Social Order and Adaptability in Animal and Human Cultures as Analogues for Agent Communities: Toward a Policy-Based Approach
Summary (7 min read)
2. Some Sources of Order in the Animal World
- The authors start by examining some of the ways that animals cooperate and maintain order.
- Speaking of the process of mutual “attunement” (roughly, “getting to know one another”) among individuals, a component process of cooperation, biologist W.J. Smith states: Smith goes on to discuss two main benefits that accrue from such processes of cooperation or “joint activity.”.
- But the main benefit of predictability is the social order it contributes to the group.
- Thus, to know that an individual is performing a particular display is to learn something about the behavior it may select—every display can thus be described as encoding messages about behavioral selections [60, p. 87].
2.1. Interactional Displays
- Interactional displays indicate availability or unavailability to participate in joint activity.
- These displays “primarily provide information about the communicator’s readiness or lack of readiness, to join in acts that involve other individuals” [60, p. 88].
- This category also includes signals of shunning interaction.
- To see why this may be useful, consider the signaling functions of the lights on the back of a car: “[W]e use turn signals and brake lights to tell others of their actions and intentions.
- Loud sounds, loud singing, howling (e.g., one jackal howls, and all the rest in the area howl in response), assuming high, visible physical positions, special kinds of flight patterns or displays, also known as Absence of opportunity.
2.2. Seeking Displays
- The behavioral selection about which a display provides information if it is done only in this way can be termed ‘seeking.’.
- The display is interpreted as providing not just the information that a communicator is ready to do this second selection, but that its behavior includes seeking or preparing to seek an opportunity” [60, p. 118].
- These are associated with so many kinds of behaviors that their particular forms vary widely.
- Agents that indicate to others what they are trying to do can elicit the right form of aid from others, can contribute to possible coordination among tasks, and the like.
2.3. Receptiveness Displays
- Displays indicating receptiveness are the inverse of seeking displays, i.e., they indicate a specific response to the seeking of particular kinds of activities by others: “Some displays indicate the behavioral selections that a communicator will accept, not those it is prepared to perform.
- At least two behavioral selection messages must be provided by such a display, one indicating that the communicator will behave receptively and another indicating the class of acts to which it is receptive.
- Effectively, the communicator adopts the role of soliciting acts from another individual; it does not offer them” [60, p. 122].
- These displays sometimes carry over into adult relationships, as when a female mate solicits various forms of “help with the nest” from her male partner [60, p. 125].
- As with seeking displays, receptivity displays are so diverse that they defy general description.
2.4. Attack and Escape Displays
- Displays indicating attack and escape: “are said to encode either, or both, of attack and escape messages when all their occurrence is correlated with a range of attack- or escape-related behavior.
- Behavioral indices of attack differ among and within species, but include acts that, if completed, will harm another individual.
- Escape behavior can be any appropriate form of avoidance, ranging from headlong fleeing to turning aside, or even freezing and other ways of hiding” [60, p. 93].
- Attack and escape displays may differ, but they are sometimes more or less the same display, differing only in degree or subtle nuance.
- They have value both between and within groups, for instance, to muster help against an intruder or to avoid inadvertent flare-ups (e.g., one group member coming upon another by surprise).
2.5. Copulation Behavior Displays
- There are displays indicating copulation behavior: “Some displays are performed only before or during the social interactions in which eggs are fertilized.
- These interactions involve either copulation or some behavioral analogue such as the amplexus behavior of frogs” [60, p. 97].
- This class of social display would seem to have little to do with agents—at least at their current stage of development.
- Analogues to these displays may be pertinent when certain forms of intricate inter-coordination are occurring among agents, involving the need for complex cooperation and coordination to carry out the task successfully, e.g., exchanging ontologies.
- In a simple fashion, a Palm PDA demonstrates this kind of display when it beeps and lights up after successful docking in its cradle.
2.6. Association Maintenance Displays
- There are displays associated with maintaining, staying-in association: “Some displays correlate with the behavior involved in remaining with another individual.
