From Generation to Access: On Artificial Dreaming as an Advanced Form of AI Art

1. Introduction

The evolution of generative artificial intelligence is continuously reshaping the media boundaries of art: from text-to-image generation to multimodal immersive content, AI art has consistently advanced along the path of "enhancing external sensory fidelity and intensifying embodied experience." As brain-computer interface (BCI) technology penetrates from laboratories into application scenarios, "artificial dreaming"—the intervention in sleep processes through technological means to generate controllable immersive dream experiences within consciousness—is gradually moving from science fiction narrative into the domain of serious technological and artistic discussion.

Open-world games represented by Grand Theft Auto (GTA) have demonstrated that subjects have a strong demand for the sense of place—"having been there, having lived in a virtual world." However, screen media inherently possess three limitations: the physical boundary of viewing, the dependence on high-stimulus themes to maintain arousal levels, and the mind-body split between character action and player body; the social and psychological controversies triggered by violent and transgressive themes have also persisted. Artificial dreaming may be regarded as a potential direction for breaking through these limitations; yet existing discussions mostly remain at the level of experiential imagination, neither clarifying the fidelity boundaries of different intervention intensities nor distinguishing between two fundamentally different application paradigms: artistic expression and entertainment experience.

Current research exhibits two evident biases. First, it emphasizes technical implementation while neglecting physiological impact: sleep is assumed by default to be merely the "carrier" of dreaming, ignoring that sleep itself undertakes core life functions such as brain repair, memory consolidation, emotional regulation, and somatic recovery, and systematic assessment of the physiological costs that artificial intervention may incur is lacking. Second, it emphasizes sensory fidelity while neglecting aesthetic logic: it assumes by default that "the more complete the modalities and the higher the fidelity, the better the experience," lacking deep reflection on the dream's own operating mechanisms and aesthetic nature, and failing to distinguish between dream-making paradigms oriented toward different goals.

In fact, the dream system formed over hundreds of millions of years of evolution in the human brain is itself a sophisticated "biologically native generative system": it takes memory fragments as input and internal visual imagery as output, has built-in stability mechanisms, emotional regulation mechanisms, and boundary protection mechanisms, and deeply serves the core restorative functions of sleep. From the perspective of neuroaesthetics, the operating rules of this system simultaneously constitute an artistic design reference and a physiological safety standard for artificial dreaming.

This paper takes a complete REM sleep-to-awakening boundary consciousness experience as its entry point, first deconstructing at the phenomenological level the full chain of dream generation and suppression, then revealing the underlying logic of the dream system from three levels—neuroanatomy, functional mechanisms, and evolutionary function; it then systematically assesses the impact of different intervention levels on sleep restoration functions across four dimensions, clarifying the correspondence between intervention intensity and fidelity; on this basis, it distinguishes between two paradigms—artistic dreaming and ludic dreaming—and infers their layered technical pathways, aesthetic cores, and multiple boundaries. The study extends the neuroaesthetic framework proposed by the author in From Alignment to Evocation [1], expanding the "alignment–evocation" dichotomy from single-image generation to continuous consciousness generation scenarios, and introducing sleep physiology and the "consciousness ecology" dimension, in an attempt to provide a theoretical reference that combines aesthetic value and physiological safety for future immersive AI dream art.

It should be noted that this paper as a whole constitutes a theoretical inference and cross-disciplinary construction based on existing neuroscience literature; some judgments regarding intervention effects, boundaries, and risks have not yet been verified by large-sample clinical experiments, and the relevant conclusions should be regarded as a hypothetical framework rather than definitive conclusions.

2. The Sleep-Boundary Consciousness Sample: Method and Phenomenological Deconstruction

2.1. Sample Nature and Analytical Method

The boundary state of consciousness—the transitional phase between sleep and wakefulness—is a natural window for observing the mechanisms of dream generation and suppression. This paper adopts a complete episode of hypnopompic hallucination during an afternoon nap as its analytical sample. The nature and inferential boundaries of the sample must first be defined: it is a phenomenological self-observation of a single subject, lacking statistical representativeness. This paper does not treat it as evidence for mechanistic conclusions, but only as a natural clue for "posing questions and illustrating mechanisms"; all mechanistic judgments regarding dream generation and sleep function are grounded in published sleep neuroscience literature. The sample can be deconstructed, along a temporal sequence, into five consecutive segments, each exhibiting distinct modal differences and consciousness characteristics.

