Dynamic identity of social agents
0 likes338 views
The document discusses the dynamic identity model (DIMA) for social agents, emphasizing that social identity influences behavior and decision-making based on context. It outlines how agents must adapt their identities—personal or social—relative to the situation, incorporating factors like salience and emotional relevance. Additionally, the document addresses limitations of the model and proposes future research directions such as exploring self-esteem's role and interpersonal relationships.
![The DIMA Model
A set of agents A = {a1,…,an}
Each agent has a list of personal characteristics Pa = [c1,…, cn] : c ∈ C
A set of social groups SG, each with a list of prototypical characteristics
SGi = [c1, …,cn] : c ∈ C
Core Problem
To determine if in a situation an agent should use its personal identity
or a social identity (i.e. to act as an individual or as a member of a
social group)](https://image.slidesharecdn.com/dynamicidentityofsocialagents-210319134002/85/Dynamic-identity-of-social-agents-5-320.jpg)
![The DIMA Model
Context is defined by
The agents present (who)
may be brought to context by reference
The theme (what, where, when)
topic, place, event → a set of characteristics T = [c1,..,cn]
The active identity is the most salient one](https://image.slidesharecdn.com/dynamicidentityofsocialagents-210319134002/85/Dynamic-identity-of-social-agents-7-320.jpg)
![The DIMA Model
Computing the salience
The context theme filters the relevant personal characteristics
Agents are described only by a subset of Pa(normalized values [0-100])
Clustering algorithm to find social groups (e.g. KNN)
Compute comparative fit
fit = α × 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑆𝐺𝑖𝑛, 𝑆𝐺𝑜𝑢𝑡 +
(1 − α) (β × 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 𝑆𝐺𝑖𝑛 + (1 − β) × 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 𝑆𝐺𝑜𝑢𝑡 )
(α and β) = [0-1] are meta-parameters](https://image.slidesharecdn.com/dynamicidentityofsocialagents-210319134002/85/Dynamic-identity-of-social-agents-9-320.jpg)
Dynamic identity of social agents
- 1. Dynamic Identity of Social Agents Rui Prada, Joana Dimas, Samuel Mascarenhas, Diogo Rato Instituto Superior Técnico, Universidade de Lisboa and INESC-ID Social Identity @ Social Simulation Fest, 19 March 2021
- 2. Motivation • The social context shapes who we are (our identity) • Who we are shapes our behaviour and interpretation of a situation • We act according to our identity • Socially aware agents should take into account social identity dynamics
- 3. Socially aware agents • Understand social context (who, what, where, when) • Adapt to social context • Update the social identity • Use the active social identity to drive the decisions
- 4. The DIMA Model Dynamic Identity Model for Agents (DIMA)1,2 • Social Identity Theory, Social Categorisation and Meta-contrast • Agents redefine their (and others) identity either as unique individuals or as members of a social group • Apply in-group / out-group dynamics do decision making 1Joana Dimas, Phil Lopes, Rui Prada: “One for all, all for one: Agents with social identities” in proceedings of CogSci‘2013 - 35th Annual Meeting of the Cognitive Science Society, pp. 2195-2200, Berlin, Germany. August 2013. CSS. 2Joana Dimas, Rui Prada: “Dynamic Identity Model for Agents” in Multi-Agent-Based Simulation XIV: International Workshop, MABS 2013, Saint Paul, USA, May 2013, Revised Selected Papers, Shah Jamal Alam, H. Van Dyke Parunak (Eds.). Lecture Notes in Computer Science, pp. 37-52. 2014. Springer Berlin Heidelberg.