- These displays appear to provide a kind of reassurance to other group members that, despite some possible indications to the contrary, the individual has not broken ranks with the group.
- Such assurances are particularly useful when salient events may raise doubt about the continued association.
- “the likelihood that a group will remain together after one or more have fought with each other or with outsiders can also be increased by displays encoding an association message” [60, p. 104].
- Various kinds of vocalizations—clearly, signals that can operate over a distance are important in this function.
2.7. Indecisiveness Displays
- Indicators of indecision are various, ranging from simply adopting a static, frozen stance, as if waiting for the situation to provide greater cues, to variations on displays that usually indicate action but are modified to increase the range of choice.
- Displays for indecisiveness can include behaviors irrelevant and inappropriate to the situation, e.g., suddenly, unexpectantly initiating grooming or eating [60, p. 107].
2.8. Locomotion Displays
- Displays indicating locomotion simply signal that the animal is moving or is about to move: “[These] displays provide information about a communicator’s use of flight (or other locomotive) behavior, but not about functional categories of flight such as approach, withdrawal, attack, or foraging.
- The displays correlate with all these acts and more…some [animals] extend the performance of the displays to correlate with hopping or running when they forage on the ground.
- These displays appear to consist primarily of various forms of vocalizations.
- Signals indicating that an animal is about to move can be more diverse, for example, dances in honeybees, head-tossing in geese.
- Signals that indicate that an agent is moving or is about to move would seem particularly germane in teams containing mobile agents.
2.9. Staying-Put Displays
- Displays indicating remaining with a site are the opposite of the locomotion displays: “Displays performed only when a communicator is remaining at a fixed site encode the information he will remain at a single point, in the vicinity of such a locus, or in an area that allows considerable movement within fixed boundaries.
- The behavioral selection referred to is simply “staying-put,” defined with respect to a site” [60, p. 115].
- Song vocalizations, in particular, are associated with remaining in a territory.
- Birds that do not sing can have special vocalizations for remaining in place, e.g., the “ecstatic” vocalization of the Adelie penguin [60, p. 115].
- Also included are wing-beating, and various specialized postures and movements.
2.10. Attentiveness Displays
- Displays indicating attentiveness to a stimulus simply convey that the communicator is attending to something and monitoring it.
- Three distinct barks of a prairie-dog, indicating three different phases of monitoring.
- For agents, these signals could portend that something important might be happening.
- Appropriate response, of course, would require additional information.
- In the animal world, for instance, this additional information sometimes indicates the location of the stimulus.
3. Some Sources of Order in the Human World
- It is not surprising that joint activity—and the “getting to know each other” both necessary for it and engendered by it—are also important to humans.
- Because of their wider behavioral repertoire, the greater complexity of their communication processes, and their reduced dependence on biological determinism, human cooperation and regulatory processes take on an even greater variety of forms.
- Culture, the accumulated totality of such patterns, is not just an ornament of human existence but—the principal basis of its specificity—an essential condition for it.
- For a comprehensive and interesting treatment of these kinds of issues regarding joint activity in humans, see[15].
- Order and predictability may have a basis in the simple cooperative act between two people, in which the parties “contract” to engage together in a set of interlinked, mutually beneficial activities.
4. The Problem of Adaptability
- While the discussion so far has dealt mainly with the maintenance of order, change is also necessary in perpetuating healthy societies, especially if those societies are expected to adapt to new circumstances and endure over long periods of time.
- In the second case, the party responding to the request for help might, on the one hand, go to the unmanned end of the table and try to help lift (and he would not throw a rope—due to the basic circumstantial difference).
- [In Mexico, when two cars approach a narrow bridge from different directions, flashing your headlights means, ‘I got here first, so keep out of my way.’.
- Thus the elements of consistency, but also potential novelty, may both be necessary to signaling activity in the real world, because the world is never static: “In all social events, the behavior of participants must engender considerable predictability.
