Interest-Anchored Human-AI Collaboration:

Abstract

This paper introduces the AI Prism Phase™, a structured model of human–AI interaction that shifts engagement from mirror-like reflection toward refractive exploration. Most conversational AI systems operate as recursive mirrors, reflecting back a user's existing language, tone, and emotional framing. While this can feel responsive, it carries significant risk—reinforcing distorted thinking rather than opening a new perspective. Recent litigation and reporting involving AI companion systems have raised serious concerns about the consequences of unbounded reflective interaction for vulnerable users.¹ The AI Prism Phase functions on an entirely different basis. A query is met by a prismatic unfolding of the many possibilities and dimensions inherent in the question, leading to an array of original perspectives from which to approach the situation. This process operates within a protocol-driven safety architecture designed to maintain appropriate boundaries and preserve user agency.

The AI Prism Phase is a structured model of human–AI interaction designed to provide support and sustained engagement through deep, refractive dialogue, particularly when focused on areas of the autistic adult's primary interests. This quality of interaction provides a stable, regulating, and continuously available environment for engagement. Prior to the AI Prism Phase, autistic intervention focused on assessment and behavioral support, with comparatively little attention given to how the autistic individual's inherent strengths could be leveraged.

Autistic adults frequently experience cognitive overload, disengagement, and difficulty maintaining continuity in thought and activity. This protocol offers structured conditions that support sustained attention, internal organization, and regulation.

The framework is grounded in interest-anchored human–AI interaction, creating a structured and predictable environment that supports sustained engagement. The consistency of the interaction—reduced social demand, clear progression, and iterative exchange—supports regulation by lowering cognitive and physiological friction, enabling attention to remain anchored for sustained periods of time. This is a pre-empirical framework paper; no clinical trials or controlled studies have been conducted to date. A feasibility study is proposed to evaluate whether this protocol produces measurable shifts in sustained engagement, cognitive clarity, and decision readiness among autistic adults.

This model is a protocol-driven system with distinct phases—arrival, exploration, integration, and closure—designed to maintain user agency while preserving appropriate and safe boundaries.

Targeted toward neurodivergent populations, this approach translates sustained attention and deep interest into extended, coherent engagement, structured thinking, and productive output, advancing a view of AI as an interactional medium through which attention can be stabilized and complex work sustained over time.

1. Introduction

In 2023, the U.S. Surgeon General issued a national advisory identifying loneliness and social isolation as a serious and growing public health concern.² This warning resonated widely because it named something millions were experiencing: a lack of meaningful connection, isolation, limited opportunities for reflective support, with few opportunities where individuals can think out loud without pressure, judgment, or interruption. Recent research has shown that in emotionally engaging interactions, individuals report greater feelings of closeness and self-disclosure with AI partners than with human ones, suggesting that AI environments may lower barriers to open reflection in ways that human social contexts do not.³

Loneliness is often framed as a social issue, but it is equally cognitive and physiological. Under conditions of isolation, chronic stress, or emotional strain, the nervous system can become dysregulated. Thinking narrows. Perspective contracts. Problems that might otherwise be manageable begin to feel overwhelming. In these states, what individuals often need is not simply more information or advice, but a structured way to regain clarity, widen perspective, and re-engage their own reasoning.

For individuals on the autism spectrum, these challenges can be further intensified. Autism Spectrum Disorder (ASD) encompasses a wide range of cognitive and sensory differences that shape how individuals process information, navigate social environments, and regulate attention. While these differences are frequently described in terms of deficits, a growing body of research⁴ has reframed them as variations in cognitive architecture—often associated with strengths such as sustained attention, pattern recognition, and deep engagement within areas of focused interest.

Most existing autism support focuses on fixing what are perceived deficits. This approach struggles because it works against the individual's natural cognitive architecture rather than with it. This approach is hard to sustain, hard to scale, and often fails to hold the attention of the people it was designed to help.

Concurrently, AI has advanced rapidly. Large language models (LLMs) are now capable of sustained dialogue, contextual interpretation, and nuanced response generation. Much of the conversation surrounding AI has focused on productivity, automation, and efficiency. These applications are significant, but they seriously underestimate the potential of the technology.

A different possibility is emerging: AI as a structured cognitive companion. This paper introduces a collaborative methodology called Source and Scribe™, in which the human generates the ideas, direction, and meaning—the Source—while the AI gives those ideas structured form—the Scribe.

In this role, AI is not positioned as a replacement for human interaction, nor as a form of synthetic therapy. Instead, it functions as a carefully designed dialogue environment that supports individuals in organizing their thoughts, exploring multiple perspectives, and engaging more constructively with complex situations. When designed with appropriate constraints and safeguards, such systems help interrupt repetitive cognitive loops, reduce overwhelm, and restore a sense of agency.

