IMAGINE: AI-Mediated Communication Model
- IMAGINE is an integrated model that theorizes real-time AI-mediated communication through a closed-loop system consisting of AA-creator, AA-receptor, and AA-negotiator.
- The model leverages continuous measurement and adaptive content generation to optimize influence in contexts like persuasion, marketing, and personalized user interactions.
- IMAGINE emphasizes ethical considerations, including privacy and fairness, while redefining media effects research with dynamic, time-sequenced methodologies.
IMAGINE, the Integrated Model of Artificial Intelligence–Mediated Communication Effects, is a conceptual architecture for theorizing communication in a scenario where artificial agents continuously measure people’s responses to media, generate or modify content in real time, and optimize that content toward specified goals of influence. The model joins three artificial agents in a synchronous closed loop—AA-creator, AA-receptor, and AA-negotiator—and is proposed as a framework for rethinking media evolution and media effects under conditions in which content production can be done without human intervention and governed by the controlled reactions of the individual to media exposure (Guerrero-Sole, 2022).
1. Definition, scope, and rationale
IMAGINE is defined as an integrated model of artificial intelligence–mediated communication effects. Its purpose is to help scholars theorize and design research for “continuous real-time connection between AI measurement of people's responses to media, and the AI creation of content, with the objective of optimizing and maximizing the processes of influence” (Guerrero-Sole, 2022). The proposal is situated in the broader context of AI transforming creativity and communication, including text-to-image and text-to-video systems, affect detection from facial expressions, and brain–computer interfaces.
The framework is explicitly future-oriented. It does not claim that fully realized systems of this kind are already standard media infrastructure. Rather, it argues that theory should adapt now to a hypothetical scenario in which media content is created dynamically, in real time, and in direct relation to measured emotional, cognitive, and behavioral responses. This emphasis makes IMAGINE not a platform specification or engineering blueprint, but a conceptual model for communication theory.
Its central concern is optimization of influence. In the model, goals may be set by a third party, such as persuasion, marketing, or health communication, or by the individual, as in accessibility, rehabilitation, or uses-and-gratifications contexts. This dual possibility is structurally important because it places persuasive communication, personalized media, and user-driven assistive systems inside the same architecture.
2. Core architecture and assumptions
IMAGINE consists of three artificial agents operating in a closed loop (Guerrero-Sole, 2022). The AA-creator generates or modifies media content in real time. The AA-receptor measures emotional and cognitive responses to content in real time, using methods such as computer vision, facial expression analysis, psychophysiological measures, and potentially noninvasive brain decoding. The AA-negotiator mediates between creation and reception by dealing with “the goals that are considered to be achieved on the receiver’s mind,” integrating goals, context, and external data to decide what content transformation should follow.
The model rests on several foundational assumptions. It assumes continuous measurement of emotional, cognitive, and behavioral responses; AI-driven content generation across text, image, video, and sound; explicit optimization toward desired outcomes such as attention, belief or attitude change, or behavior; bidirectionality between reception and emission; real-time synchronicity between sensing and generation; adaptive learning toward convergence; and ephemeral uniqueness, since content is created on the fly from transient response states. The result is a communication process in which each exposure may be unique, ephemeral, and leave no traces.
The architecture is organized into a measurement layer, a content generation layer, a user state representation, an optimization objective, a feedback/control loop, and environmental or contextual factors. The measurement layer transforms raw signals into interpretable user-state variables such as attention, liking, trustworthiness, arousal, and related cognitive or affective states. The generation layer includes neural networks, transformers, GANs, diffusion models, text-to-image/video/sound systems, and reenactment or editing systems that can alter voice, facial cues, or narrative elements. The optimization objective specifies desired levels or thresholds over time. The AA-negotiator then fuses goal definitions with user state and context to steer content updates.
3. Closed-loop process and conceptual formalization
The process described by IMAGINE is iterative and synchronous (Guerrero-Sole, 2022). First, user responses are captured through sensors such as cameras, EEG, MEG, EDR, and possibly fMRI or BCI systems, and are transformed into estimates of emotional and cognitive states. Second, the AA-negotiator receives these state estimates together with current goals, contextual data, external data, and constraints, and decides what content change is most likely to move user states toward desired values. Third, the AA-creator synthesizes or adjusts content in real time, for example by changing visual or sonic features, narrative pacing, facial expressions, or aesthetic style. Fourth, the adapted content is delivered, and the AA-receptor immediately measures the updated response. The loop then continues.
