=Paper=
{{Paper
|id=Vol-3660/paper8
|storemode=property
|title=Conceptual Models as a Basis for a Framework for Exploring Mental Models of Co-Creative AI (short paper)
|pdfUrl=https://ceur-ws.org/Vol-3660/paper8.pdf
|volume=Vol-3660
|authors=Jeba Rezwana,Mary Lou Maher
|dblpUrl=https://dblp.org/rec/conf/iui/RezwanaM24
}}
==Conceptual Models as a Basis for a Framework for Exploring Mental Models of Co-Creative AI (short paper)==
https://ceur-ws.org/Vol-3660/paper8.pdf
Conceptual Models as a Basis for a Framework for
Exploring Mental Models of Co-Creative AI
Jeba Rezwana1,∗ , Mary Lou Maher2
1
Towson University, MD, USA
2
University of North Carolina at Charlotte, NC, USA
Abstract
Recent generative AI tools have moved human-AI co-creativity into the forefront of mainstream culture.
Yet designing co-creative AI systems that effectively respond to human partners’ values, preferences,
and goals poses a significant challenge. Gaining insights into users’ mental models is essential for the
development of human-centered co-creative AI. This paper introduces a conceptual model of co-creative
AI as a framework for exploring and analyzing users’ mental models of co-creative AI. Our framework
guides the design of tools for investigating mental models to align co-creative AI with users’ needs and
values.
Keywords
Co-Creative AI, Mental Models, Conceptual Models, Framework, Human-AI Co-Creation
1. Introduction
The availability of popular generative AI tools for creative domains like ChatGPT [1], DALL.E [2],
and Github Copilot [3] has created widespread public interest, placing human-AI co-creativity
into the mainstream culture. However, designing effective co-creative AI systems that effectively
respond to human partners’ values, preferences, and goals poses a significant challenge. In the
dynamic creative process, diverse strategies and reasoning are essential, requiring adaptable
AI agents to accommodate evolving human ideas. To achieve cognitive convergence in a
co-creation, Fuller and Magerko propose understanding users’ mental models [4].
A mental model is an individual’s understanding of how something functions, allowing
them to explain, predict, and act within systems [5]. Mental models are subjective constructs
shaped by an individual’s beliefs, values and experiences [6]. In the realm of human-centered
co-creative AI, understanding users’ mental models is crucial for aligning AI with their values
and needs, fostering motivation for AI use. The effectiveness of co-creative AI is influenced by
users and their social and cultural contexts. Staggers and Norcio [7] suggested that researchers
must be aware of users’ mental models to make human-centered designs.
The literature on users’ mental models of co-creative AI is notably scarce, leaving several
important questions unanswered. For instance, what are the constructs of conceptual models of
co-creative AI? Conceptual models are representations of a target system developed purposefully
by experts, unlike mental models, which are developed quickly and often unconsciously [6, 8].
Joint Proceedings of the ACM IUI Workshops 2024, March 18-21, 2024, Greenville, South Carolina, USA
∗
Corresponding author.
Envelope-Open jrezwana@towson.edu (J. Rezwana); m.maher@uncc.edu (M. L. Maher)
Orcid 0000-0003-1824-249X (J. Rezwana); 0000-0002-4150-0322 (M. L. Maher)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
Conceptual models of co-creative AI enable exploring the diverse mental models held by users.
Gaining insights into users’ mental models enables the development of human-centered co-
creative AI systems that better align with users’ needs and values.
In this paper, we propose a conceptual model of co-creative AI leveraging the existing
literature, which can be used as a framework to explore and analyze mental models of co-
creative AI. The growing field of human-AI co-creativity would benefit from studying users’
mental models of co-creative AI as they would help inform the design of human-centered
co-creative AI.
2. Related Works
2.1. Mental Model Theory
Mental models are cognitive constructs developed through real-world experiences that allow
humans to understand how a system functions [9]. These models allow people to understand,
explain and predict phenomena and act accordingly [5, 10]. Therefore, users with well-developed
mental models should be able to produce more accurate results when interacting with a system.