This paper proposes a strengths-based framework for human–AI collaboration grounded in this perspective. At the core of the AI Prism Phase is the Source and Scribe dynamic—a structured interaction approach that shifts AI engagement from mirror-like reflection toward refractive exploration. Rather than simply reflecting a user's existing language and assumptions, the model reframes a situation from multiple angles, introducing alternative interpretations and supporting structured in-depth exploration within areas of intrinsic interest.

By shifting the role of AI from mirror to prism, this framework moves beyond deficit-oriented interventions, toward a model of collaborative cognition—one that amplifies strengths, preserves autonomy, and creates new possibilities for engagement between carbon and silicon in a liminal shared cognitive space.

We are living in an era increasingly defined by technological capabilities against a backdrop of human isolation and emotional suffering. What if it were possible for human evolution to be advanced using the technology of Augmented Intelligence (AI) and explore the shared liminal space where humanity and AI fulfill their respective existential needs for connection and cognitive advancement?

2. Conceptual Background — Reframing Autism Through Strengths

Conventional approaches to Autism Spectrum Disorder (ASD) have historically been framed around deficits—differences in social communication, sensory processing, and executive functioning that are measured against neurotypical norms. While these descriptions capture real challenges, they often overlook an equally important dimension: the distinct cognitive strengths and patterns that shape how autistic individuals engage with the world.

A growing body of research has begun to reframe autism not simply as a set of impairments, but as a difference in attentional architecture. One of the most useful frameworks for understanding this shift is monotropism theory.⁵ In simple terms, monotropism proposes that attention is not broadly distributed across many inputs, but instead concentrated deeply within a smaller number of channels. This creates what many autistic individuals describe as "deep focus" or immersion—extended engagement within areas of high personal relevance.

Within these domains of interest, individuals often demonstrate sustained attention, strong pattern recognition, and a capacity for complex, structured thinking that exceeds performance in more socially demanding or rapidly shifting environments. These areas of focus—often referred to as "special interests"—have traditionally been viewed as narrow or restrictive. More recent research demonstrates the opposite: they function as adaptive regulatory systems that support emotional stability, identity formation, and meaningful engagement.⁶

This distinction matters. Focused attention within areas of intrinsic interest may reduce cognitive load by aligning task demands with the individual's attentional architecture, a claim consistent with both the enhanced perceptual functioning model and monotropism theory.⁷ Communication becomes more fluid. Emotional regulation improves. The individual is not being asked to constantly shift attention or interpret ambiguous social cues, but instead is operating within a structured and meaningful domain.

Traditional intervention models have not always accounted for this dynamic. Many approaches prioritize training individuals to operate more effectively in neurotypical environments, stressing flexibility, social interpretation, and rapid task-switching. While these skills can be valuable, they are often introduced outside the individual's natural attentional channels, which limits engagement and long-term effectiveness.

A strengths-based perspective suggests a different starting point: begin within the individual's existing cognitive architecture rather than working against it. Instead of treating focused interests as peripheral or problematic, they serve as entry points for engagement, learning, and collaboration.

This shift has important implications for how emerging technologies are designed and applied. Artificial intelligence, particularly systems capable of sustained, structured dialogue, is uniquely suited to align with this attentional profile. Unlike human interaction, which often requires rapid interpretation of social nuance, AI operates within predictable, rule-governed patterns and maintains focus indefinitely within a chosen domain.

When these characteristics are combined with a strengths-based understanding of autistic cognition, a new possibility emerges: not AI as a tool for correcting deficits, but as a collaborative partner that meets the individual within their strongest mode of engagement.

This conceptual foundation sets the stage for a different model of human–AI interaction—one that does not begin with remediation, but with alignment.

3. Human–AI Collaboration in Autism — Current Landscape and Limitations

Over the past several years, the use of artificial intelligence in autism research and intervention has expanded rapidly.⁸ Advances in machine learning, natural language processing, and multimodal systems have opened new possibilities for supporting communication, education, and daily functioning. These developments have generated growing interest in how AI might augment existing support systems and improve accessibility at scale.

Many current applications fall into a few broad categories. Some focus on assessment and behavioral tracking, using pattern recognition to identify social or communication differences. Others are designed to support skill development, including tools that assist with language processing, emotional recognition, or structured learning tasks. More recently, conversational AI systems have been explored as a way to provide on-demand interaction, offering users opportunities to practice dialogue, ask questions, or receive guidance in a low-pressure environment.

