Short Review: DAUTENHAHN, Kerstin [1997] I could be you - The phenomenological dimension of social understanding

1. Starting with the assumptions (A1) that the identification of conspecifics and (A2) imitation are two crucial points in the construction of social robots (p.3) she was challenged by the questions, what does make up a social interaction situation? and What makes the difference between the exchange of communication signals and understanding? Additionally she assumes (A3) that these mechanisms do develop (p.4) and that (A4) experiental bodily understanding plays a crucial role (p.4).
   
   
2. Looking to whole societies of organisms she distinguishes between anonymous societies (like in the case of insects) and individual societies (p.4). In D's notion of social intelligence is the focus of attention the directed interaction between individuals (p.5).
   
   
3. From a methodological point of view DA. distinguishes the phenomenological approach from a computational approach (p.5). And she states "In this [phenomenological] conception of understanding we consider each other's action and words not as physically caused, but as a dialogical relationship in which we interpret meaning from each other's gestures on the basis of our reality (continuous, lived experience in humanly meaningful contexts)." (p.5). It is her intention, to overcome the computationalistic/ phenomenological gap, to mediate between the 'inside' and 'outside' dimension of embodied cognition by the construction of embodied robotic agents (p.5). Later on she states "Empirical studies addressing phenomenological phenomena could give us insight about mechanisms like empathy, autobiographic memory and embodiment which could help us to bridge the gap between the two domains" (p29). And she continues: "The better we understand the human psychology and human internal dynamics the more we can hope to explain embodiment and empathic understanding on a scientific basis. This knowledge can then be applied to artifacts. It is necessary to ground the computationalistic, conceptual approach to (social) understanding in empathic and experiental dynamic processes within the system. Without such a grounding we are in danger of building systems with much less social expertise than autistic people or even dogs" (p.29).
   
   
4. But then DA. Concludes "Because of the different nature of robot bodies, experiences, and internal processes, the phenomenological world of humans and robots will, in my view, always be different. Nevertheless, the way different species of animals can communicate shows us that interspecies communication can in fact work" (p.29).
   
   
5. DA. establishes now the concepts of Sympathy and Empathy (p.7). In D's view is empathy a social, interpersonal means of 'mind-reading' by pushing oneself into a state of engagement in other persons' psychological and emotional matters (p.7). She assumes a possible evolutionary step from sympathy to empathy: "The state of active openess has possible been one important step from the more 'immediate' process of sympathyto the more cognitive level of empathy and understanding" (p.9).
   
   
6. Empathy requires remembering processes which re-construct experiences on the basis of current situations (p.9). Following Rosenfield DA. assumes: (R1) There is no memory but the process of remembering. R2) Memories do not consist of static items ... but they result out of a reconstruction process. (R3) The body is the point of reference for all remembering events. (4) Body, time, and the concept of 'self' are strongly interrelated (p.10). Citing HOBSON DA. unerlines that a truly interpersonal understanding can only develop because of the corresponding intimacy between the perceptible expressiveness of a person's 'body' and the expressiveness of certain aspects of that person's 'mind' (p.16). DA. postulates here a life-long perspective, autobiography as an ongoing re-construction of the own history and creating the concept of individual personality (p.10).
   
   
7. With the concept of embodied cognition DA. underlines the importance of the body for cognition. (B1) Every natural system has a unique body which implies a unique viewpoint which in turn implies a unique cognition. (B2) The active exploration of the environment through body movements is highly important for learning about the environment and the development of cognition (P.10). Furthermore (B3) do humans use their bodies intensively as social tools (p.10). According to SYNNOTT one should therefore regard the body as a dynamic social category which has to be correlated with the body image which any individual is using if it perceives its own body (p.11).
   
   
8. In contrast to the individual aspects DA. points out that for social interaction and communication a minimum of consensus on the 'common world view' has to be present, otherwise are individuals with complex cognitive systems running the risc to deviate from the 'normal' behavior (p.11). For the coordination and convergence of behavior plays language a prominent part: "Language does not only function to acquire nowledge about 'behavioral characteristics' of others, but also to get to know the internal 'states' of others, i.e. their feelings, attitudes (...) etc." (p.12). And DA. thinks that it was the pressure to communicate under social conditions, to be cooperative and form alliances, which has probably led to the evolution of a 'technical means' (language) to communicate social affairs, individual personal traits and attitudes...(p.12). Humans are biased towards categorizing the worlDA. We share this cognitive feature with the rest of the animal world, but the 'invention' of symbols and language gave an important push in this direction... Categories help us to reduce the complexity of living and non-living objects. Taxonomy ... can in this context be regarded as a means of describing and 'sorting' the huge number of animal species..(p.13).
   