2.2. Full-Chain Phenomenological Reconstruction

I. Standard REM Silent-Film Stage. In the mid-sleep period, a typical REM dream was entered: visual imagery was clear and complete, presenting a continuous narrative scene, yet throughout there was no internally generated auditory signal; plot progression relied entirely on semantic perception at the level of intention, conforming to the general characteristics of the "dream silent film."

II. Boundary-State Auditory Hallucination Emergence. At the critical moment approaching awakening, a creaking door-hinge sound of approximately 2–3 seconds was suddenly inserted into the dream narrative. Its auditory texture was highly realistic, subjectively indistinguishable from real hearing; the sound did not serve as a gradual foreshadowing of the plot, but appeared in an intrusive manner.

III. Instantaneous Emotional Eruption. The auditory hallucination immediately induced intense panic, subjectively experienced as a "sensation of the soul being extracted," with emotional arousal speed and intensity significantly exceeding visual horror scenes in ordinary dreams.

IV. Somatic Tremor Response. The emotional eruption was accompanied by violent, neural-discharge-like trembling throughout the body, subjectively interpreted as "the body striving to pull the soul back into the flesh," but in fact a hypnic jerk during the sleep-to-wakefulness transition.

V. Post-Hoc Narrative Suture. After full awakening, the brain automatically wove three independent physiological events—the visual dream, the intrusive auditory hallucination, and the somatic tremor, which had occurred in parallel—into a narrative with complete causal logic.

2.3. Formulation of Core Questions

This sample presents three counterintuitive features that constitute the core inquiries of this paper. First, why are standard dreams naturally vision-dominant and generally lacking in high-fidelity internal audition? Is this an evolutionary accident, or does a deep functional constraint exist? Second, why, once audition appears at the boundary state, is its emotional impact far stronger than that of visual imagery? What is its relationship to the sensory-emotional characteristics of the waking state? Third, if artificial technology intervenes in dream generation, what levels of fidelity can be achieved by interventions of different intensities, and to what degree will they affect the restorative functions of sleep itself? Answers to these questions cannot remain at the level of individual experience; they must penetrate the underlying mechanisms of sleep neuroscience: first understanding the operating logic and physiological functions of natural dreams, then assessing the risks and value of artificial intervention.

3. The Biologically Native Generative System: Four Layers of Constraints in Dream Generation

Dream generation during REM sleep is essentially a spontaneous internal generative process within the brain, driven by specific neural pathways and subject to strict functional regulation. This system has, through evolution, formed four mutually nested layers of constraint—the biological prototype that artificial dreaming must reference.

3.1. Anatomical Constraint: The Visual Bias of PGO Waves

The most core electrophysiological marker of REM sleep is the PGO wave (ponto-geniculo-occipital wave), which is the primary driving signal for dream imagery generation [2]. PGO waves originate from cholinergic neurons in the pontine brainstem, and their conduction exhibits high anatomical specificity: they pass sequentially through the thalamic lateral geniculate nucleus (the visual signal relay nucleus), ultimately projecting to the primary visual cortex in the occipital lobe. This pathway anatomically determines that the internal generative signals of REM sleep naturally preferentially perfuse the visual system: wave after wave of PGO waves continuously stimulate the visual thalamus and cortex, directly triggering high-fidelity visual imagery that constitutes the main content of dreams. In contrast, the thalamic medial geniculate nucleus corresponding to the auditory pathway does not possess an equally intense, directionally internal drive from the pons; although the auditory cortex exhibits scattered activation during REM, it lacks a high-intensity perfusion mechanism similar to PGO waves, making it difficult to generate stable, high-fidelity internal auditory imagery [3]. The silent-film property of dreams is, first and foremost, specified by neuroanatomical structure.

