Towards Social Identity in Socio-Cognitive Agents

Round 1

Reviewer 1 Report

This is a well-written paper that examines an interesting topic that is presented in an easy-to-follow way.

Some minor points:

• Please defines AI the first time.
• I suggest you better stress the gap in the literature and the importance of the proposed framework. Why is your framework important?
• In the proposed framework, it would be interesting to understand how some socio-economic aspects can be categorized in different points proposed. I think, for example, to personal attitudes, cultural aspects characterizing the place where one is born, religion, education received, the social trust, and the economic context. In addition, it could be interesting to report some empirical cases to clarify better the points described.
• What possible fields of application of the proposed framework and possible future research evolutions? Which areas, in particular, can benefit from your framework?

Author Response

Firstly, we would like to thank the reviewer for the effort and feedback. We considered the suggestions and conducted several modifications to the documents towards solving the mentioned issues. In particular, we identify the potential fields of application that can benefit from our model as well as the research avenues that can be pursued based on this work.

Additionally, we conducted a literature review to better position our contribution among other socio-cognitive architectures and we expect it exposes the gap in the literature that our model attempts to address.

As suggested by the reviewers, we introduced a new diagram to illustrate the architecture of the cognitive social frames model to clarify how the different concepts interact with each other. This addition not only helps to understand how the multiple stages of the agent’s mechanism affect the memory of the agent, but also supports the presentation of a use case added in a new subsection.

Finally, after analysing the paragraph on socio-economic aspects, we decided to mention in the newly added use case subsection (requested by other reviewers) a scenario that contemplates the examples brought up by the reviewer. In particular, as recommended by the reviewer, we address how some of the aspects characterizing each individual may affect and be affected by the context (such as the economic or cultural ones).  Still, besides addressing these concepts in the document, we would like to briefly mention in this reply that most of the examples given by the reviewer (e.g. personal attitudes, cultural aspects, religion) are characteristics that can shape the agent’s behaviour in multiple phases of our model. For instance, one’s religion can affect not only what the agent considers as appropriate behaviour: by deploying specific cognitive resources from the CSFs (e.g. a specific posture) but also one’s interpretation of its surroundings: by assigning social meaning to perceptions that are uniquely related to that religion (e.g. a temple is perceived as an holy place for a worshipper, while it might be perceived as just a monument for tourists). Furthemore, if an agent includes perceptions of surrounding social actor’s actions in its social context, this will influence, as described above, the frames that became salient, hence modifying its behaviour. 

Although out of scope of this work, it is important to mention that one’s actions will be influenced by the social context and also affect others’ social context. Thus, when one’s actions directly manipulate resources that are used and shared by a population of agents, it can potentially lead to distinct agents ascribing different values to the same object based on their frame of reference, potentially enabling researchers with a novel way of modelling socio-economic dynamics in large populations based on the agent’s context.

Reviewer 2 Report

[ Major ]

1. The manuscript provides a good summary of the interplay between cognition and social behavior (Page 1). Page 2 argues the need to design socio-cognitive systems, which are valid, grounded by proper references. However, the idea of modeling socio-cognition for agents (and also social robots) is not new. There is ample literature on the subject matter, including models considering many constructs (e.g., social context) the authors mentioned in the paper. For the article to be published, the authors need to conduct an integrated literature review on socio-cognitive agents, clearly highlighting the gap in research, ideally before 2. Design Principles.

2. The Design Principles need more work to be useful to the designers/developers/engineers of any agents. A design principle should be general and universal to some degree, but the paper's principles are too general, lacking specificity. At the moment, the principles read as a general list of human social characteristics. Should you swap "social-cognitive agents" to "humans." nearly all principles are valid. The point is that a more precise definition of the agent should be defined because it entails different implications and will affect the design principles. Virtual humans, machines, robots, chatbots, ECA (Embodied Conversational Agent) all have different characteristics. For example, a social robot has a physical presence that elicits different social-psychological responses from humans. Currently, the authors are using the term AI, autonomous machines, social agents, and social robots interchangeably. I suggest constraining the author's operational definition of agent and providing a design principle that is more ecologically valid.

