The theory of social functions: challenges for computational social science and multi-agent learning
Introduction
The social paradigm is rapidly growing within AI because of the situated and interactive perspective (Bobrow, 1991) and of Agent-oriented computing and Multi-Agent Systems (MAS) (Gasser, 1991, Hunhs & Singh, 1998). Such a paradigm will strongly contribute — mainly thanks to Agent-Based Social Simulation — to the birth of the ‘Computational Social Sciences’ Carley, 2000, Müller et al., 1998, Castelfranchi, 1998d. Social sciences will contribute to the design and understanding of artificial societies, cyber-organisations and computer-mediated interaction, while the sciences of the artificial will transform the social sciences, providing experimental platforms, operational and formal conceptualisations, and new models of social phenomena. A significant inter-disciplinary fertilisation is expected like that which, in the 1960s and 1970s, gave birth to Cognitive Science.
The basic claims of this paper are as follows:
- •
The main contribution of AI (and, in particular, of cognitive-Agent modelling and MAS) entering the social simulation domain will be an impressive advance in the theory of the micro–macro link. In particular, the foundational theoretical problem of the social sciences — the possibility of unconscious, unplanned emergent forms of cooperation, organisation and intelligence among intentional, planning agents (the ‘vexata quaestio’ of the ‘invisible hand’, of the ‘spontaneous social order’ but also of ‘social functions’) — will eventually be clarified.
- •
A very serious problem for the theory (and architecture) of cognitive agents is how to reconcile the ‘external’ teleology of behaviour with the ‘internal’ teleology governing it; how to reconcile intentionality, deliberation, and planning with playing social functions and contributing to the social order.
- •
To solve these foundational and architectural problems, complex models of learning are needed, where learning does not operate within a ‘reactive’ architecture made of simple rules, classifiers, associations, and stereotypic behaviours, but operates upon high level anticipatory cognitive representations (beliefs, goals) which govern intentional action. A theory of the relationships between individual intentional behaviour, reinforcement learning, and the feedback of collective emerging effects is needed.
I will present a critical characterisation of the problem of self-organising social phenomena and functions among intentional agents, discussing both unsatisfactory accounts in social theory and in MAS, and the hard theoretical problems to be solved.
I will also try to sketch a possible line of reconciliation between emergence and cognition, by building a notion of behavioural function of intentional action. To do this, I have to build on the unintended social effects of the agents’ behaviours, and on some sort of reinforcement learning dealing with beliefs and intentions.
Section snippets
MAS, agent-based social simulation and their promises
Computer simulation of behavioural and social phenomena is a successful and rapidly growing interdisciplinary area Conte & Gilbert, 1995, Troitzsch, 1997. Suffice to mention the renewed interest of sociologists and economists, testified by several workshops in the international conferences of sociology, economics, and game theory, several ‘social’ papers in the new area of Artificial Life (ALife), papers in the Journal of Mathematical Sociology, books on simulating organisations Masuch, 1995,
An emergent confusion
The triumphant notion of ‘emergence’ has a bad conceptual and epistemological status. Its different meanings exemplify the confusion and the need for a discussion. This is particularly important since, in my view, only computer simulation of emerging phenomena (including social ones) can finally provide some clear notions of ‘emergence’. My aim is also to stress which notion of emergence is really needed, and how to model it on the basis of selection processes (evolution or learning).
‘Emergent’
Social functions and cognition
The aim of this section is to analyse the crucial relationship between social ‘functions’ and cognitive agents’ mental representations. This relationship is crucial for at least two reasons:
(a) on the one hand, no theory of social functions is possible and tenable without clearly solving this problem (see Section 4.2);
(b) on the other hand, without a theory of emerging functions among cognitive agents social behaviour cannot be fully explained.6
Cognitive requirements for a theory of social functions
In order to account for the functional character of intentional actions, from the cognitive point of view, on the one hand we have an architectural problem, on the other we need a sophisticated model of intentional action.
How social functions are implemented through cognitive representations
After the previous characterisation of the critical points in the notion of social ‘function’, and the necessary specifications about intention and cognitive architecture, we can try to sketch the ‘internal’ mechanism(s) for external functions impinging on intentional actions.
I will first describe an abstract simplified model of ‘auto-functions’, be them either ‘kako-functions’ or ‘eufunctions’ relative to the goals or interests of the agents they emerge from. Second, I will exemplify this
Kako- or eu-functions: relative to whom or what?
In what sense is the ‘clean-street’ habit good and the ‘dirty-street’ habit bad? As we saw, ‘good’ (eu) and ‘bad’ (kakos) must be relative to the goals or interests of some system/agent (Miceli & Castelfranchi, 1989). In fact, so far we have referred the bad character of these functions to the involved agents’ goals or interests. So, the habits of dirtying streets is bad relative to B1 and its related goals or to the cleanliness and aesthetic interests of the agent. With respect to those goals
Why Elster and Hayek are wrong
On the basis of this cognitive characterisation of functions, let me now summarise the main points that Elster’s criticism and proposal about the notion of function do not take into account. I will also discuss the limits of Hayek’s view of spontaneous order as necessarily advantageous for the agents.