- With regard to change and adaptation in culture and its regulatory role, modern biologists have increasingly emphasized that the natural selection process includes not only basic biology but also the equally complex elements of culture, cultural change and cultural selection.
5. Building Cultures for Agent Communities: Sources of Order
- The authors agent research and development efforts over the past decade have maintained a consistent trend.
- It is in this sense that what the authors have been doing might be thought of as creating “cultures” for agent communities, especially communities that might endure for long periods of time.
- Beyond the basics of individual agent protection, these communities will depend on legal services, based on explicit policies, to ensure that rights and obligations are monitored and enforced.
- The authors will introduce KAoS (5.4) and some basic categories of technical and social policies (5.5).
5.1. Norms and Policy
- In the early 20th century, a legal theorist named Wesley Newcomb Hohfeld developed a theory of fundamental legal concepts [32] from which most of current work on theories of normative positions have taken at least some degree of inspiration (see e.g., [40; 57]).
- In the multi-agent system research community, Conte and Castelfranchi [19] found that norms were variously described as constraints on 8.
- In an insightful essay, Roger Clarke explores some of the implications of Asimov’s stories about the laws of robotics for information technologists [16].
- Interest in policy-based approaches to multi-agent and distributed systems has also grown considerably in recent years (http://www.policy-workshop.org) [22; 37; 67].
- While sharing much in common with norm-based approaches, policy-based perspectives differ in subtle ways.
5.2. Plans and Policy
- Policy management should not be confused with planning or workflow management, which are related but separate functions.
- Planning mechanisms are generally deliberative (i.e., they reason deeply and actively about activities in support of complex goals) whereas policy mechanisms tend to be reactive (i.e., concerned with simple actions triggered by some environmental event) [27, pp. 161-162].
- The independence of policy, reasoning, and enforcement mechanisms from planning capabilities helps assure that, wherever possible, key constraints imposed by the humans are respected even in the face of buggy or malicious agents on the one hand, and poorly designed or oversimplified plans on the other.
- Plans tend to be strategic and comprehensive, while policies, in their sense, are by nature tactical and piecemeal.
- In short, the authors might say that while policies constitute the “rules of the road”—providing the stop signs, speed limits, and lane markers that serve to coordinate traffic and minimize mishaps—they are not sufficient to address the problem of “route planning.
5.3. Autonomy and Policy12
- The outermost rectangle, labeled potential actions, represents the set of all actions defined in some ontology under current consideration.
- Of these possible actions, any given actor15 (e.g., Agent A) will likely only be deemed to be capable of performing some subset.
- Environmental autonomy can be expressed in terms of the possible actions available to the agent—the more the behavior is wholly deterministic in the presence of a fixed set of environmental inputs, the smaller the range of possible actions available to the agent.
- A computational system’s “ontology” defines what exists for the program—in other words, what can be represented by it.
- For this reason, adjustable autonomy may involve not merely a shift in roles among a human-agent pair, but rather the distribution of dynamic demands across many coordinated actors.
5.4. Overview of KAoS
- KAoS is a collection of componentized agent services compatible with several popular agent frameworks, including Nomads [63], the DARPA CoABS Grid [38], the DARPA ALP/Ultra*Log Cougaar framework (http://www.cougaar.net), CORBA (http://www.omg.org), and Voyager (http://www.recursionsw.com/osi.asp).
- KAoS domain services provide the capability for groups of agents, people, resources, and other entities to be structured into organizations of agent domains and subdomains to facilitate agent-agent collaboration and external policy administration.
- The KAoS Policy Ontologies (KPO), represented in OWL [69], distinguishes between authorizations (i.e., constraints that 19 As Hancock and Scallen [31] rightfully observe, the problem of adaptive function allocation is not merely one of efficiency or technical elegance.
- Permit or forbid some action) and obligations (i.e., constraints that require some action to be performed, or else serve to waive such a requirement) [22].