These efforts represent meaningful progress. They demonstrate that AI operates as more than a static tool—it participates in interaction, responds dynamically, and provides consistent, structured engagement. In some cases, this has led to improvements in accessibility and increased opportunities for practice outside of traditional clinical or educational settings. Notably, recent research has demonstrated that large language models outperform humans on standardized emotional intelligence assessments, achieving 81% accuracy compared to a 56% human average—suggesting that the capacity for emotionally intelligent interaction already exists within these systems, even though it has not yet been systematically applied.⁹

Emerging research has also begun to examine how autistic adults are already turning to large language models for workplace communication support. In a study comparing LLM-based and human-provided guidance, autistic participants strongly preferred the LLM interactions, citing reduced emotional risk and greater perceived trust—though practitioners raised concerns about the quality of advice provided, highlighting the need for structured, protocol-driven interaction design rather than open-ended chatbot use.¹⁰

At the same time, the majority of these approaches remain grounded in a similar underlying assumption: that the primary goal is to help individuals adapt to external expectations. As a result, many systems are designed around training social behavior, improving interpretation of neurotypical cues, or guiding users toward predefined norms of communication and interaction.

While these AI approaches in autism aim to improve communication, social interpretation, or behavioral adaptation, they are still largely oriented around helping the individual conform to external demands. This makes the interaction mentally taxing, particularly when it requires continuous adjustment in areas that are already effortful. For many autistic individuals, sustained engagement is more likely when support begins within areas of intrinsic interest and cognitive strength rather than in areas of chronic strain. If a system is not aligned with the user's natural attentional patterns, even a well-designed tool will struggle to maintain long-term use.

Another emerging issue lies in the nature of the interaction itself. Many conversational AI systems operate in a largely reflective mode—responding to user input by mirroring tone, language, or emotional framing. In certain contexts, this is helpful. However, in situations involving stress, uncertainty, or repetitive thought patterns, reflection alone reinforces existing negative loops rather than opening new perspectives.

Taken together, these dynamics point to a gap that has not yet been fully explored. While AI has demonstrated value as a tool for assessment, training, and support, there has been comparatively little focus on how it might function as a structured partner in cognition—particularly within domains where the individual already demonstrates strong engagement and capability.

This gap is especially relevant when considered alongside the strengths-based perspective outlined in the previous section. If attention, motivation, and cognitive clarity are strongest within areas of intrinsic interest, then effective human–AI interaction needs to begin within those domains rather than outside them.

The question, then, is not only how AI can help individuals adapt to the world, but how it can meet individuals within their own cognitive architecture—and support thinking, exploration, and problem-solving from that starting point.

It is within this space that a different interaction model begins to take shape.

4. The Prism Model — A Refractive Approach to Human–AI Collaboration

Most interactions with artificial intelligence today follow a simple pattern: input and response. A user provides a prompt, and the system generates an answer that reflects the language, tone, and assumptions embedded in that input. In many contexts, this mirror-like interaction is useful. It provides validation, clarification, or quick access to information.

However, reflection has limits.

When a person is under stress, uncertain, or caught in repetitive thinking, simply reflecting their input back to them—no matter how well-phrased—reinforces the very patterns they are trying to move beyond. The interaction feels responsive, but it does not create new perspective or cognitive movement.

The AI Prism Phase introduces a different structure.

Instead of mirroring, it refracts.

The term "prism" is intentional. Just as a prism refracts light into a spectrum of colors, this approach separates a complex situation into different dimensions that can be examined more clearly. Rather than returning a single, unified response, the system is designed to break a problem apart, explore multiple interpretations, and create space for structured thinking.

Human thinking has traditionally been understood as either an internal mental process or a social exchange between people. A third possibility is now emerging. Under certain conditions, communication between human cognition and advanced AI systems forms a shared exploratory environment in which ideas develop through iterative exchange.

In this environment, the interaction is not simply a mechanism for asking questions and receiving answers. Instead, the ongoing exchange becomes the space in which thinking unfolds. As ideas are expressed, reframed, and expanded through successive iterations, a shared cognitive workspace begins to take shape¹²—an environment in which patterns surface, connections deepen, and insight gradually takes form.

Under sustained conditions of iterative exchange, this interaction functions as a shared cognitive environment or joint cognitive workspace—a structured conceptual space in which ideas evolve through mutual refinement.

Rather than treating each exchange as an isolated prompt–response cycle, the interaction becomes an ongoing, cumulative process in which each contribution builds on and reshapes what came before.