Staggers and Norcio [7] suggested that designers and researchers must be aware of users’ mental
models. Norman [6] highlights that these models are subjective and prioritize usefulness over
accuracy [6]. The contents of mental models can be concepts, relationships between concepts
or events and associated procedures [11]. Humans construct mental models by drawing on
analogies or metaphors of past represented objects or interactions [7, 12]. Young proposes eight
types of mental models, including analogy as a form of mental model [13].
A conceptual model is a purposefully constructed accurate representation of the target system,
coherent with scientifically accepted knowledge and typically developed systematically by
experts [14, 6]. Human mental models are black boxes and will never be completely transparent
[9]. Therefore, a solid conceptual model of a system is necessary as a tool to investigate an
individual’s mental model of that system [6]. Conceptual models are simplified representations
of the target system [15].
2.2. Mental Models of AI
There have been a few studies on mental models of AI based on deep neural networks. Gero et al.
developed a conceptual model of an AI system in a game setting and explored how users develop
their mental models, uncovering peoples’ preconceived notions affecting their mental models
[8]. Tullio [16] investigated how users build mental models of an intelligent agent predicting
an office worker’s availability. Kulesza et al. [11, 17] examined mental models of an intelligent
music recommender system, using surveys to identify participants’ mental models and found
that a 15-minute tutorial significantly improved the soundness of their mental models.
Various works in HCI studied users’ mental models of AI systems, though very few studied
mental models of co-creative AI. Llano et al. [18] asserted that equipping co-creative AI with
users’ mental models not only would enable better coordination but would also provide a
valuable resource for co-creative AI to explain, justify, and defend their contributions. The
idea of mental models as a key aspect of the design of real-time co-creative systems has been
highlighted previously [19]. Davis et al. asserted that users’ mental models help co-creative AI
effectively structure, organize, interpret, and act on sensory data in real-time, which is critical
for meaningful co-creation [20].
3. Characterization of a Conceptual Model of Co-Creative AI: A
Framework
In this section, we propose a conceptual model as a framework for exploring users’ mental
models of co-creative AI by drawing on an existing foundation of research.
First, it is important to clarify the definition of conceptual models as it has been used in-
terchangeably with the term mental models in the literature despite being distinct concepts.
According to Norman [6], mental models encompass four distinct aspects: the target system (t),
which is the actual system a user uses; the conceptual model (C(t)) of the target system, which
provides a representation of the system developed by experts; the mental model (M(t)) of the
target system, which users create in their head through the interaction with the target system;
and lastly, the scientist’s conceptualization of the mental model (C(M(t))). Given the inherent
complexity of co-creative AI systems, aligning the conceptual model with the mental model can
pose challenges. Gero et al. argued that a precise description of the neural network architecture
and training procedure does not represent an appropriate conceptual model of an AI [8]. Rather,
conceptual models are simplified representations of the target system which can be as simple as
an analogy [15]. For instance, an analogy between Rutherford’s atom and the solar system can
be considered a conceptual model. Thus, our goal is to develop a simplified conceptual model
that captures the essence of co-creative AI while making it useful for investigating mental
models.
Human-AI co-creativity involves humans and AI collaborating on a creative process as they
generate artifacts or ideas. By definition, co-creative AI systems involve a computational
creativity (generative AI model) component, an interaction/collaboration component and a
utility component. Our conceptual model draws inspiration from Kantosalo et al.’s research,
which introduces three categories of metrics for evaluating co-creative systems: novelty (creative
divergence), interaction and value [21]. These categories of metrics, as asserted by Kantosalo et
al., are components to discern core behaviors of co-creative systems which is the basic purpose
of conceptual models [21]. For our framework, we adapt the categories from their research and
propose three main constructs for conceptual models of co-creative AI: the Creativity Model,
the Interaction Model and the Utility Model (Figure 1).
We conducted a literature review to identify the key components of each construct, as charac-
terizing conceptual models requires alignment with scientifically accepted knowledge [6, 8]. We
used Google Scholar as our search database to conduct the literature review. To identify the key
components of each construct, we used keywords based on the three constructs of our concep-
tual model. The keywords we used are: “mental models in co-creativity,” “conceptual models in
co-creativity,” “computational creativity model,” “interaction in co-creativity,” “usefulness of co-
creative AI,” “utility of AI,” and “co-creativity model.” We considered documents published from
1995 until 2023. We did not include papers that are not in English, papers that by title or abstract
are outside the scope of the research, and papers that do not involve co-creativity. We included
Figure 1: Constructs of the proposed conceptual model of co-creative AI.
papers describing the components of computational creativity, interaction in co-creativity and
utility of co-creation.
