the actors – the human and the robot. Similar to HHI, both the robot as well as the human are embodied and share the same physical space. For the situations that contribute to the model, it is further assumed that

Figure 2-1. Levels of specificity of the situation

The degree of specificity increases from the outer to the inner circle.

both actors act autonomously. However, as has been reported in Wrede et al. (to appear), HRI is asymmetric because the robot’s and the user’s modalities and abilities differ. For example, robots without arms cannot use gestures, some robots are not mobile, and most robots cannot perceive smells. Because robots have different modalities and “sensors” to humans, they perceive the situation differently. Apart from the modalities, also the mental abilities vary. Even though a robot might speak with the user, most robots only have a very basic vocabulary and need more time for computation than humans do. The repertoire of elements of the robot in general is restricted to a certain set that, usually, is smaller than the set of a human and also timing is often asymmetric in HRI situations. For this reason, turn-taking in HRI might differ from HHI. The language and speech factor in this case is restricted by the abilities of the robot that forms a part of the social situation as such.

Another characteristic of HRI situations concerns the fact that even though robots are social interaction partners, they are also machines. This is important to keep in mind because humans have different attitudes towards machines than towards other humans. People who are afraid of using a computer (or a robot) are certainly not necessarily afraid of talking to another human.

On the other hand, some people have social relationships with or via their computers but avoid relationships with people in real life. People’s attitudes toward HRI might also be connected to their experiences. To date, only a few people have experienced HRI situations with social robots. That is why they do not have any concrete knowledge about them but need to transfer to HRI what they know from HHI and interaction with other machines.

The next level of specificity depicted in Figure 2-1 is scenarios. As described in Section 1.2, the scenario that serves as an example here is the home tour. The scenario in HRI determines what roles the actors take. The home tour focuses on the teaching of the robot. The general goal is that the robot acquires knowledge about objects and rooms. Hence, the robot is the learner while the user is the teacher. These roles imply rules for the interaction, for example, that the user is free to guide the robot around in order to teach it some object or room and the robot follows.

This also requires that the users are more skilled with respect to the subject, which is the home and objects within it, and that they know how to teach the robot. The scenario also implies that the robot has the skill to learn (or at least to give the feedback that it has learned something) and to drive around. Hence, scenarios are connected to certain abilities of the robot. They are also intertwined with the environmental setting. The boundary of the setting is the apartment. It contains props such as furniture and the objects the robot needs to learn, and modifiers, for example, the color of the walls. These factors with respect to the scenario describe given aspects of the situation.

From an even more specific point of view, situations can be described with respect to the tasks.

On the task level, also the goals and the goal structure are more specific. One goal of the human might be to show a certain object to the robot. Showing the object might include subtasks such as getting the robot’s attention, saying the name of the object, and receiving an appropriate feedback. On this level of specificity, concrete repertoires of elements and sequences of behavior can be identified which will be demonstrated in Chapters 4 and 5. Also typical difficulties can be recognized which will be achieved here with Systemic Interaction Analysis (Sections 3.2.6, 5.2, and 5.3). On the task level, language and speech are even more restricted since the robot only does certain things as a reaction to specific commands, for example, the amount of commands to make the robot follow the user is limited.

On the level of a specific situation all factors describe a concrete part of the interaction with a certain beginning and end. Every situation can be described on this level. The description includes concrete actions, specific elements chosen from the repertoire in a certain sequence, specific goals, a specific environment, and specific difficulties.

Finally, it needs to be considered that most situations researched in HRI are not only HRI situations but also study or experiment situations. Therefore, the humans are not only users of the robot but also participants in the study. In this role, they will probably want to do a good job or to fulfill the expectations of the experimenter. If the experimenter is present, the social behavior of the subjects might change. That is the reason why the experimenter in all the studies presented here remained in the background (see Section 1.4). It was not always possible that he or she would leave the room due to security reasons; actually, the feeling of being alone in a first-contact situation in a foreign environment with a mobile robot might also scare the participants which might influence the interaction more than the presence of the experimenter.

Finally, the study situation could also be stressful because people are in the focus of attention and they might be afraid of failure. According to Argyle, Furnham, and Graham (1981) both these factors cause difficulties. This was considered when designing the user studies in order to keep difficulties to a minimum.

To conclude, it has been shown that HRI can be analyzed as a social situation on different levels of specificity. The lower levels of specificity (situation in general, HRI situation) allow for a general comparison of HRI with other types of interaction (for example, HHI and interaction with virtual agents). The scenario level enables comparison between HRI scenarios, the goals that they include, the roles and the rules that they imply, the skills that are necessary, and the environment that they should be situated in. The task level allows for the comparison of

behavior within and between tasks. Behaviors that are used within the tasks can be identified and compared to behaviors that occur in the context of other tasks. The highest level of specificity, specific situations, allows for in-depth descriptions of situations and for the identification of similarities between situations.

In the following, the levels of situation in general and HRI situations play a role insofar as at various points parallels to HHI are drawn to show relations and distinctions. The scenario level allows to compare different scenarios in HRI such as the home tour (see Section 1.2). The task level and the level of specific situations are the main levels of analysis here, because the goal is to find out what the users do in specific situations, how they try to complete tasks, whether their behavior follows certain sequences, and which difficulties arise.