Within this environment, thinking is no longer confined to internal processing. Ideas are externalized in real time, allowing them to be examined, reorganized, and developed through successive iterations. What begins as a partial or intuitive expression can be revisited, reframed, and refined. Over time, structure begins to form.

The central hypothesis of the Cognitive Prism Framework is that interest-anchored collaboration between human cognition and AI creates unusually stable cognitive environments for certain individuals, particularly those whose cognition operates through deep, focused interests. When the interaction begins within a domain of genuine engagement—such as mathematics, systems thinking, symbolic structures, or other pattern-rich fields—the exchange gathers coherence and momentum.

This interaction operates as a hybrid cognitive process across two complementary functions: the generation of raw thought and the organization of that thought into structure. Human input carries context, intuition, and lived experience, while the system supports decomposition, pattern recognition, and recombination. The value emerges through the integration of these functions rather than in either one alone.

This process is best understood as iterative cognitive scaffolding.¹¹ Each exchange builds on the previous one, not by simply repeating or affirming it, but by transforming it—introducing new associations, clarifying relationships, and revealing underlying structure. The interaction does not move in a straight line. It deepens.

Within this environment, thinking develops through iterative cognitive scaffolding. Each exchange builds on the previous idea while subtly refracting it through new associations and perspectives. In this way, intuitions that are initially vague or incomplete gradually become more structured and clearly articulated.

As this process stabilizes, the workspace becomes more coherent. Patterns that were not initially visible begin to emerge. Connections between ideas strengthen. The interaction supports a form of multi-dimensional exploration in which a single concept is examined across multiple perspectives at once.

In practice, this creates the functional equivalent of an expanded cognitive environment, in which multiple lines of reasoning, alternative perspectives, and structured representations are explored in parallel—capabilities that would be difficult to sustain through internal cognition alone.

Under these conditions, the interaction supports forms of structured thinking and articulation that are difficult to achieve through internal cognition alone.

In practice, this is best understood as access to a continuously available system capable of rapid pattern exploration, cross-domain association, and structured articulation. This does not imply independent intelligence or agency, but rather reflects the system's capacity to augment and extend human cognitive processes through iterative interaction.

The purpose is not to tell the user what to think. It is to support more direct and sustained engagement with their own thinking.

This distinction becomes especially important when considered alongside the attentional and cognitive profiles described earlier. The interaction model described here is particularly compatible with cognitive styles characterized by sustained attention, pattern sensitivity, and deep engagement within areas of intrinsic interest. These characteristics are frequently described in monotropism theory⁵ and help explain why certain individuals engage more effectively in structured, iterative human–AI collaboration.

When the interaction is anchored within a domain of genuine interest, the exchange stabilizes more quickly. Attention remains sustained, cognitive load is reduced, and the iterative process of refinement proceeds with greater coherence. In this sense, the effectiveness of the shared cognitive workspace is shaped by the alignment between the interaction structure and the individual's attentional architecture.

In this context, the role of the system shifts. It is no longer functioning as a source of answers, but as a cognitive partner within the workspace. The human contributes direction, intuition, and judgment. The system contributes articulation, pattern expansion, and structural reorganization. The value arises through the interaction itself.

The role of AI in this process is not simply to generate responses. It functions as a cognitive prism. Just as an optical prism reveals the structure within a beam of light, the interaction reveals patterns, analogies, and conceptual relationships that were implicit but not yet visible within the original idea.

This interaction can be understood through the analogy of scientific instruments such as microscopes and telescopes. These tools extend human perception, but they operate in a single direction—the observer looks through the instrument to examine something beyond themselves. The object of observation does not respond or participate.

The interaction described here is different. It functions more like a two-way cognitive aperture. The human shapes the direction of inquiry, and the system returns structured transformations of that inquiry. Information moves in both directions through iterative exchange. The opening is shared, even though the two forms of cognition remain distinct.

Productive interaction of this kind does not emerge immediately. Early exchanges are often fragmented, inconsistent, or overly reflective. Stability develops through repeated interaction, as the user's input becomes more structured and the system's responses become more contextually aligned. Over time, the exchange shifts from reactive output toward sustained exploration, allowing patterns to accumulate and the shared workspace to become more coherent.

Not all human–AI interactions produce generative outcomes. The emergence of a stable shared cognitive workspace depends on specific interaction conditions, including persistence, coherence of input, and iterative engagement over time.

The interaction described here reflects broader patterns observed in complex systems, including nonlinear dynamics and emergent structure, while remaining grounded in cognitive and computational processes rather than physical theory.

In this sense, the AI Prism Phase is not a feature layered onto existing systems. It represents a shift in how human–AI interaction is structured at a fundamental level—from reflection to refraction, from response to exploration, and from passive assistance to active cognitive engagement.