- 5. The DIMA Model A set of agents A = {a1,…,an} Each agent has a list of personal characteristics Pa = [c1,…, cn] : c ∈ C A set of social groups SG, each with a list of prototypical characteristics SGi = [c1, …,cn] : c ∈ C Core Problem To determine if in a situation an agent should use its personal identity or a social identity (i.e. to act as an individual or as a member of a social group)
- 6. The DIMA Model Set of identities One personal identity per agent One social identity per social group Agents have an accessibility value per social identity Each agent may have more than one social identity Determine the active identity congruent with the context One of the above (personal or social)
- 7. The DIMA Model Context is defined by The agents present (who) may be brought to context by reference The theme (what, where, when) topic, place, event → a set of characteristics T = [c1,..,cn] The active identity is the most salient one
- 8. The DIMA Model Social identity salience 𝒔𝒂𝒍𝒊𝒆𝒏𝒄𝒆 𝑆𝐼, 𝑎, 𝑐𝑜𝑛𝑡𝑒𝑥𝑡 = 𝑓𝑖𝑡𝑆𝐼, 𝑐𝑜𝑛𝑡𝑒𝑥𝑡 × 𝑎𝑐𝑐𝑒𝑠𝑠𝑖𝑏𝑖𝑙𝑖𝑡𝑦𝑆𝐼, 𝑎 Fit: fits the situation (comparative and normative) Accessibility: recall past experiences, emotional relevance/attachment
- 9. The DIMA Model Computing the salience The context theme filters the relevant personal characteristics Agents are described only by a subset of Pa(normalized values [0-100]) Clustering algorithm to find social groups (e.g. KNN) Compute comparative fit fit = α × 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑆𝐺𝑖𝑛, 𝑆𝐺𝑜𝑢𝑡 + (1 − α) (β × 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 𝑆𝐺𝑖𝑛 + (1 − β) × 𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑖𝑜𝑛 𝑆𝐺𝑜𝑢𝑡 ) (α and β) = [0-1] are meta-parameters
- 10. The DIMA Model salience (a) > salience (b) > salience (c) = 0
- 11. The DIMA Model If more than one social group is identified in the context Try to recall the normative social identity 𝑁𝑜𝑟𝑚𝑆𝐺 = 𝑎𝑟𝑔 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑆𝐺𝑘, 𝑆𝐺𝑖𝑛 < 𝑡ℎ ∶ 𝑆𝐺𝑘 𝑖𝑛 𝑎𝑔𝑒𝑛𝑡’𝑠 𝐾𝐵 𝐼𝐹 𝑁𝑜𝑟𝑚𝑆𝐺 ≠ ∅, 𝑡ℎ𝑒 𝒔𝒂𝒍𝒊𝒆𝒏𝒕 𝒊𝒅𝒆𝒏𝒕𝒊𝒕𝒚 = 𝑎𝑟𝑔𝑚𝑎𝑥 (𝑎𝑐𝑐𝑒𝑠𝑠𝑖𝑏𝑖𝑙𝑖𝑡𝑦(𝑆𝐺𝑛)) IF 𝑁𝑜𝑟𝑚𝑆𝐺 ≠ ∅, then the agent is in the presence of a new social group (ad-hoc, non normative), the 𝒔𝒂𝒍𝒊𝒆𝒏𝒕 𝒊𝒅𝒆𝒏𝒕𝒊𝒕𝒚 = 𝑐𝑒𝑛𝑡𝑟𝑜𝑖𝑑(𝑆𝐺𝑖𝑛)
- 12. The DIMA Model Active Identity = personal identity →(salience) salient social identity E.g. if the salience is above a threshold the active identity is the social identity If only one social group is identified in the context, the active identity is the personal identity
- 13. The DIMA Model The agent’s decisions are based on The characteristics of the active identity The salience of the social identity The membership regarding SGin and SGout The prototypical characteristics of the others After taking an action the emotional relevance of the outcome is assessed The accessibility of the social identity is updated 𝑎𝑐𝑐𝑒𝑠𝑠𝑖𝑏𝑖𝑙𝑖𝑡𝑦 𝑆𝐺𝑖𝑛 𝑡 + 1 = 𝑎𝑐𝑐𝑒𝑠𝑠𝑖𝑏𝑖𝑙𝑡𝑦 𝑆𝐺𝑖𝑛 𝑡 + 𝑒𝑚𝑜𝑡𝑅𝑒𝑙𝑒𝑣 𝑜𝑢𝑡𝑐𝑜𝑚𝑒 𝑡 × 𝑠𝑎𝑙𝑖𝑒𝑛𝑐𝑒(𝑆𝐺𝑖𝑛)t
- 14. The DIMA Model
- 15. Application to a Dictator Game Repeated multi-agent dictator game offer = base ± f(salience(SI)) – ingroup / outgroup bias 10 agents (two groups) Varying size of groups (5 vs 5, 4 vs 6 and 2 vs 8) Results Salience of social identity of smaller groups was higher Accumulated wealth of smaller groups was higher Effects of ingroup favouritism are stronger
- 16. Group based emotions1 Displaying emotions as individual or group Determine the cognitive unit for the emotional appraisal Positive effects for group identification, trust and likability 1Filipa Correia, Samuel Mascarenhas, Rui Prada, Francisco S. Melo, Ana Paiva: "Group-based emotions in teams of humans and robots" in proceedings of HRI'18 - International Conference on Human-Robot Interaction, pp. 261-269, Chicago, IL, USA, March 2018. ACM/IEEE.
- 17. The DIMA Model Limitations • The comparative fit is predominant • Works better with ad-hoc social identities (minimal group paradigm) • The context theme is directly defined by a set of characteristics • Need to improve the definition of contextual features • The personal identity and social identity characteristics belong to the same set and are of numeric in nature • Better support for (cognitive) social features
- 18. Cognitive Social Frames1 1Rato, Diogo, Samuel Mascarenhas, and Rui Prada. "Towards Social Identity in Socio-Cognitive Agents." arXiv preprint arXiv:2001.07142 (2020).
- 19. Ideas for the future • Use self-esteem as a driving force of social agents • Explore optimal distinctiveness theory • Social identity and inter-personal relationships
- 20. Thank you Dynamic Identity of Social Agents Rui Prada rui.prada@tecnico.ulisboa.pt Social Identity @ Social Simulation Fest, 19 March 2021