3.2. Functional Constraint: The Thalamic Gating "External-Open, Internal-Closed" Strategy

The thalamus is the obligatory relay station for all sensory signals entering the cortex except olfaction, and is the core gate for sleep–wakefulness switching. During sleep, the thalamus executes differentiated gating strategies for different senses [3,4]. For the visual channel, eyelid closure has already physically blocked external light signals; the lateral geniculate nucleus's "external input gate" is completely closed, while the "internal generation gate" is fully open, allowing PGO waves to reach the cortex without interference; the visual system thereby fully shifts to inward operation, with no confusion between internal and external signals. For the auditory channel, evolutionary logic (here "evolutionary logic" refers to functional-level adaptive explanation, not causal proof at the phylogenetic level) determines that audition is the only remote sense that must maintain round-the-clock vigilance during sleep; the medial geniculate nucleus's "external input gate" must remain semi-open, ensuring that real external sounds are transmitted to the cortex at any time; as a corollary, the "internal generation gate" is actively tightened, strictly limiting the projection of internal neural activity to the auditory cortex. The essence of this asymmetric strategy is signal-to-noise ratio optimization: ensuring the clarity of external real danger signals, and avoiding confusion between internal auditory hallucinations and external real sounds that would cause frequent false alarms and make sleep unsustainable.

3.3. Emotional Constraint: The Safety Threshold of the Auditory Short Pathway

The deeper motivation for the functional constraint lies at the emotional level. Audition possesses a subcortical rapid pathway: sound information, after relay through the medial geniculate nucleus, can project directly to the emotional core—the amygdala—without undergoing complex semantic processing in the prefrontal cortex [5]. This short pathway has extremely low latency and extremely high arousal efficiency; it is an emotional shortcut preserved by evolution for auditory vigilance. Precisely because the auditory–amygdala pathway's arousal efficiency is too high, the dream system must strictly suppress internal auditory generation: if standard REM dreams could stably generate high-fidelity auditory imagery, abnormal sounds in dreams would frequently and instantaneously activate the amygdala, causing sympathetic excitation and sharp heart rate increases, repeatedly interrupting sleep so that REM cycles could not be maintained.

This echoes the "auditory–emotion direct pathway" thesis proposed by the author in From Alignment to Evocation [1]: the highly efficient survival warning channel in the waking state must be actively closed in the sleep internal generative system, in exchange for the continuity and stability of the entire generative system's operation. In the sample, a few seconds of door-hinge creaking was sufficient to detonate intense panic—precisely the direct manifestation of sound bypassing the cortical rational channel and reaching the amygdala directly.

3.4. Boundary Constraint: The Leakage Effect During Gate Reconstruction

The pre-awakening auditory hallucination is not an "audio upgrade" of the dream, but a brief leakage of the sleep suppression mechanism during the system-switching period, which in turn verifies the existence of auditory suppression in standard REM. During the transition from REM to wakefulness, the thalamic gating mode undergoes overall reconstruction; in this unstable intermediate state, the auditory "internal generation gate" may experience instantaneous relaxation, allowing internal neural discharges to break through suppression and flood into the auditory cortex, forming highly realistic auditory hallucinations [6]. Its typical features are suddenness, high fidelity, and strong emotionality. This phenomenon reveals a systemic principle: the switching state is the most fragile segment of generative stability, and simultaneously the high-risk segment with the strongest experiential realism. Even the naturally evolved dream system leaks at its boundaries; artificial systems with frequent mode switching have all the more reason to take this as a cautionary lesson.

4. Sleep Restoration Function Baseline and Graded Assessment of Artificial Intervention

4.1. Assessment Baseline: The Four Major Restoration Functions of Sleep

Dreams are not a byproduct of sleep, but a key implementation vehicle for sleep's core restorative functions. REM sleep and dream activity deeply serve four major physiological functions, which constitute the baseline scale for assessing the impact of artificial dreaming.

First, brain neural repair. During sleep, the activity of the brain's glymphatic system is significantly enhanced, clearing metabolic wastes such as β-amyloid accumulated during the day through cerebrospinal fluid circulation and perfusion; during sleep, the interstitial space of brain tissue expands, and the rate of metabolic waste clearance is significantly higher than in the waking state [7,8]. At the same time, synaptic pruning during sleep weakens redundant connections and consolidates important synapses, achieving energy-efficiency optimization of neural networks [9]; the maintenance of both glymphatic perfusion and synaptic homeostasis is highly dependent on the integrity of the sleep cycle [10].