The model distinguishes between convergence and divergence. The process is convergent when changes guided by the AA-negotiator move measured variables toward goal thresholds. It is divergent when they do not, in which case adaptation continues. This makes IMAGINE a theory of time-sequenced adjustment rather than one-shot message design. Reception affects emission continuously, and the output of one stage becomes the input of the next.
The paper introduces three conceptual parameters: ac, the intelligence level of creation; ar, the intelligence level of reception; and an, the intelligence level of negotiation. These variables are not embedded in explicit equations, objective functions, update rules, or response functions. The formalization is qualitative and table-based.
| ac, ar, an | Communication model |
|---|---|
| ac=0, ar=0, an=0 | broadcasting model |
| ac=1, ar=1, an=1 | “ideal cycle of AI-mediated communication” |
The model also refers conceptually to an efficiency measure, with the ideal cycle corresponding to efficiency equal to 1, but it does not supply a mathematical definition. This is significant because it marks the framework as conceptual rather than computationally specified. A plausible implication is that later empirical work would need to operationalize user state, objectives, and control policies in ways the original paper deliberately leaves open.
4. Implications for media evolution and media-effects theory
IMAGINE is proposed as a bridge between theories of media evolution and theories of media effects (Guerrero-Sole, 2022). In relation to media evolution, it highlights a long-term shift from separate, post-exposure measurement instruments such as surveys and interviews toward embedded, pervasive, and transparent forms of measurement inside platforms, including likes, follows, and attention tracking. On this reading, the model crystallizes an “ideal cycle” in which intended effects are maximized and unintended effects minimized.
In relation to media-effects theory, IMAGINE reframes effects research from discrete outcomes and self-report instruments to continuous, time-sequenced psychophysiological dynamics. The paper anticipates that “experiments with discrete variables” may be replaced by continuous transformation of content under real-time modulation of measured variables. This suggests a move away from static message comparisons toward adaptive trajectories of content-response interaction.
The model also extends persuasion logics by adding control over measured response itself, not only over message creation. It therefore positions influence as a closed-loop optimization problem. At the same time, it links this logic to uses-and-gratifications by allowing goals to be set either by institutions or by users themselves. The resulting tension—industry-driven goals versus user-driven goals—is one of the model’s central theoretical stakes.
The paper further suggests an effectiveness gradient. As ac, ar, and an approach 1, the system approaches an “ideal cycle” that resembles the hypodermic needle model in the sense of stronger, more reliable short-term influence on targeted variables. Conversely, lower intelligence levels imply minimal or no measurable effects. It also suggests that adaptive closed-loop content should outperform static messages on short-term goals such as attention, arousal, parasocial interaction, and compliance, especially in immersive or VR contexts and where human-like social cues are available.
5. Illustrative instantiations: parasocial interaction and real-time beautification
The paper develops parasocial interaction and real-time beautification as two concrete instantiations of the model (Guerrero-Sole, 2022). In the parasocial interaction case, the AA-receptor measures variables such as likeability, credibility, trustworthiness, attention, and emotional states through computer vision, psychophysiological signals, and potentially voice analysis. The AA-creator then modifies source characteristics including facial expressions, voice cues, body shape, camera framing, direct address, and 3D sound in order to increase perceived attractiveness, human-likeness, and similarity. The AA-negotiator uses these measurements to tune content dynamically so as to maximize trust and credibility and induce outcomes such as willingness to buy, willingness to vote, or message agreement. The resulting content is highly personalized and ephemeral.
The parasocial interaction example is important because it models how face-to-face cues can be computationally simulated and iteratively adjusted. Synthetic faces, social-trait modeling, and reenactment systems can be used to manipulate cues associated with dominance or trust. The model’s formulation is explicitly bidirectional: outputs measured by the AA-receptor become inputs for the AA-negotiator and, through it, for the AA-creator.