Concluding remarks
I hope that, after this long and tangled argumentation, it will be clearer why only Computational Social Science and, in particular, Multi-Agent-Based Social Simulation (SS) could probably deal with this kind of problem. Moreover, the task of SS is not only to predict emerging social effects or the experimentation of possible policies. I believe that the contribution of SS to the theoretical development of the cognitive and social sciences could be really remarkable. SS can provide not only an
Acknowledgements
This research forms part of a 20-year project pursued at the IP-CNR aimed at reconciling scientific teleological approaches and a theory of goal-governed agents with the theory of goal-oriented systems and of functional activities (Castelfranchi, 1982). This work on functions would not be possible without years of reflection and collaboration with Rosaria Conte and Maria Miceli on these topics. I would like to thank Rosaria, Maria, and Rafael Bordini for their precious comments. I am also in
References (93)
Intelligence without representation
Artificial Intelligence
(1991)Modelling social action for AI agents
Artificial Intelligence
(1998)Social conceptions of knowledge and action: DAI foundations and open systems semantics
Artificial Intelligence
(1991)Biological function
Animal Behavior
(1978)Learning, action, and consciousness: a hybrid approach towards modeling consciousness
Neural Networks
(1997)- et al.
Some experiments with a hybrid model for learning sequential decision making
Information Sciences
(1998) - Agre, P. E. (1989). The dynamic structure of everyday life. Phd Thesis. Boston: Department of Electrical Engineering...
The architecture of cognition
(1983)- et al.
The unbearable automaticity of being
American Psychologist
(1999)
Norms of cooperation
Ethics
Dimensions of interaction
AI Magazine
Effects pervers et ordre social
An invitation to reflexive sociology
Etiological theories of function: a geographical survey
Evolutionary epistemology
Computational social science: agents, interaction, and dynamics
Scopi esterni (External ends)
Rassegna Italiana di Sociologia
Social power: a missed point in DAI, MAS and HCI
Reasons: belief support and goal dynamics
Mathware and Soft Computing
Individual social action
Challenges for agent-based social simulation. The theory of social functions
Through the minds of the agents
Journal of Artificial Societies and Social Simulation
Simulating with cognitive agents: the importance of cognitive emergence
Emergence and cognition: towards a synthetic paradigm in AI and cognitive science
Per una teoria (pessimistica) della mano invisibile e dell’ordine spontaneo (For a pessimistic theory of the invisible hand and spontaneous social order)
Through the agents’ minds: cognitive mediators of social action
Affective appraisal vs. cognitive evaluation in social emotions and interactions
Emerging functionalities among intelligent systems: co-operation within and without minds
AI & Society
Cultural transmission and evolution. A quantitative approach
Just how (un)realistic are evolutionary algoritms as representations of social processes?
Journal of Artificial Societies and Social Simulation
Influence. The psychology of persuasion
Rational interaction as the basis for communication
The necessity of intelligent agents in social simulation
Introduction: computer simulation for social theory
Functional analysis
Descartes’ error
MANTA: new experimental results on the emergence of (artificial) ant societies
Division of labour and social co-ordination modes: a simple simulation model
Functional analysis in anthropology and sociology: an interpretative essay
Annual Review of Anthropology
Marxism, functionalism and game-theory: the case for methodological individualism
Theory and Society
Cited by (93)
A Biased Inferential Naivety learning model for a network of agents
2022, Cognitive Systems ResearchCitation Excerpt :The process of obtaining information and updating beliefs in the context of society is called social learning, which is an important subject of research in various disciplines. It is applied in the context of distributed signal processing in the wireless sensor networks for distributed detection, distributed estimation, and cognitive agent modeling (Borkar & Varaiya, 1982; Krishnamurthy & Poor, 2013; Tsitsiklis & Athans, 1984; Wang & Djuric, 2015; Castelfranchi, 2001). It is also applied to solve coordination and consensus problems in control theory (Blondel et al., 2009; Bullo et al., 2009).
Multi-agent-based modelling and simulation of high-speed train
2020, Computers and Electrical EngineeringCitation Excerpt :The agent has a flexible organisational framework and evolutionary mechanism. MABMS has been applied in many areas, such as the modelling of organisational psychology [5], engineering team work [6], social functions [7], consumer behaviour [8], friendship in social networks [9], traffic and transportation simula- tion [10], population oscillations [11], dynamics of contagious disease spread [12], and military applications [13]. With respect to the HST, most researchers have adopted the methodology of mechanical systems and structural dynamics simulation using the automatic dynamic analysis of mechanical systems (ADAMS) software [14].
A Theory of Tutelary Relationships
2023, A Theory of Tutelary RelationshipsA Survey of Deep Q-Networks used for Reinforcement Learning: State of the Art
2023, Lecture Notes on Data Engineering and Communications TechnologiesDynamical systems as a level of cognitive analysis of multi-agent learning: Algorithmic foundations of temporal-difference learning dynamics
2022, Neural Computing and Applications