Second, memory consolidation and integration. The REM period is a critical stage for the integration of declarative and procedural memory; the hippocampus and neocortex undergo synchronous oscillations, transferring daytime temporary memories to long-term cortical storage and completing the refinement, reorganization, and association of memories [11]; cued activation during slow-wave sleep can likewise directionally promote declarative memory consolidation [12]. The imagery fragments in dreams are mostly drawn from recent and remote memories—precisely the phenomenological presentation of this reorganization process.

Third, emotional regulation and trauma extinction. During REM, the amygdala is active while prefrontal regulation is relatively weakened; the brain reprocesses daytime emotional experiences in the safe environment of sleep, completing emotional extinction and trauma desensitization [13]; a normal REM sleep architecture is an important physiological basis for maintaining emotional homeostasis.

Fourth, somatic physiological recovery. During sleep, sympathetic nervous activity decreases and parasympathetic activity dominates; heart rate and blood pressure decline, and the body enters a state of repair and energy reserve; during sleep, immune cell activity is enhanced and cytokine secretion increases, allowing immune function to be repaired and improved [14].

4.2. Assessment Framework: Three Levels of Intervention and Four Dimensions

This paper classifies the intervention intensity of artificial dreaming into three levels. Mild intervention centers on subthreshold sensory stimulation, such as weak sounds or odor cue implantation during REM, without altering the basic structure and cycle of sleep; moderate intervention centers on neurofeedback regulation and lucid dream induction, which can guide dream themes and emotions, with slight impact on sleep depth but without disrupting the integrity of the sleep cycle; severe intervention centers on invasive cortical electrical stimulation and deep brain stimulation, directly driving neural pathways to generate imagery and deeply altering sleep's neural activity patterns.

It should be stated in advance: the following graded assessments are all built upon existing sleep physiology research conclusions, combined with logical inference regarding the intervention characteristics of artificial dreaming; head-to-head direct comparison data across different intervention intensities are currently relatively scarce, and in particular, the long-term impact of severe invasive intervention on human sleep lacks large-sample longitudinal research support. The relevant judgments should be regarded as theoretical predictions rather than clinical conclusions. The impact of the three levels of intervention is assessed dimension by dimension below.

4.3. Impact on Brain Neural Repair Function

Mild intervention has no significant negative impact. The timing of subthreshold stimulation is limited to the REM window, with intensity controlled below the arousal threshold, not interrupting the sleep cycle nor altering the alternating structure of NREM and REM; existing evidence has not found that low-intensity auditory or olfactory stimulation affects glymphatic perfusion and synaptic pruning processes [10].

Moderate intervention has slight impact. Lucid dream induction and neurofeedback regulation may slightly elevate cortical arousal levels during REM and alter REM sleep depth, but generally do not disrupt cycle integrity; regular lucid dream experience has not shown clear evidence of sleep quality decline [15]. Only when lucid dream frequency is excessively pursued, leading to increased nocturnal awakenings and sleep fragmentation, does it indirectly reduce neural repair efficiency.

Severe intervention carries significant potential risks. Invasive electrical stimulation directly intervenes in the sleep electrical activity of the brainstem, thalamus, and cortex, potentially disrupting normal sleep cycle architecture, causing shortened REM duration and reduced sleep depth; and since both glymphatic clearance and synaptic pruning presuppose an intact sleep architecture [8,9], cycle disruption will indirectly inhibit metabolic waste clearance and neural network homeostasis optimization. It must be emphasized that this risk judgment is currently based mainly on mechanistic inference and animal experimental evidence; systematic data on the long-term direct impact of invasive stimulation on human glymphatic function remain lacking, and it should not be stated as a clinically confirmed conclusion.

4.4. Impact on Memory Consolidation Function

Mild intervention can positively reinforce memory consolidation. This is currently the only positive dream-making mode that has been relatively fully confirmed. Research shows that presenting odor or sound cues related to learning content during sleep can directionally activate corresponding memory representations, reinforce the consolidation process, and improve post-awakening memory performance [12]; its principle is to use external cues to guide memory reorganization during sleep without disrupting sleep architecture.

Moderate intervention can enhance specific types of memory. In the lucid dream state, individuals can actively rehearse procedural skills such as motor actions and musical instrument performance within the dream, with behavioral effects similar to motor imagery training in the waking state (although the neural mechanisms of the two are not entirely identical) [16]; however, once lucid dreaming causes sleep fragmentation, it instead impairs overall memory consolidation efficiency.