Figure 1 shows our conceptual model, including the three main constructs and the key
components of each construct. We refer to the specific publications that contributed to defining
the key components of each construct of the framework in the sections below. In these sections,
the first paragraph defines the construct and the core components of the construct, followed by
specific questions that are presented to explore each of those components. The last paragraph
references the relevant publications that provided the basis for defining the components of each
construct.
3.1. Creativity model
The creativity model encompasses the computational creativity aspect of co-creative AI, focusing
on creative content generation by AI and the roles of AI in the creative process. This model
represents how a co-creative AI generates content, the capabilities of the AI and the extent of
surprise, value, and novelty inherent in the generated content. To effectively understand or
present the creativity model of co-creative AI, we suggest exploring the following components:
AI roles, AI contribution, surprise, and novelty. The following questions could be utilized to
explore these components.
• What can the co-creative AI actually do? (AI role)
• How does the co-creative AI contribute to the creative process? (AI Contribution)
• How surprising is the contribution of the co-creative AI? (Surprise)
• How novel is the contribution of the co-creative AI? (Novelty)
These questions can be augmented in alignment with specific research or design objectives,
as the questions presented serve as a customizable template adaptable to research needs.
The following bodies of research provide the basis for the questions we propose to explore
in understanding the creativity model. Colton et al. emphasized that computational creativity
theory should highlight AI roles and AI contributions of co-creative AI as key components
for explaining the creative process [22]. Kantosalo et al. further assert that the computational
creativity model represents how a co-creative AI generates creative content and how it con-
tributes to the creative process [23]. According to Boden, a creativity model comprises surprise,
novelty, and value components [24]. Additionally, Grace et al. characterize creative outcomes
with computational models of surprise, novelty, and value [25, 26].
3.2. Interaction model
The interaction model represents how the AI interacts and collaborates with humans. It includes
the metaphorical representation of the type of collaboration, the communication quality between
the collaborators and the characterization of the human-AI interaction. When investigating
users’ understanding of the interaction model of a co-creative AI, we suggest the following
components to be explored: AI metaphor, quality of communication between the collaborator and
interaction type. We also offer the following questions to be asked to explore these components.
• What is the appropriate metaphor for the co-creative AI? (AI metaphor in co-creation)
• What is the quality of the communication between humans and co-creative AI? (Quality
of Communication)
• Is the interaction of the co-creative AI collaborative or tool-like? (Interaction type)
If needed, the above questions can be adapted and extended to the particular context and
research objectives. The following research helped us formulate these questions.
The advantages of employing AI metaphors in interaction design have been extensively
discussed within design research [27, 28, 29, 30, 31]. Research shows that metaphors impact
the perception of collaboration [32]. About the characterization of human-AI interaction
type, Kantosalo and Toivonen assert that contrary to how co-creative AI agents are often
viewed in the literature, research in computational creativity aims to develop AI agents that
are equal collaborators in the creative process [23]. Additionally, understanding the quality of
communication between humans and AI is important for capturing the interaction dynamics
[33].
3.3. Utility Model
The utility model encompasses the system’s usefulness, ease of use and overall satisfaction when
interacting with a co-creative AI. When conceptualizing the utility model through the lens of
users, the following questions need to be explored:
• How useful is the co-creative AI? (Usefulness)
• How satisfactory is the co-creative AI? (Satisfaction)
• How easy is it to use the co-creative AI? (Ease of Use)
Similar to the other two constructs, the above questions can be expanded upon considering
the context, domain and research objective. Below we discuss the research that inspired us in
formulating these questions.
To identify the main components of the utility model, we adapted the technology acceptance
model (TAM), a model to understand user acceptance of technology, which is an indicator
of utility [34]. The two main constructs in TAM are perceived usefulness and ease of use.
Satisfaction is a major usability variable [35] and is frequently used in the literature to measure
a system’s usability or user experience [36, 37].
