As artificial intelligence systems become increasingly integrated into human creative practice, a central question emerges: What kind of cognitive theory best explains human-AI co-creation?
Traditional AI systems have largely been grounded in information processing models of cognition. Within this framework, intelligence is often understood as symbolic manipulation, internal representation, planning, optimization, and goal-directed computation. While these models have been extraordinarily successful in many domains, they struggle to fully explain the dynamics of co-creative interaction.
Co-creation is fundamentally improvisational, participatory, adaptive, relational, and emergent. Meaning emerges continuously through interaction between collaborators rather than existing fully formed inside either participant. This is why enaction provides such a powerful paradigm for co-creative AI.
Enaction shifts the focus of cognition away from isolated internal computation and toward interaction itself. Rather than viewing intelligence as something that happens solely inside the mind or machine, enaction proposes that cognition emerges through active engagement with environments and other agents. This interaction-centered perspective aligns remarkably well with the dynamics of human-AI co-creation.
The cognitive science theory of enaction emerged through the work of researchers such as Francisco Varela, Evan Thompson, and Eleanor Rosch, who proposed that cognition is not merely computation over internal representations. Instead, cognition emerges through embodied interaction with the world.
Within enaction, perception is active, meaning is constructed through engagement, and intelligence arises through adaptive interaction. Agents continuously regulate their relationship with the environment in a process known as sense-making.
Meaning is therefore not passively received from the world, nor is it fully preconstructed internally. Instead, meaning emerges through participation.
This becomes especially important in creative interaction. Creativity is rarely static or predetermined. Creative processes evolve dynamically through feedback, interpretation, experimentation, coordination, and adaptation. These dynamics are precisely what enaction is designed to explain.
Many traditional AI paradigms implicitly treat creativity as idea generation, search through solution spaces, symbolic recombination, or optimization over representations. These approaches work reasonably well for autonomous generation systems.
However, co-creative AI is fundamentally different.
In co-creative systems, humans adapt to AI, AI adapts to humans, interaction histories matter, meaning evolves dynamically, and creative trajectories emerge through participation itself. The creative process cannot be fully reduced to isolated inputs and outputs.
As argued in The Five Pillars of Enaction as a Theoretical Framework for Co-Creative Artificial Intelligence, co-creation is inherently dynamic, open-ended, improvisational, and interaction-centered. Traditional information-processing models struggle to fully capture these relational dynamics. Enaction, by contrast, places interaction at the center of cognition itself.
The ICCC 2024 Best Paper, The Five Pillars of Enaction as a Theoretical Framework for Co-Creative Artificial Intelligence, proposed a formal enactive framework for analyzing and designing co-creative AI systems.
The framework identified five core pillars of enaction: autonomy, sense-making, embodiment, emergence, and experience. Together, these pillars provide a conceptual structure for understanding co-creative interaction.
In enaction, autonomy refers to the self-organizing nature of living cognitive systems. Within co-creative AI, autonomy becomes relational rather than absolute. A co-creative system must maintain enough independence to contribute meaningfully while still remaining responsive to human collaborators.
This creates a delicate balance: too little autonomy reduces the system to a passive tool, while too much autonomy transforms it into an isolated generator disconnected from collaboration. Co-creative AI therefore operates within a dynamic space of shared agency.
Sense-making is perhaps the most important enactive concept for co-creative AI. Through interaction, collaborators continuously construct meaning together. During co-creation, each action reshapes interpretation, interpretation reorganizes future action, and new affordances emerge recursively.
Creative interaction therefore becomes an evolving process of participatory meaning construction. This idea directly influenced the development of Creative Sense-Making as a framework for modeling and quantifying interaction dynamics during co-creation.
Enaction emphasizes that cognition is fundamentally embodied. Embodiment does not necessarily require biological bodies alone. In co-creative AI, embodiment can emerge through interfaces, interaction loops, sensorimotor coupling, drawing gestures, conversational exchange, or shared manipulation of creative artifacts.
The important insight is that cognition is shaped through active engagement rather than detached symbolic reasoning alone. This became especially important in systems such as The Drawing Apprentice, where meaning emerged through reciprocal sketch interaction between human and AI participants.
One of the defining features of co-creation is emergence. Creative outcomes often cannot be predicted in advance because they arise through interaction between coupled participants. The resulting creativity belongs neither solely to the human nor solely to the AI.
Instead, creativity emerges from the interaction itself.
This idea appears repeatedly throughout co-creative AI research. Interaction dynamics, participatory sense-making, conceptual shifts, collaborative trajectories, and adaptive coordination all contribute to emergent creativity. The enactive framework provides a natural theoretical basis for understanding these dynamics.
The final pillar emphasizes lived experience. Creative interaction is not purely computational. It is experiential.
Human collaborators experience surprise, tension, rhythm, inspiration, ambiguity, and engagement during co-creation. These experiential dimensions fundamentally shape the creative process.
Traditional computational frameworks often struggle to account for these phenomenological qualities. Enaction, however, explicitly acknowledges that cognition and experience are inseparable. This becomes critically important in the design of human-centered co-creative systems.
The Drawing Apprentice became one of the earliest practical demonstrations of enactive principles within co-creative AI. Rather than functioning as a static image generator, the system participated dynamically in collaborative drawing sessions with users.
The interaction unfolded through turn-taking, mutual adaptation, sketch interpretation, conceptual shifts, and evolving collaborative trajectories. The system therefore functioned less like a tool and more like an interactive creative partner.
This represented a major shift in computational creativity research. Rather than asking, “Can AI autonomously create art?” the Drawing Apprentice instead explored: “How do humans and AI systems create together through interaction?”
That question remains central to co-creative AI today.
One of the most important implications of enaction is that intelligence itself may be fundamentally relational.
From an enactive perspective, intelligence is not isolated computation, cognition is not detached representation, and creativity is not solitary production. Instead, intelligence emerges through interaction, meaning emerges through participation, and creativity emerges through adaptive coupling.
This perspective increasingly aligns with modern developments in human-centered AI, hybrid intelligence, participatory AI, interactive machine learning, and collaborative cognition.
Recent frameworks such as The Co-Creative Design Framework for Hybrid Intelligence continue extending these ideas into contemporary generative AI systems by emphasizing agency, interaction dynamics, and communication as central dimensions of co-creative design.
The rise of generative AI has renewed interest in many ideas explored earlier within co-creative AI research. However, enaction reveals an important distinction between generation and participation.
Generative systems primarily produce outputs. Enactive co-creative systems participate in evolving interaction.
This difference is profound.
A generative system may create content independently. An enactive co-creative system helps construct meaning dynamically alongside human collaborators. This interaction changes the human, the AI, and the creative process itself.
Enaction provides one of the strongest theoretical foundations currently available for understanding co-creative AI. It explains why interaction matters, participation matters, adaptation matters, and why meaning cannot be reduced to static computation alone.
As AI systems become increasingly integrated into creative practice, education, research, and everyday cognition, interaction-centered frameworks will likely become increasingly important.
The future of AI may therefore depend not merely on building more intelligent systems. It may depend on building systems capable of participating meaningfully in shared processes of human sense-making and creative emergence.