Severe intervention carries risks of memory disruption. Invasive stimulation directly acts on memory-related brain regions such as the hippocampus and medial temporal lobe, potentially interfering with the natural processes of memory encoding, consolidation, and retrieval; existing research on deep brain stimulation also indicates that precisely regulating memory-related circuits is highly difficult, and improper stimulation carries the possibility of inducing memory confusion and false memories [17], with long-term safety data currently almost entirely absent.

4.5. Impact on Emotional Regulation Function

Mild intervention can assist emotional repair. Experimental evidence indicates that the hedonic valence of odor cues presented during REM systematically affects dream emotions: positive odors bring more positive dream emotions, while negative odors bring more negative dream emotions [18]. Accordingly, positive sensory cues can increase the proportion of positively valenced dreams and assist nocturnal emotional extinction. This is also a key dividing line between artificial dreaming and violent games: violent games arouse emotions through strong external stimuli, and prolonged play is accompanied by emotional desensitization, increased aggressive tendencies, and comes at the cost of sleep deprivation; mild dream-making, by contrast, works gently within sleep's natural emotional regulation framework and possesses emotional healing potential.

Moderate intervention is dual-natured. Controllable dream-making can provide a safe channel for emotional catharsis, approximating nocturnal emotional exposure therapy; however, once emotional stimulation spirals out of control and frequently triggers intense fear and anxiety, it instead disrupts emotional regulation function, leading to increased nightmares and aggravated daytime anxiety.

Severe intervention carries the highest risk of emotional overload. When directly regulating emotional nuclei such as the amygdala, the difficulty of finely tuning emotional intensity is far greater than with cortical stimulation; excessive stimulation can trigger nocturnal emotional overload and panic attacks, severely disrupting sleep's emotional regulation function, and its risks must be weighed cautiously against therapeutic benefits.

4.6. Impact on Somatic Physiological Recovery

Mild intervention has no significant physiological impact. Subthreshold stimulation does not activate the sympathetic nervous system, has no significant effect on heart rate, blood pressure, or respiratory rhythm during sleep, and does not interfere with the immune repair process; for somatic recovery, it is essentially equivalent to natural sleep.

Moderate intervention has limited impact but requires intensity control. Lucid dreaming generally does not significantly alter autonomic sleep rhythms; if intense emotional reactions occur frequently in dreams, however, it will activate the hypothalamic–pituitary–adrenal axis, elevate nocturnal cortisol levels, enhance sympathetic excitability, and weaken physiological recovery effects—human cortisol secretion has a strict circadian rhythm and is deeply coupled with sleep architecture [19]; the disturbance of this rhythm by repeated nocturnal awakenings warrants particular vigilance.

Severe intervention causes significant physiological interference. Invasive brain stimulation directly affects the brainstem autonomic nerve centers, interfering with heart rate, blood pressure, and respiratory rhythm, and disrupting the parasympathetic-dominant repair state; whether it weakens nocturnal immune cell activity in the long term remains to be assessed.

4.7. Assessment Summary

Table 1 summarizes the impact of the three levels of intervention on the four major restoration functions. The overall pattern is: the impact of artificial dreaming on sleep restoration functions is highly correlated with intervention intensity. Mild-to-moderate non-invasive dream-making has minimal negative impact on sleep architecture and restoration functions, and possesses positive value in the dimensions of memory consolidation and emotional regulation, making it the direction with the highest safety and the greatest practical feasibility; severe invasive dream-making carries significant potential risks across the dimensions of neural repair, memory consolidation, and somatic recovery, and currently lacks a safety basis for application to healthy populations.

This conclusion also provides a theoretical basis, at the physiological level, for the judgment that "dream-making has fewer negative impacts than violent games"—but this judgment applies only to properly designed mild-to-moderate intervention scenarios and still requires further empirical verification: all-night violent gaming is essentially sleep deprivation, directly destroying all sleep restoration functions; whereas reasonably designed mild-to-moderate dream-making theoretically achieves immersive experience within the sleep process, without depriving sleep, and may even assist emotional and memory functions—but this judgment holds only under the premise of preserving sleep architecture integrity.