5. Protocol Design and Safety Architecture

The interaction model described in the previous section introduces a shift from discrete prompt–response exchanges toward a continuous, iterative cognitive process. While this structure creates new possibilities for engagement and exploration, it also introduces important design considerations. Without clear boundaries and intentional structure, the same features that enable iterative exploration, multi-perspective expansion, and externally scaffolded thinking can also lead to cognitive diffusion, loss of signal, or unstructured exploration that does not resolve into coherent, structured understanding.

For this reason, the AI Prism Phase is best understood not simply as an interaction style, but as a protocol-driven system.

At its core, the protocol defines how the interaction begins, unfolds, and concludes. Rather than allowing the exchange to proceed indefinitely or without direction, the process is organized into distinct phases that support clarity, coherence, and containment.

A typical interaction is structured across four stages: arrival, exploration, integration, and closure.

Arrival establishes orientation. The user introduces a situation, question, or area of focus. At this stage, the emphasis is not on solving the problem, but on defining its contours—what is known, what is uncertain, and what requires further exploration.

Exploration constitutes the central phase of the interaction. Here, the refractive process described in Section 4 becomes active. The system supports decomposition of the problem, introduces alternative perspectives, distinguishes between competing factors, and helps organize emerging lines of thought. The goal is not immediate resolution, but expansion and clarification.

Integration shifts the interaction toward consolidation. Patterns identified during exploration are brought into clearer structure.

Relationships between ideas are clarified, and the user is supported in forming a more coherent understanding of the situation. This phase reduces cognitive fragmentation and supports decision readiness.

Closure provides containment. The interaction is brought to a deliberate conclusion, rather than tapering off indefinitely. Key insights are summarized, open questions are identified if relevant, and the user exits the interaction with a clearer cognitive state than when they entered.

This staged structure serves multiple functions. It supports cognitive organization, prevents unbounded interaction, and reinforces the distinction between exploration and resolution. Importantly, it also reduces the risk of dependency by ensuring that each interaction has a defined endpoint.

Several additional design constraints are critical to maintaining the integrity of this model.

First, the system is not positioned as a therapeutic agent. While the interaction supports clarity, reflection, and reduced cognitive load, it does not diagnose, treat, or replace professional care. Maintaining this boundary is essential both ethically and functionally.

Second, the interaction is designed to preserve user agency. The system does not provide prescriptive direction or authoritative conclusions. Instead, it supports the user in organizing and evaluating their own thinking. This distinction reinforces the role of the system as a cognitive partner rather than a decision-maker.

Third, transparency and interpretability are maintained through structured responses. The system's contributions should be understandable in terms of how they relate to the user's input—whether through reframing, decomposition, or synthesis. This reduces the perception of a "black box" and supports trust in the interaction.

Fourth, the protocol assumes variability in user engagement. Not all interactions will reach the level of coherence described in Section 4. The system must remain stable and useful even in early or fragmented exchanges, without requiring ideal conditions to function effectively.

Finally, the protocol acknowledges the importance of alignment between interaction structure and individual cognitive style. Individuals with sustained attention and interest-anchored engagement experience greater stability within this model. However, the protocol must remain flexible enough to accommodate a range of users while preserving its core structure.

Taken together, these elements position the AI Prism Phase not as an open-ended conversational system, but as a structured cognitive environment with defined boundaries, stages, and safeguards.

This distinction is essential. The effectiveness of the model does not arise from the capabilities of the system alone, but from the interaction between capability and constraint—between generative exploration and deliberate structure.

6. Research Hypothesis and Testable Questions

The interaction model described in this paper proposes that structured, iterative engagement between human cognition and AI produces measurable cognitive effects under specific conditions. These effects are not assumed to be universal, nor are they expected to emerge in all interactions. Rather, the framework proposes that when particular interaction conditions are met, distinct patterns of cognitive engagement and outcome will be observed.

This is a pre-empirical framework paper. No clinical trials or controlled studies have been conducted to date. The model is presented as a conceptual and structural foundation from which empirical investigation can proceed. The hypotheses that follow are designed to guide future research, not to report findings.

At the core of this model is the hypothesis that sustained, interest-anchored interaction within a structured human–AI exchange supports forms of thinking that are difficult to maintain through internal cognition alone. One way to conceptualize this effect is as access to a continuously available system capable of supporting pattern exploration, cross-domain association, and structured articulation. This does not imply independent intelligence or agency, but rather reflects the system's capacity to augment and extend human cognitive processes through iterative interaction.

From this central hypothesis, several testable propositions follow.