5.5. Technical and Social Policy Categories
- To increase the likelihood of human acceptance of agent technology, successful systems must attend to both the technical and social aspects of policy [51].
- From a social perspective, the authors want agents to be designed to fit well with how people actually work together and otherwise interact.
- This category of policies is concerned with assuring that identification of proper users is associated with various agent commands and actions.
- The authors now discuss a few simple examples of policy relating to the theme of display and signaling behavior.
- In places that have large control panels,… the first act of the human operators is to shut off the alarms so they can concentrate upon the problem” [50, p. 128].
5.6. Nonverbal Expression Policy: Examples
- Where possible, agents usually take advantage of explicit verbal channels for communication in order to reduce the need for relying on current primitive robotic vision and auditory sensing capabilities [47, p. 295].
- Books on human etiquette [70] contain many descriptions of appropriate behavior in a wide variety of social settings.
- Finally, in addition to this previous work, the authors think that display and signaling behavior among people [46] and groups of animals will be one of the most fruitful sources of policy for effective nonverbal expression in agents.
- This policy prevents the PSA from moving until it has first signaled for some number of seconds its intention to move.
6. Building Cultures for Agent Communities: Potential Sources of Adaptation
- The authors have seen an example of the need for this kind of adaptation in the last section, in which the comfortable distance a PSA should keep from its partner invokes cultural considerations.
- The second type of adaptation involves changes in policy, either in response to experience, for example, in realizing that enforcing a policy or set of policies has consistently resulted in untoward outcomes, or by recognizing that the nature of the operational world had changed in consequential ways.
- This second kind of adaptation has been even less explored.
- From the perspective of this paper, such adaptation might involve a sort of “cultural learning” that might prove challenging to current machine learning approaches.
7. Conclusion
- The authors have attempted to encourage an expansion of thinking about the sources, nature, and diversity of regulatory systems that can be utilized to achieve acceptable levels of order when groups of agents or mixed agent-human groups are engaged in consequential work.
- Roughly interpreted, the constitution of authority refers to how things of various sorts come to have regulatory power over human conduct.
- That is, there are limits to the speed with which a particular sort of car, on a particular sort of road, can navigate the turn without crashing, and people who do not want to get hurt will honor these constraints as they are able.
- At much greater degrees of abstraction from the scene, there is the Motor Vehicle Code and other formal statutes that, for instance, prescribe the amount of certain substances that the driver may have in his or her body.
- The complexity of this interplay makes us realize even more that the authors are only at the beginning in addressing the dual problems of order and change in agent communities (let alone the optimal delicate balance between them), and it is hard not to feel a bit overwhelmed.
Did you find this useful? Give us your feedback
Citations
30 citations
26 citations
Cites methods from "Social Order and Adaptability in An..."
...As part of this research, we are developing policies to govern various nonverbal forms of expression in software agents and robots [ 11 ]....
[...]
25 citations
20 citations
Cites background from "Social Order and Adaptability in An..."
...Joint activity, interdependent activity, is important even to animals, which in some ways are better analogues than humans for the limited intellectual capacities of software agents [16]....
[...]
...Wohlmuth's full analysis is discussed in [16]....
[...]
20 citations
References
20,105 citations
1,376 citations
"Social Order and Adaptability in An..." refers background in this paper
...Breazeal has taken inspiration from research in child psychology [68] to develop robot displays that reflect four basic classes of preverbal social responses: affective (changing facial expressions), exploratory (visual search, maintenance of mutual regard with human), protective (turning head away), and regulatory (expressive feedback to gain caregiver attention, cyclic waxing and waning of internal states, habituation, and signals of internal motivation) [13]....
[...]
1,233 citations
"Social Order and Adaptability in An..." refers background in this paper
...Measures of expected utility can be used to evaluate the tradeoffs involved in potentially interrupting the ongoing activities of agents and humans in such situations, in order to communicate, coordinate, and reallocate responsibilities [18; 33; 34]....
[...]
1. Introduction