In the real-time beautification case, the AA-receptor tracks responses to face aesthetics, including arousal, pleasantness, attention, and social evaluations such as trustworthiness. The AA-creator applies beautification algorithms, including data-driven facial reshaping, deepfake or face-swap systems, and avatar customization, modifying symmetry, averageness, sexually dimorphic traits, and expressions to optimize perceived attractiveness or credibility. These changes can be applied either to a media figure or to the user’s own self-presentation in real time.
The paper notes that attractiveness has demonstrable behavioral consequences, including learning effectiveness, help-seeking success, gaming performance, purchase intentions, and loyalty. Within IMAGINE, the AA-negotiator monitors whether beautification increases targeted outcomes such as social presence, parasocial interaction intensity, or brand attitudes, and continues adaptation until threshold values are approached. Because these modifications occur during exposure, the message becomes unique and ephemeral for each iteration.
6. Research designs, measurement strategies, and analytical implications
IMAGINE is also presented as a guide for empirical operationalization (Guerrero-Sole, 2022). The paper canvasses closed-loop experiments in which content features are programmatically adjusted based on real-time signals, with adaptive and non-adaptive conditions compared on targeted outcomes. It also points to BCI-mediated interaction studies using noninvasive EEG or MEG, and possibly fMRI in laboratory settings, as well as affective-computing studies that track facial expressions and attention during adaptive parasocial or beautification exposures.
The proposed measurement repertoire includes psychophysiology such as EDR, EEG, and EMG; neuroimaging such as fMRI; computer vision and voice-affect analysis; and behavioral telemetry such as gaze, visual orientation, interaction logs, help-seeking, and choice behavior, including buying or voting proxies. Outcomes are framed in classical media-effects terms—knowledge, beliefs, attitudes, affects, physiological responses, and behaviors—while parasocial indicators include perceived authenticity, liking, similarity, social presence, trustworthiness, credibility, enjoyment, and supportive intentions.
Methodologically, the paper recommends dynamic, time-series, and state-space modeling of continuous response trajectories, aligned with dynamic-process and time-sequenced approaches. It also suggests training models to predict which content feature changes move targeted states toward goals and evaluating them by convergence speed and stability. This implies a major methodological shift: self-reports and discrete statistical tests are characterized as poorly suited to continuous adaptive processes.
The paper also discusses self-effects. When users themselves define the optimization objective—such as BCI-based writing or speaking, or beautified self-presentation—the content they create for themselves may produce measurable changes in their own beliefs, attitudes, and behaviors. This extends the framework beyond persuasion or marketing into accessibility, rehabilitation, and other user-directed domains.
7. Ethical, epistemic, and future issues
IMAGINE places privacy and mental autonomy at the center of its ethical analysis (Guerrero-Sole, 2022). Because the model presumes intimate monitoring of thoughts, emotions, and bodily responses, even noninvasive recording may threaten privacy and autonomy if deployed without robust consent and governance. The optimization of influence also raises risks of manipulation and deception, described in the paper through concerns about “Artificial Intelligence Manipulation Effects.”
Bias and fairness are equally central. Perception and generation systems may encode and amplify attractiveness standards or demographic stereotyping, so algorithmic fairness is treated as a design priority. Transparency is likewise emphasized: if exposures increasingly function as experiments, users should know what is measured, how it is used, and toward what goals. The paper explicitly raises the question of whether AI is serving the user or the media industry and platform economics.
The framework also warns against overpromising. Emotion inference from facial expressions is described as controversial and often unreliable, with context dependency and overpromising presented as major risks. Noninvasive brain decoding is treated as promising but emergent, with significant noise and interpretation challenges. These cautions are not peripheral; they define the epistemic limits of the framework.
For future work, the paper calls for dynamic, continuous, closed-loop research designs; learning systems capable of subtle message modification; stronger ethical frameworks; and further work on how to operationalize goals responsibly, audit AA-negotiator decisions, protect autonomy in immersive environments, and balance industry and user objectives. A plausible implication is that IMAGINE’s long-term value lies less in any single implementation than in establishing a vocabulary for analyzing machine-mediated influence where measurement, generation, and optimization become synchronically linked.