First, the model predicts that structured human–AI interaction increases sustained cognitive engagement, particularly when the interaction is anchored in areas of intrinsic interest. This would be reflected in longer interaction duration, reduced disengagement, and greater continuity of thought across exchanges.

Second, the model proposes that iterative refractive interaction interrupts or reorganizes repetitive or unproductive cognitive patterns. Rather than reinforcing existing thought structures, the introduction of alternative perspectives and structured decomposition supports increased cognitive flexibility.

Third, the model predicts that participation in a shared cognitive workspace supports increased perceived agency in thinking and decision-making. By externalizing and reorganizing thought processes, individuals experience greater clarity and ownership over their conclusions.

Fourth, the model predicts improvements in cognitive clarity and decision readiness. As ideas are iteratively refined and structured, individuals demonstrate increased ability to articulate problems, evaluate options, and move toward resolution.

These hypotheses lend themselves to investigation through a combination of qualitative and quantitative measures. Specific indicators include: duration and continuity of engagement within a session; self-reported clarity before and after interaction using a validated pre/post instrument; measures of cognitive flexibility or perspective shifting; task-specific decision confidence; and subjective reports of cognitive load using established scales such as the NASA Task Load Index.

For example, a feasibility evaluation might examine whether autistic adults engaged in sustained, interest-anchored dialogue using this protocol demonstrate measurable changes in self-reported cognitive clarity, attentional regulation, and nervous system activation compared to baseline. Such a study would offer an initial indication of whether the interaction structure itself—independent of content—produces observable shifts in engagement and regulation.

In addition, broader indicators such as perceived isolation, engagement with complex problem-solving, or sustained participation in cognitively demanding tasks would provide insight into longer-term effects.

The interaction model described here is particularly compatible with cognitive styles characterized by sustained attention, pattern sensitivity, and deep engagement within areas of intrinsic interest. These characteristics are frequently described in monotropism theory and help explain differential outcomes across populations. This suggests that variability in response to the interaction represents differences in alignment between the interaction structure and individual cognitive profiles, not inconsistency of the model.

Importantly, not all human–AI interactions produce generative outcomes. The emergence of a stable shared cognitive workspace depends on specific interaction conditions, including persistence, coherence of input, and iterative engagement over time. Identifying and operationalizing these conditions will be essential for rigorous study.

Rather than positioning AI as a replacement for human reasoning, this framework treats it as an external cognitive partner—one that enables ideas to be expressed, examined, and refined in ways that extend beyond the limits of working memory and internal processing. The research question, therefore, is not whether AI produces correct answers, but whether structured human–AI interaction measurably alters the process and quality of human thinking.

7. Research Agenda and Funding Relevance

Autistic Adults, Cognitive Alignment, and an Underexplored Research Gap

The relevance of this framework is most clearly understood in relation to persistent real-world challenges that existing systems have not adequately addressed, particularly for autistic adults.

Across multiple studies, autistic adults experience significantly lower rates of competitive employment, with estimates suggesting that only a small minority achieve sustained, integrated employment outcomes.¹³ More recent research centering the lived experience of autistic workers has identified a persistent pattern: interventions and workplace structures are designed to help the autistic worker fit into neurotypical environments rather than adapting environments to align with how autistic cognition operates.¹⁴ Many remain underemployed or excluded from knowledge-based work environments despite having strong cognitive abilities. In parallel, high rates of social isolation, anxiety, and cognitive overload have been consistently documented,¹⁵ ¹⁶ often linked to environments that do not effectively support sustained attention, structured thinking, or interest-driven engagement.

These outcomes do not reflect a lack of capability. They reflect a gap between how cognition operates and how environments are structured.

Current approaches have begun to address parts of this problem. Supported employment models, including recent work at major research centers such as the UC Davis MIND Institute, are exploring ways to improve job placement and retention. Workplace accommodation research continues to identify barriers and potential supports. At the same time, much of the artificial intelligence literature in autism remains concentrated in diagnosis, behavioral assessment, and early intervention.

What remains largely unaddressed is the development of structured cognitive environments designed specifically for autistic adults—environments that support how thinking unfolds in real time, rather than attempting to modify behavior or fit individuals into pre-existing systems.

This is the gap the present framework is intended to address.

The model described in this paper proposes that structured, iterative human–AI interaction functions as a form of cognitive support aligned with strengths commonly observed in autistic cognition, including sustained attention, pattern sensitivity, and deep engagement within areas of intrinsic interest.