5. Artificial Dreaming as an Advanced Form of AI Art

5.1. Paradigm Shift: From "Exogenous Generation" to "Endogenous Generation"

Conventional AI art (images, video, music) belongs to exogenous generation: the model outputs results to external media such as screens and speakers, and the subject receives them through the senses; the work exists independently of the subject's consciousness, and the artistic experience occurs in the interaction between subject and work. Artificial dreaming, by contrast, belongs to endogenous generation: generative signals are directly connected to the consciousness-generation pathway through neural guidance or brain-computer interfaces, constructing immersive experience within the subject's consciousness; it does not output an independent "work," but directly generates aesthetic experience itself—experience and consciousness are inseparable, and the subject cannot easily exit as they would by turning off a screen.

From exogenous to endogenous, the artistic carrier penetrates from external physical media into consciousness itself; this is an essential leap in the media dimension of AI art. The fundamental divide between endogenous and exogenous does not lie in "whether a device is worn," but in whether the generative signal enters experience through the subject's own consciousness-generation pathway—VR remains exogenous, because what it generates is an external world model independent of consciousness; dreaming is endogenous, because what it generates is the operating state of consciousness itself.

5.2. Experiential Upgrade: Breaking Through the Three Boundaries of Screen Media

First, dissolving the ontological boundary of media: from "viewing immersion" to "existential experience." No matter how realistic the imagery, the screen player remains a viewer separated by a window rather than a being within space. Dreaming removes this membrane: the subject is not looking at a world, but "living right inside that world"; spatial experience is directly constructed within consciousness, and the depth of this existential experience is beyond the reach of external media.

Second, liberating the thematic boundary: from high-stimulus dependence to the possibility of everyday aesthetics. External media signals must undergo complete cortical processing, and their emotional conversion rate is relatively low; to maintain arousal levels, open-world games are compelled to rely on high-conflict, high-stimulus themes, forming a path dependence on violence and transgression. Endogenous signals directly activate the memory and emotional systems, and their emotional conversion rate is far higher; everyday scenes such as an afternoon alley, a seaside walk, or the reappearance of an old residence are sufficient to produce deep experience, thereby potentially reducing at a fundamental level the latent social and psychological costs of immersive experience, making everyday aesthetics and healing themes possible.

Third, bridging the mind-body separation boundary: from mind-body split to embodied generation. In screen games, the character runs and fights while the player's body sits sedentary; the split between operation and experience brings sedentary burden and embodied absence. Dreaming occurs in sleep or sleep-like states; the bodily sense in the dream is endogenously generated by the brain and fully synchronized with consciousness. Walking and action in the dream activate the corresponding somatosensory cortex and produce real bodily memories [20], whereby the sense of place acquires a complete embodied anchor.

5.3. Value Positioning: Supplement, Not Replacement

It can be foreseen that artificial dreaming and open-world games are not a relationship of replacement, but complementary media forms. The core appeal of games lies in rules, challenges, competition, and social interaction; dream narratives are nonlinear, lack clear goals and a fair competitive framework, and are difficult to support synchronous multiplayer interaction—these are not the strengths of dreaming. The unique value of dreaming lies in inner exploration, emotional experience, and the construction of a sense of place, filling the domain of endogenous experience that games find difficult to reach. It can achieve deep embodied immersion unattainable by screen media without disrupting basic sleep functions, but it need not aim to replace games.

6. Layered Technical Pathways and Design Constraints

Based on the four layers of constraint in biological dreams and the four major functions of sleep, all dream-making technologies should follow five general principles: vision as the primary modality, restrained use of audition, emotional safety thresholds, smooth system-boundary transitions, and priority protection of sleep architecture. The following discussion proceeds along two pathways: non-invasive and invasive.

6.1. Non-Invasive Pathway: From External Induction to Autonomous Regulation

The non-invasive pathway does not implant devices, but guides dreams through external sensory stimulation, neurofeedback, and cognitive training. It is safe, low in ethical risk, and has minimal impact on sleep function, making it the primary form of realization in the near-to-mid term. It comprises three progressive stages.