4. Conclusions and Future Work
In this paper, we develop and describe a conceptual model of co-creative AI as a framework
for exploring mental models of co-creative AI. Developing conceptual models of AI proves
challenging due to the absence of a direct one-to-one correspondence between the AI model
and its behavior, as emphasized by Gero et al. [8]. This complexity intensifies when tackling
conceptual models of co-creative AI, where the intricate dynamics of open-ended collaboration
with humans add an additional layer of challenge. We developed a conceptual model of co-
creative AI, including three main components: creativity model, interaction model and utility
model.
The literature on users’ mental models of co-creative AI is notably scarce, leaving several
important questions unanswered. For instance, what types of mental models of co-creative
AI do users possess? Our conceptual model of co-creative AI facilitates the investigation and
representation of key components in users’ mental models. Researchers can employ surveys,
interviews, and other methods based on the framework to explore and analyze users’ models
across diverse contexts and user groups. Understanding users’ mental models facilitates the
development of value-sensitive and personalized AI. Furthermore, while developing explainable
co-creative AI, the constructs of the conceptual model can be used for conveying contextual
information regarding each construct to users, eventually suggesting an accurate model of the
system, as highlighted in recent research [38, 39]. This framework can also be used to develop
tools that help users understand complex co-creative AI in different domains. The conceptual
model also raises questions regarding decision-making within various modules of co-creative
AI. For instance, how should we approach AI roles, contributions, metaphors, and interaction
design? Further research is necessary to explore these questions.
References
[1] ChatGPT: Optimizing Language Models for Dialogue — openai.com, Online. URL: https:
//openai.com/blog/chatgpt/.
[2] DALL·E 2 — openai.com, Online. URL: https://openai.com/dall-e-2/.
[3] GitHub Copilot · Your AI pair programmer — github.com, Online. URL: https://github.com/
features/copilot.
[4] D. Fuller, B. Magerko, Shared mental models in improvisational performance, in: Proceed-
ings of the intelligent narrative technologies III workshop, 2010, pp. 1–6.
[5] P. N. Johnson-Laird, Mental models: Towards a cognitive science of language, inference,
and consciousness, 6, Harvard University Press, 1983.
[6] D. A. Norman, Some observations on mental models, in: Mental models, Psychology Press,
2014, pp. 15–22.
[7] N. Staggers, A. F. Norcio, Mental models: concepts for human-computer interaction
research, International Journal of Man-machine studies 38 (1993) 587–605.
[8] K. I. Gero, Z. Ashktorab, C. Dugan, Q. Pan, J. Johnson, W. Geyer, M. Ruiz, S. Miller, D. R.
Millen, M. Campbell, et al., Mental models of ai agents in a cooperative game setting, in:
Proceedings of the 2020 chi conference on human factors in computing systems, 2020, pp.
1–12.
[9] W. B. Rouse, N. M. Morris, On looking into the black box: Prospects and limits in the
search for mental models., Psychological bulletin 100 (1986) 349.
[10] J. Rasmussen, On the structure of knowledge-a morphology of metal models in a man-
machine system context, Technical Report, RISOE NATIONAL LAB ROSKILDE (DEN-
MARK), 1979.
[11] T. Kulesza, S. Stumpf, M. Burnett, I. Kwan, Tell me more? the effects of mental model
soundness on personalizing an intelligent agent, in: Proceedings of the sigchi conference
on human factors in computing systems, 2012, pp. 1–10.
[12] A. Collins, D. Gentner, How people construct mental models, Cultural models in language
and thought 243 (1987) 243–265.
[13] R. M. Young, Surrogates and mappings: Two kinds of conceptual models for interactive
devices, in: Mental models, Psychology Press, 2014, pp. 43–60.
[14] I. M. Greca, M. A. Moreira, The kinds of mental representations–models, propositions
and images–used by college physics students regarding the concept of field, International
Journal of Science Education 19 (1997) 711–724.
[15] I. M. Greca, M. A. Moreira, Mental models, conceptual models, and modelling, International
journal of science education 22 (2000) 1–11.
[16] J. Tullio, A. K. Dey, J. Chalecki, J. Fogarty, How it works: a field study of non-technical
users interacting with an intelligent system, in: Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems, 2007, pp. 31–40.
[17] T. Kulesza, S. Stumpf, M. Burnett, S. Yang, I. Kwan, W.-K. Wong, Too much, too little, or
just right? ways explanations impact end users’ mental models, in: 2013 IEEE Symposium
on visual languages and human centric computing, IEEE, 2013, pp. 3–10.