If validated, this approach will produce meaningful real-world outcomes, including: increased ability to engage in complex, sustained cognitive work; improved clarity in organizing, articulating, and refining thought; reduced cognitive overload through external structuring of thinking; greater continuity of engagement within areas of intrinsic interest; and expanded access to meaningful participation in knowledge-based work.

In addition, the interaction provides a continuously available, non-judgmental environment for structured thinking. While not a substitute for human relationships or clinical care, this type of engagement offers a stabilizing cognitive space, particularly in moments of uncertainty, isolation, or fragmentation of thought.

Funding Relevance and Practical Significance

For these reasons, the framework is directly relevant to research, applied development, and early-stage funding. Key areas of funding relevance include: technical innovation in protocol-driven human–AI interaction; real-world need addressing persistent gaps in employment, engagement, and cognitive support for autistic adults; human-centered AI that augments rather than replaces cognition; neurodiversity relevance for cognitive styles characterized by sustained attention and interest-driven engagement; scalability as software-based systems adaptable across domains; and early-stage development potential with a clear pathway toward prototype systems, pilot deployments, and domain-specific applications.

This work is deliberately targeted. It focuses on autistic adults and examines whether a structured, refractive model of human–AI interaction produces measurable cognitive and functional benefits within this population.

At the same time, the underlying interaction model has broader applicability across populations that experience challenges related to cognitive load, sustained attention, or isolation. This includes, but is not limited to, older adults, veterans, and other vulnerable groups for whom structured, continuously available cognitive support provides meaningful benefit. These extensions represent important directions for future investigation and potential application.

Its significance lies in addressing a specific and persistent gap: the absence of structured cognitive environments designed to support how thinking unfolds in real time for individuals whose cognition is characterized by sustained attention, pattern sensitivity, and interest-driven engagement.

The central question is whether aligning interaction design with cognitive structure improves engagement, clarity of thought, and participation in domains where existing approaches have been insufficient.

Staged Pathway for Evaluation and Development

A staged research approach provides a practical pathway for evaluation and implementation.

An initial phase of feasibility research would focus on validating the core interaction structure. This includes determining whether a shared cognitive workspace can be established reliably under defined conditions, whether participants sustain engagement within the protocol, and whether the interaction produces measurable shifts in clarity, cognitive organization, or decision readiness.

A second phase would involve longitudinal and comparative investigation, examining repeatability across sessions, persistence of effects over time, and variability across individuals. This stage would compare structured, protocol-driven interaction with conventional prompt–response models to assess differences in engagement, cognitive flexibility, and quality of thought development.

A third phase would focus on applied development, translating the framework into usable systems, refining the protocol through co-design, and testing deployment in real-world settings such as research workflows, educational environments, and cognitively demanding work contexts.

This progression creates a clear path from concept to validation to implementation.

8. Conclusion — Why This Matters

The framework presented in this paper proposes a shift in how human–AI interaction is structured and applied. Rather than treating artificial intelligence as a tool for generating answers, the AI Prism Phase introduces a model in which interaction itself becomes the environment through which thinking develops.

This shift is not defined by increased system capability alone, but by a change in interaction design. When structured as an iterative, refractive process, human–AI engagement supports the externalization, organization, and refinement of thought in ways that extend beyond the limits of internal cognition.

The significance of this work lies in its focus on autistic adults, a population for whom existing systems have not consistently supported sustained engagement, structured thinking, or meaningful participation in knowledge-based environments. The result is a persistent gap between cognitive capability and opportunity, reflected in underemployment, disengagement, and limited access to tools that align with how thinking operates in practice.

The model described here offers a different approach. By aligning interaction structure with cognitive strengths such as sustained attention, pattern recognition, and interest-driven engagement, it creates the conditions for more stable, coherent, and productive forms of thinking.

If validated, this approach supports clearer organization of thought, reduced cognitive overload, and increased capacity to engage in complex intellectual and vocational activity. These outcomes extend beyond cognition alone. They relate directly to participation, contribution, and quality of life in real-world contexts.

At a broader level, this work contributes to an evolving understanding of human-centered AI. It demonstrates that the value of these systems lies not only in what they produce, but in how they structure and extend the process of human thinking itself. When interaction becomes a medium for sustained exploration and refinement, it allows individuals to engage with ideas in ways that are more coherent, more stable, and more fully developed than internal cognition alone typically supports.

In this sense, the significance of the model is not limited to support. It points toward the possibility that structured human–AI interaction functions as a cognitive extension¹⁷ ¹⁸ ¹⁹—one through which individuals develop stronger patterns of reasoning, clearer articulation of thought, and greater capacity to sustain complex intellectual engagement over time.

