From: AAAI Technical Report FS-99-01. Compilation copyright © 1999, AAAI (www.aaai.org). All rights reserved. The Lemur’s Tale - Story-Telling in Primates and Other Socially Intelligent Agents Kerstin Dautenhahn Department of Cybernetics, University of Reading Whiteknights, PO Box 225 Reading RG6 6AY, United Kingdom [email protected] Abstract Conditions1and2canbeappliedtomanysystems from machines to languages. Conditions 3 and 4 are This paper addresses the relationship between social particularly suited to social organization. intelligence and narrative intelligence, with a partic- ular emphasis on 1) the phylogenetic origins of pri- Item 1 in the list above addresses important issues mate (narrative) intelligence, and 2) the ontogenetic of group size. According to Dunbar (Dunbar 1993) origin of autobiographical stories. The ‘Narrative In- group-size is a function of relative neocortical volume telligence Hypothesis‘ (NIH) is introduced according in nonhuman primates. In human societies 150 appears to which the evolutionary origin of stories and narra- tivity was correlated with increasing social dynamics to be the upper group size limit which still allows social in primate societies, in particular the need to commu- contacts that can be regularly maintained, allowing ef- nicate about third-party relationships. Requirements fective coordination of tasks and information-flow via for artificial socially intelligent story-tellers are out- direct person-to-person contacts. Such a figure derived lined, and the issue of testing social intelligence is dis- from the analysis of historical as well as contemporary cussed. human societies. Dunbar suggests that 1) there is a cognitive limit to the number of individuals with whom Complexity of Primate Societies any one person can maintain stable relationships (de- Many researchers in primatology have contributed pending on personal knowledge, face-to-face interac- to the Social Intelligence Hypothesis (SIH), some- tions), 2) that this limit is a direct function of relative times also called Machiavellian Intelligence Hypothe- neocortex size, and 3) that this in turn limits group sis (Byrne & Whiten 1988), according to which social size. Dunbar proposes that in order to preserve sta- dynamics were among the crucial selective pressures bility and coherence human language has provided an which have driven the evolution of primate intelligence. efficient means of social bonding, which is provided in Identifying friends and allies, predicting others’ be- non-human primate societies by direct physical contact haviour, knowing how to form alliances, manipulating during social grooming (allowing only much smaller group members, making war, love and peace, are im- groups). Following this argument, language allowed portant ingredients of primate politics (de Waal 1982), an increase in group size while still preserving stability and differences between social complexity in human so- and cohesion within the group. cieties and non-human primate societies appear to be Thus, a primary role of language might have been relatively small. to communicate about social issues, to get to know How can we characterize social complexity? Accord- other group members, to synchronize group behavior. ing to (Philips & Austad 1996) complexity is a func- Language is based on representations and the possi- tion of: 1) the number of functionally distinct elements bility to combine them in arbitrary ways. Represen- (parts, jobs, roles), 2) the number of ways in which tations need not be ‘symbols’, they can be spatial or these elements can interact to perpetuate the system visual in nature, and can be verbal or nonverbal. Al- or to promote its goals (or, if it is an artifact, the goals though apes can be trained to use keyboards or a sub- of its users), 3) the number of different elements (parts, set of American Sign Language in order to communi- jobs, roles) any individual within the system can as- cate with humans, they have not developed such repre- sume at different times or at a given time, and 4) the sentational systems in the wild. Gorillas do not exten- capacity of the system to transform itself to meet new sively sign to each other, neither do they draw figures contingencies (i.e. the capacity of the system to pro- in the sand. Obviously there has not been a selective duce new elements or new relations between elements). advantage for them in developing representational sys- tems. As Oliphant (Oliphant 1999) points out, a rep- Narrativity, the capacity to communicate in terms of resentational system which can learn ‘word-meaning’ stories is therefore regarded an efficient means to com- associations need not be computationally very expen- municate social matters, and the origin of narratives sive. Brainsize can therefore not be responsible for might therefore have been a crucial milestone in the the fact that humans use representational systems and evolution of primate social intelligence (Read & Miller chimpanzees in the wild do not. However, the form of 1995). According to what we call the ‘Narrative In- language as such is meaningless, it requires the cog- telligence Hypothesis‘ (NIH) the evolutionary origin of nitive effort to give meaning and to put messages in stories and narrativity was correlated with increasing context. The ability to construct and give meaning to social dynamics in primate societies, in particular the representations is a ‘computationally’ expensive pro- need to communicate about third-party relationships. cess, e.g. it requires identification and interpretation In the following sections we analyze the primate social of the context of the communicative event, such as the field, and in more depth social understanding and the ‘personality’ of the sender (is he trustworthy?), the re- role of narrative in autobiography. lationship between ‘sender’ and ‘recipient’ of a message Evidence suggests that the evolution of the human (potential mate?), important third-party relationships, story-telling mind was strongly correlated with the evo- positions in the group hierarchy etc. Thus, one and the lution of complex mechanisms of social understanding same ‘message’ can have potentially many different in- and a complex social field. This suggests that if we terpretations and ‘meanings’, depending on the com- intend to develop a socially intelligent agent (Dauten- plexity of the primate social field (discussed below), hahn 1998) which can truly understand and respond the number of different roles an individual can have, to stories in human-agent interaction then we need to and the potential to create new roles and relationships. model at least to a certain extend social relationships and primate social life. Although humans use gestures, ‘body language’ and other non-verbal means to convey (social) meaning, hu- The Primate Social Field man communication is dominated by verbal communi- cation which is serial in nature. Thus, given the serial The primate family tree split up about forty mil- communication channel of human language, what is lion years ago into prosimians which might resemble the best means to communicate social issues, namely early arboreal primates (e.g. lemurs), and anthropoids learning about the who, what, and why? Physical (monkeys, apes, incl. humans). The problems of social grooming, the main group cohesion mechanism in non- life are especially complex for species whose cognitive human primates is ‘holistic’, parallel, spatial, sensual. skills create a complex ‘social field’ which is based on How can a stream of words convey meaning such as several fundamental components: bodily grooming does? Narrative structure seems to 1) Individuals specifically recognize other individu- be particularly suited: usually a narrative is giving a als in their groups. Primate societies are ‘individual- certain introduction of the characters (making contact ized societies’. It is thought that many mammalian between individuals, actors, listener and speaker), de- species are able to recognize individual group mates velops a plot, namely a sequence of actions that con- and remember past interactions with them but it is vey meaning (value, pleasurable, unpleasurable), usu- not established whether they understand third party ally with a high point and a resolution (reinforcement relationships, which would seem to be a skill only pos- or break-up of relationships), and focuses on ‘unusual’ sessed by primates. Two separate mechanisms have events rather than stereotypical events. In this way, been proposed for kin-recognition: early familiarity stories seem to give language a structure which resem- (i.e. previous experience with individuals in question) bles (and goes beyond) physical grooming, namely re- and phenotypic matching (using visual or non-visual placing physical presence and actions by the creation cues). Generally it is assumed that kin recognition in of a mental picture of physical actions, providing the primates depends on previous experience. However, stage, actors, intentions and a storyline. Thus, both chimpanzees have been shown to match related but story-telling and grooming are social bonding mecha- unknown individuals by visual cues, in the same way nisms, and humans use language extensively to discuss as humans can match persons in a family album. In social matters. According to Dunbar (1993) people the wild, chimpanzees form loosely organized fission- spend about 60 percent of conversations on gossiping fusion communities where even closely related indi- about relationships and personal