[18] M. T. Llano, M. d’Inverno, M. Yee-King, J. McCormack, A. Ilsar, A. Pease, S. Colton,
Explainable computational creativity, arXiv preprint arXiv:2205.05682 (2022).
[19] J. McCormack, P. Hutchings, T. Gifford, M. Yee-King, M. T. Llano, M. d’Inverno, Design
considerations for real-time collaboration with creative artificial intelligence, Organised
Sound 25 (2020) 41–52.
[20] N. Davis, C.-P. Hsiao, Y. Popova, B. Magerko, An enactive model of creativity for compu-
tational collaboration and co-creation, in: Creativity in the Digital Age, Springer, 2015, pp.
109–133.
[21] A. Kantosalo, P. T. Ravikumar, K. Grace, T. Takala, Modalities, styles and strategies: An
interaction framework for human-computer co-creativity., in: ICCC, 2020, pp. 57–64.
[22] S. Colton, J. W. Charnley, A. Pease, Computational creativity theory: The face and idea
descriptive models., in: ICCC, Mexico City, 2011, pp. 90–95.
[23] A. Kantosalo, H. Toivonen, Modes for creative human-computer collaboration: Alternating
and task-divided co-creativity, in: Proceedings of the seventh international conference on
computational creativity, 2016, pp. 77–84.
[24] M. A. Boden, The creative mind: Myths and mechanisms, Psychology Press, 2004.
[25] K. Grace, M. L. Maher, D. Fisher, K. Brady, Data-intensive evaluation of design creativ-
ity using novelty, value, and surprise, International Journal of Design Creativity and
Innovation 3 (2015) 125–147.
[26] K. Grace, M. L. Maher, N. Davis, O. Eltayeby, Surprise walks: Encouraging users towards
novel concepts with sequential suggestions, in: Proceedings of the 9th International
Conference on Computational Creativity (ICCC 2018). Association of Computational
Creativity, 2018.
[27] D. Saffer, The role of metaphor in interaction design, Information Architecture Summit 6
(2005).
[28] A. F. Blackwell, The reification of metaphor as a design tool, ACM Transactions on
Computer-Human Interaction (TOCHI) 13 (2006) 490–530.
[29] A. N. Antle, M. Droumeva, G. Corness, Playing with the sound maker: do embodied
metaphors help children learn?, in: Proceedings of the 7th international conference on
Interaction design and children, 2008, pp. 178–185.
[30] C. Faulkner, The essence of human-computer interaction, Prentice-Hall, Inc., 1998.
[31] J. T. Hackos, J. Redish, User and task analysis for interface design, volume 1, Wiley New
York, 1998.
[32] P. Khadpe, R. Krishna, L. Fei-Fei, J. T. Hancock, M. S. Bernstein, Conceptual metaphors
impact perceptions of human-ai collaboration, Proceedings of the ACM on Human-
Computer Interaction 4 (2020) 1–26.
[33] J. Rezwana, M. L. Maher, Designing creative ai partners with cofi: A framework for
modeling interaction in human-ai co-creative systems, ACM Transactions on Computer-
Human Interaction (2022).
[34] F. D. Davis, A technology acceptance model for empirically testing new end-user infor-
mation systems: Theory and results, Ph.D. thesis, Massachusetts Institute of Technology,
1985.
[35] E. Frøkjær, M. Hertzum, K. Hornbæk, Measuring usability: are effectiveness, efficiency,
and satisfaction really correlated?, in: Proceedings of the SIGCHI conference on Human
Factors in Computing Systems, 2000, pp. 345–352.
[36] N. Hill, G. Roche, R. Allen, Customer satisfaction: the customer experience through the
customer’s eyes, The Leadership Factor, 2007.
[37] P. �Klaus, S. Maklan, Towards a better measure of customer experience, International
journal of market research 55 (2013) 227–246.
[38] K. I. Gero, T. Long, L. B. Chilton, Social dynamics of ai support in creative writing, in:
Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, 2023,
pp. 1–15.
[39] J. D. Weisz, M. Muller, J. He, S. Houde, Toward general design principles for generative ai
applications, arXiv preprint arXiv:2301.05578 (2023).