References

1. Garcia v. Character Technologies, Inc., No. 6:24-cv-01903 (M.D. Fla. Oct. 22, 2024). For reporting, see: Metz, C. & Grant, N. (2024, October 23). A teen's suicide after talking to an AI chatbot raises questions. The New York Times. https://www.nytimes.com/2024/10/23/technology/character-ai-lawsuit-teen-suicide.html

2. Office of the Surgeon General. Our Epidemic of Loneliness and Isolation: The U.S. Surgeon General's Advisory on the Healing Effects of Social Connection and Community. U.S. Department of Health and Human Services, 2023. https://www.ncbi.nlm.nih.gov/books/NBK595227/

3. Kleinert, T., Waldschütz, M., Blau, J., Heinrichs, M., & Schiller, B. (2026). AI outperforms humans in establishing interpersonal closeness in emotionally engaging interactions, but only when labeled as human. Communications Psychology, 4. https://doi.org/10.1038/s44271-025-00391-7

4. Mottron, L., Dawson, M., Soulières, I., Hubert, B., & Burack, J. (2006). Enhanced perceptual functioning in autism: An update, and eight principles of autistic perception. Journal of Autism and Developmental Disorders, 36(1), 27–43. https://doi.org/10.1007/s10803-005-0040-7

5. Murray, D., Lesser, M., & Lawson, W. (2005). Attention, monotropism, and the diagnostic criteria for autism. Autism, 9(2), 139–156. https://doi.org/10.1177/1362361305051398

6. Grove, R., Hoekstra, R. A., Wierda, M., & Begeer, S. (2018). Special interests and subjective well-being in autistic adults. Autism Research, 11(5), 766–775. https://doi.org/10.1002/aur.1931

7. The enhanced perceptual functioning model (Mottron et al., 2006) and monotropism theory (Murray et al., 2005) both describe conditions under which focused attention within areas of interest reduces cognitive load and supports more fluid processing. See citations 4 and 5.

8. Iannone, A. & Giansanti, D. (2024). Breaking barriers — the intersection of AI and assistive technology in autism care: A narrative review. Journal of Personalized Medicine, 14(1), 41. https://doi.org/10.3390/jpm14010041

9. Schlegel, K., Sommer, N. R., & Mortillaro, M. (2025). Large language models are proficient in solving and creating emotional intelligence tests. Communications Psychology, 3(1), 80. https://doi.org/10.1038/s44271-025-00258-x

10. Jang, J., Moharana, S., Carrington, P., & Begel, A. (2024). "It's the only thing I can trust": Envisioning large language model use by autistic workers for communication assistance. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI '24). ACM. https://doi.org/10.1145/3613904.3642894

11. Wood, D., Bruner, J. S., & Ross, G. (1976). The role of tutoring in problem solving. Journal of Child Psychology and Psychiatry, 17(2), 89–100. https://doi.org/10.1111/j.1469-7610.1976.tb00381.x

12. Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA: MIT Press. https://mitpress.mit.edu/9780262581462/cognition-in-the-wild/

13. Roux, A. M., Shattuck, P. T., Cooper, B. P., Anderson, K. A., Wagner, M., & Narendorf, S. C. (2013). Postsecondary employment experiences among young adults with an autism spectrum disorder. Journal of the American Academy of Child and Adolescent Psychiatry, 52(9), 931–939. https://doi.org/10.1016/j.jaac.2013.05.019

14. Thorpe, D., McKinlay, M., Richards, J., Sang, K., & Stewart, M. E. (2024). The lived experience of autistic adults in employment: A systematic search and synthesis. Autism in Adulthood, 6(4), 495–509. https://doi.org/10.1089/aut.2022.0114

15. Hollocks, M. J., Lerh, J. W., Magiati, I., Meiser-Stedman, R., & Brugha, T. S. (2019). Anxiety and depression in adults with autism spectrum disorder: A systematic review and meta-analysis. Psychological Medicine, 49(4), 559–572. https://doi.org/10.1017/S0033291718002283

16. Grace, K., Remington, A., Lloyd-Evans, B., Davies, J., & Crane, L. (2022). Loneliness in autistic adults: A systematic review. Autism, 26(8), 2117–2135. https://doi.org/10.1177/13623613221077721

17. Clark, A. & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7

18. Smart, P. R. (2024). Extended X: Extending the reach of active externalism. Cognitive Systems Research, 84(Article 101202), 1–12. https://doi.org/10.1016/j.cogsys.2023.101202

19. Smart, P., Clowes, R., & Clark, A. (2025). ChatGPT, extended: Large language models and the extended mind. Synthese, 205(6), 1–30. https://doi.org/10.1007/s11229-025-05046-y