   
9. In socially living species the recognition of group members and the defence of the own group against members of other groups is a central point (p.13). Categories, which are here important are: member - non-member, normal - different/ abnormal (p.13).
   
   
10. introduces then the concept of autism (BLEULER), which has been differentiated in the following by ASPERGER (Asperger-Syndrom := autistic personality distrubances) and by KANNER (Kanner-Syndrom := infantile autism)(p.14). The main characteristics of autism are (Au1) Impaired social relationships. (Au2) Impairement of communication skills and fantasy. (Au3) Significantly reduced repertoire of activities and interests. (Au5) Abnormal responses to sensations and to the stimuli from the worlDA. (Au6) Even if autistic people treat other people in an emotionless way, they obviously possess and express emotions. (Au7) About 75% of the autistic people have impaired intellectual skills. (Au8) Generally they do not seem to care much about their body (self-injurious behavior, eating disorders ...). (p.14f).
    
    
11. DA. concludes that autistic people who are missing any kind of empathic relationships to the animate world are missing bodily-re-experiencing (resonating) with others (p.16). The principle of matching the own dynamics (as internal experiences or external movements) to the dynamics of the environment seems to be an important aspect in human development (p.16). "There is much empirical evidence for a natural, 'unfolding' capability of coordinated, synchronized movements in child development as a means to build up relationships to the social environment. Coordinated action with conspecifics is a major achievement of children's first 3 years of life. So, the tendency of synchronization of the own world and the world of others could be the basic mechanism which is judged as positive" (p.17). But "for interpreting human behavior it is necessary not only to analyze the current immediate situation of a human, but his/her social and biographical context. This context is usually not directly visible in the facial expressions or the behavior of humans, but has to be inferred and constructed on taking into account the historical aspects, the situatedness of another's mind in time and space" (p.17f).
    
    
12. Biographic re-construction includes the attribution of mental states (...) which is central to the theory of mind approach (p.18). According to PREMACK and WOODRUFF an individual has a theory of mind if it imputes mental states to himself and to others. A system of inferences of this kind is properly viewed as a theory because such states are not directly observable, and the system can be used to make predictions about the behavior of others (p.18). And DA. considers biographical re-construction processess inside an agent using the agent's body and experiences as the point of reference (p.19).
    
    
13. DA. then discusses the question how 'experiencing other minds' could possibly be modelled in artifacts (p.19). She suggests that the concept of 'experiential understanding' can only be described by dynamic mechanisms of resonance and synchronization (p.20). DA. takes therefore the assumption that cognitive systems should by dynamic systems which DA. sees as an opposite to computational models (p.20). And she concludes that the metapher of viewing artifacts as dynamic systems, studying interactions between an artifact and its environment, and correlating them with dynamics inside the agent could be a useful approach to experiental grounding of 'social understanding' in robots (p.20). Therefore she does not accept the rational approach with BDI agents (p.20).
    
    
14. DA. presents then the theory of social competence according to PREMACK D.P. and J. The TSC consists of three units: (i) intentional system, (ii) social system, and (iii) theory of mind system (p.21).
    
    
15. Ad (i): The intentional system identifies the class of items which the theory comprises. It is activated by perceptual inputs, namely by self-propelled movements in space. These objects are interpreted as intentional, as being the locus of internal cause and being engaged in goal-directed behavior. Humans distinguish physical objects (which only move by influence of another object) and animate objects (which move both in space and place). Movement in place is interpreted as animate but not intentional (p.21).
    
    
16. Ad (ii): The social system specifies the changes which the intentional objects undergo. It is activated by the intentional system. Value is domain specific. A positive (approach) or negativ (withdrawl) is assigned. Criteria for this are intensity of motion (caressing, helping or hitting, hurting). Infants expect reciprocation which preserve value in interactions. The concept of power is used to distinguish between possession and group. The infant expects group members to share reciprocation, to act alike, and to act positively to one another (p.21).
    
    
17. Ad (iii): The social system sends information to the theory of mind system. Ist output are explanations, states of mind, perception, desire and belief and ist various variations. These mental states are used to explain the actions. (p.21f).
    
    
18. DA. supposes that humans are in their perception and interpretation of autonomous artifacts influenced by the same processes which they use to interpret humans and other animals (p.22). This assumption has been confirmed by the observations which DA. could make in connection with some robot experiments, where the robot was not looking similar to any existing natural living and intentional species (p.26). She identified six factors:
    
    
19. Humans are biologically biased towards identifying moving objects.
    
    
20. Since robots move autonomously there are no external causes visible.
    
    
21. People wanted to find explanations for the goal-directed behavior of the 'intentional agent'.
    
    
22. The human observer could interact with the agent.
    
    
23. There might be a tendency to project knowledge about importance and pleasureable meaning of balancing behavior to the robots.
    