• Sensory Implantation Stage: Seed-Injection Dreaming (Mild Intervention). This stage leverages the partial permeability of the senses during sleep. Within the REM window identified by wearable EEG devices, subthreshold signals such as directed audio, olfactory release, and weak tactile stimulation are used to implant "seed" material into the dream; the overall dream narrative is still spontaneously generated by the brain, and the seeds integrate into and alter scenes, props, or plot directions. The key points for sleep protection are: intensity strictly below the arousal threshold, prioritizing low-arousal stimuli such as olfaction and low-frequency sound, avoiding sudden loud sounds, and ensuring that the NREM–REM alternating structure is not disrupted. This stage already possesses preliminary feasibility: the influence of sleep-period cues on dream themes and emotional valence, as well as the directional promotion of memory consolidation, have both been experimentally confirmed [12,18].
• Neurofeedback Stage: Closed-Loop Regulated Dreaming (Moderate Intervention). The core is to achieve "dynamic consciousness alignment": high-precision wearable EEG is used to parse sleep staging, cortical activation patterns, and emotional indicators in real time, and artificial intelligence accordingly dynamically generates guiding stimuli and regulates scene tension and narrative progression, forming a closed loop. The key points for sleep protection are: real-time monitoring of sleep depth to prevent overstimulation-induced awakening; establishing a dynamic threshold for emotional intensity so that stimulation is automatically reduced when amygdala activation is too strong; ensuring cycle integrity and preventing fragmentation. The main difficulty lies in the contradiction between the precision and latency of EEG decoding and the nonlinearity of dream narrative.
• Lucid Dream Programming Stage: Autonomous Sandbox Dreaming (Moderate Intervention). By combining lucid dream training methods such as MILD and WBTB, neurofeedback cues, and pre-sleep narrative guidance, the experiencer maintains lucid awareness within the dream and autonomously controls scenes and rules, embodying both experiencer and creator, similar to the creative mode of a sandbox game. The key points for sleep protection are: controlling the frequency of occurrence to avoid pursuing lucid dreams at the cost of sleep fragmentation; guiding a smooth transition to awakening to avoid intense emotional impact in the boundary state; ensuring sufficient deep sleep time. Systematic reviews show that after standardized training, approximately 20%–30% of people can regularly experience lucid dreams [21], and neurofeedback can further improve induction success rates.

6.2. Invasive Pathway: From Cortical Stimulation to Consciousness Integration

The invasive pathway directly intervenes in neural pathways through implanted brain-computer interfaces. It offers high precision and deep experience, but its technical difficulty, ethical risk, and sleep impact are all significantly higher. It belongs to the long-term theoretical pathway and comprises three levels.

• Cortical Stimulation Level: Artificial PGO Wave-Style Imagery Generation (Severe Intervention). Through microelectrode arrays implanted in the visual cortex, cortical functional columns are stimulated in specific patterns to generate visual imagery ranging from light points and shapes to complete scenes, equivalent to artificially manufacturing PGO waves; it can generate stable and clear visual material, but cannot yet drive a complete narrative. The key points for sleep protection are: preferentially activating the visual pathway and strictly restraining stimulation of the auditory cortex; matching the stimulation rhythm to natural PGO waves and setting a duration limit to avoid excessive occupation of the REM window. Existing invasive brain-computer interfaces have already enabled blind individuals to obtain visual perception of letters and simple shapes, verifying the feasibility of cortical microstimulation for generating visual imagery [22].
• Pathway Regulation Level: Thalamic–Limbic System Integrated Regulation (Severe Intervention). Using deep brain stimulation to regulate thalamic gating states, simulating the "external-blocked, internal-open" mode during sleep, while simultaneously regulating limbic structures such as the amygdala and hippocampus to control emotional curves and memory recall, it can generate continuous narratives and actively switch specific modalities on and off. The key points for sleep protection are: replicating natural gating logic and retaining external alerting channels; using audition only as emotional punctuation rather than background sound; real-time monitoring of autonomic indicators. This level exhibits large individual differences and high regulatory difficulty, with side-effect risks significantly higher than cortical stimulation.
• Consciousness Integration Level: Deep Fusion of Memory–Generation Systems (Severe Intervention). By reading the neural encoding of memories and converting them into dream material, while simultaneously writing the generated experience back into the memory system, it can highly personalize the reproduction of past scenes and form "real" embodied memories. The key points for sleep protection and ethics are: strictly setting memory boundaries to prevent confusion between real memories and dream memories; retaining clear boundaries between dream entry and exit and a reality anchor; prohibiting deep intervention in the hippocampal memory encoding rhythm. This level involves memory modification and consciousness integration, and carries the most profound ethical controversies, and should be strictly limited to therapeutic research.