3. The authors should provide a representative case study (or a scenario) that explains the CSF model's interaction for the readership to benefit from the paper. The authors provided a real-life example on Page 11 (e.g., football); the authors may consider expanding them to include critical components of CSF and the agent's interaction with humans to reveal how social relationship and context take into consideration.

[ Minor ]

1. The conclusion needs more discussion. What are the limitations of the work? What is the next step for the research community as a whole? A discussion tieing back to the literature review should be provided.

2. Emotion or affect should not be ignored in any socio-cognitive system. 

3. The abstract needs to be specific. For example, the first sentence ("designed around some specific units of social behavior that address particular challenges") is too general without meaningful info. Please consider writing specific and representative units of social behavior and challenges instead of being abstract and general.

Author Response

Firstly, we would like to thank the reviewer for the invested time and the detailed suggestions. 

    One of the major revisions suggested by the reviewer focuses on the lack of a clear positioning of our contribution due to the absence of a literature review. As such, we conducted a literature review to better position our work in comparison to others before describing the Design Principles. Besides mentioning other cognitive models and architectures that endow agents with similar social capabilities to our model, we clearly described in the introduction our model objectives and what gap it attempts to address. 

Additionally, the reviewer mentions that the terms AI, autonomous machines, social agents, and social robots have been used interchangeably. We conducted a review of the terminology adopted and, although these terms are used throughout the Design Principles sections to mention the literature on social cognitive agents that spans from theoretical frameworks to the application on virtual characters or robots, the summarized design principles, i.e. the final sentence of each subsection, now only mentions the term agent, as suggested by the reviewer. Furthermore, while defining them, our goal was to provide a high level direction to the design and development of socio cognitive models for autonomous agents, without specifics on each application area, since we believe some concepts are shared across multiple domains (e.g. robots to games). Still, we agree with the reviewer that the domain of application will affect the design principles, but that is a matter for their interpretation and implementation on particular scenarios rather than something to include in the principles themselves. Also, it is important to note that the design principles are guided by human cognition since our goal is to identify directions that help to create agents that can socially interact with humans in a similar way to how humans perceive, reason and interact with/about each other. 

The reviewer also mentioned the importance of emotions and affect in socio-cognitive systems. Models for emotions and affect have also been researched towards creating better socio-cognitive agents. Indeed, considering that our goal is to create better human-agent interactions, emotional responses should also be focused on when deploying such agents. However, our model does not enforce a specific emotional appraisal approach. Instead, our contribution focuses on identifying the adequacy of a behaviour to the interpretation of the surrounding environment. Nonetheless, the inclusion of emotional appraisal mechanisms as cognitive resources can contribute to the identification of affordable emotional responses based on the context.

As suggested by the reviewers, we introduced a new diagram to illustrate the architecture of the cognitive social frames model to clarify how the different concepts interact with each other. This addition not only helps to understand how the multiple stages of the agent’s mechanism affect the memory of the agent, but also supports the presentation of a use case added in a new subsection. As suggested, we added a running example to demonstrate how the model is affected by other social actors. Reflecting the reviewers suggestions, we expect that these two additions will contribute to a clearer and more readable document.

    Regarding some of the minor revisions suggested by the reviewer, we added specific units of social behaviour to the abstract of the document and improved the conclusion of the work. In particular, we identify areas of application that can benefit from our contribution, the limitations of our model as well as research opportunities for future works.

Round 2

Reviewer 2 Report

Dear Authors,

I have read the manuscript carefully. The paper now has a more straightforward goal, striking a balance between breadth (i.e., concepts applicable to multiple domains) and depth (i.e., socio-cognitive models for autonomous agents). The added figures help the readership to understand the framework better.

I think the work is now substantial enough for publication, conditional on the following:

• The paper should still mention the importance of emotion/affect concerning the socio-cognitive model. The argument provided in the point-by-point letter would do but should be included in the discussion.

Author Response

We would like to thank the reviewer for the feedback.
We also believe the document was greatly improved after the comments and suggestions.

We included a paragraph including the relevancy of affect and emotions on socio-cognitive systems. The paragraph is very similar to the response provided in the previous reply.

Again, all authors would like to thank the reviewer for the invested time and the insightful comments.