    
24. The people felt touched by the riscy situations which they diagnosed as dangerous for the agent (p.25f).

GENERAL COMMENTS

1. The basic idea seems to me convincing and is very close to the view I adopt in the Wittgensteinagents [WAs] approach. Thus the article is very stimulating and helps to clarify this approach.
   
   
2. The only weak points of the article I can found are located in the methodological area. The usage of the term 'phenomenological' is not clear as well as the interrelationship between 'empirical studies' and the 'phenomenological phenomena'. Moreover is the supposed difference between 'dynamic systems' and 'computational systems' in my understanding not a necessary one. In consequence of the vague usage of the terms 'empirical' and 'phenomenological' remains in the whole article some uncertainty about the usage of all the concepts describing either observable attitudes or 'internal' properties of the object under investigation. This is especially problematic where DA states that language mediates internal states.

FREE ADAPTATION OF IDEAS

1. In the WA-approach I try to establish an internal architecture of agents, which gives at least one example how two agents can synchronize their internal states with respect to a certain language L. The two main concepts are the non-linguistic NLNG structures and the linguistic structures LNG which constitute adaptive systems of their own and which are interrelated. The non-linguistic structure NLNG together with the linguistic structure LNG compose the body of the system.
   
   
2. In the formal theories of these agents one can show that the synchronization of bodies bod_i and bod_j of two systems i and j presupposes a sufficient isomorphism between bod_i and bod_j.
   
   
3. The WA-approach is situated in a whole bunch of different theory paradigms. The most important ones here are (i) the behavioristic theory T_sr related to the obervable stimuli and responses of a system, (ii) the physiological theory T_n regarding the structure of the physiology of the body (including the brain), and (iii) a phenomenological theory T_phen dealing with the phenomena in a subjective perspective. These basic theories can be related by correlating the data of each domain of investigation: (iv) in the physiologically enhanced behavioristic theory T_snr one correlates stimulus-response pairs with physiological facts and (v) in a physiologically structured phenomenological theory T_n_phen one correlates the subset of physiologically characterized phenomena with all the other phenomena. This allows finally (vi) the correlation of the physiologically enhanced behavioristic Theory T_snr with the physiologically structured phenomenological theory T_n_phen resulting in the supertheory T_snr_phen.
   
   
4. The concept of mind is in this context related to the phenomenological theory T_phen. If one assumes a mapping from the phenomenological theory T_phen to the behavioral interpreted physiological theory T_snr then one could introduce the concept of the embodiment of mind as that correlation of the phenomenological structure to the physiological one. Empirical studies can help to clarify properties of the mind then only in the sense that (a) those studies differentiate the behavioral and physiological parts of the supertheory T_snr_phen and then (b) - by correlation - this correlation induces some additional distinctions in the realm of phenomenology which would not be possible by phenomenology alone on account the limited scope of the phenomenological data.
   
   
5. In the WA-approach we assume pre-scientifically that the objects under investigation are living systems. This implies properties like birth, learning, complex adaptation processes, interactions with the environment etc. We assume furthermore that all living systems are finite systems. Socalled 'continuous' processes are here therefore understood as artifacts of the used method of measurement which happens in all cases where the describing categories are less fine grained then the structures under investigation.
   
   
6. The construction of formal theories of these systems within a process of a continuous situated discourse leads to symbolically represented data D and formal structures STR, which can be combined to a theory T. As long as this theory T tries to 'mimick' all essential properties of the living system this theory T will represent dynamic living systems within the limits which are due to every theory construction process.
   
   
7. Because we assume that every living object under investigation is principially a finite system it is principially possible to construct a computational model CM_T which is a one-to-one-mapping of the underlying theory T and therefore represents the same dynamic system as the theory T. A distinction between computational models and dynamic systems is therefore not necessary.
   
   
8. As far as a synchronization of the 'external' properties P_w of the world with the 'internal' properties IS of the system is assumed, it is necessary to assume (i) some mapping (called stimulation) from P_w into IS resulting in some IS_w, (ii) a mapping (called perception) from IS_w x IS into IS, (iii) a mapping (called acting) from IS into IS_a and (iv) (effecting) from IS_a into P_w, and (v) some adapting mapping (called learning) from IS to IS resulting in a process of more and more imitation of the perceived properties through internally caused actions. All these mappings must be able not only to represent 'single' events but 'state changes' correlating with 'processes'.
   
   
9. These general claims have to become more and more finegrained to cope with the diversities and subtleties which can be observed within the realm of perception, conceptualization, inferencing, planning etc. Very important concepts which have to be reconstructed are e.g. remembering, emotions, evaluation, and symbolization.