6.3. Correspondence Between Intervention Intensity and Dream Fidelity

The "fidelity" of artificial dreaming is not a single metric; it can be deconstructed into four core dimensions: visual resolution, spatial continuity, interactive feedback, and modal completeness. Different intervention intensities correspond to different upper limits of fidelity, and there exists a significant gap from the public perception of "an open-world dream like GTA," which must be clarified one by one.

• Mild Intervention: Can Only Achieve "Theme-Guided Dreaming," Far Short of GTA-Level Fidelity

The essence of mild intervention is to implant "sensory seeds" into the native dream, rather than generating a complete dream from scratch. The overall narrative, scene construction, and logical organization of the dream are still spontaneously generated by the brain; external cues serve only to guide themes, implant elements, and regulate emotions. Its visual fidelity and spatial continuity depend entirely on the level of the brain's native dreams, typically exhibiting the natural dream characteristics of fragmentation, jumpiness, and blurriness; it cannot generate stable, high-resolution open-world scenes, nor achieve continuous interactive feedback and complete sound effects.

Existing experiments confirm that olfactory or auditory cues during REM can alter dream thematic tendencies and emotional valence, but cannot control the specific imagery and plot direction of the dream [12,18]. In other words, mild intervention can "influence dreams," but cannot "manufacture dreams," let alone produce GTA-level controllable immersive experience.

2.

Moderate Intervention: Can Achieve "Autonomous Controllable Dreaming," Approaching a "Low-Spec Open-World" Experience, but with Enormous Individual Differences

Through lucid dream training and neurofeedback guidance, individuals can maintain metacognition within the dream and achieve active control over scenes, characters, and actions, enhancing spatial continuity and interactivity to a certain degree. Some skilled lucid dreamers can construct relatively continuous open scenes and achieve basic experiences such as free walking and object interaction, similar to the immersion level of early 3D games.

However, this fidelity exhibits extremely strong individual differences: the vast majority of people cannot stably maintain high-definition scenes and continuous physical rules, and the dream still retains its native characteristics of blurriness and jumpiness; at the same time, stable multimodal full sound effects cannot be achieved, and audition remains scattered and sudden. The experiential ceiling of this level is far below the sensory standards of modern open-world games such as GTA, and it is highly dependent on the individual user's ability, lacking the basis for standardized mass production.

3.

Severe Intervention: Theoretically Can Approach "High-Fidelity Generative Dreaming," but Current Technology Is Far from Achieving It, and the Sleep Cost Is Extremely High

Through high-density microelectrode arrays directly stimulating the visual cortex, high-resolution visual imagery can theoretically be generated; combined with limbic system regulation, continuous narratives and stable emotions can be constructed, progressively approaching GTA-level visual fidelity and spatial interactivity. However, to achieve a complete open-world experience, precise regulation of the hippocampus's spatial representation, the somatosensory cortex's tactile feedback, and the auditory cortex's environmental sound effects is also required, and its technical complexity rises exponentially, far exceeding the capability boundaries of current invasive brain-computer interfaces.

More critically, the high-intensity, multi-brain-region stimulation required to achieve this fidelity would deeply interfere with the natural sleep cycle and neural repair functions; it is essentially trading sleep health for immersive experience, and lacks rationality for daily application to healthy populations.

Summary: In brief, if the "high-fidelity, fully interactive, open-world" standard of GTA-like open-world games is taken as the benchmark, mild intervention cannot achieve it at all, moderate intervention can barely reach a low-spec experience with enormous individual differences, and only severe invasive intervention theoretically holds the possibility of approaching it—but at the cost of significant sleep health. This also means that ludic dreaming that pursues extreme realism is inherently in conflict with the principle of "protecting sleep restoration functions"; mild-to-moderate intervention is better suited to an aesthetic-experience route rather than a high-fidelity entertainment route.

6.4. Pathway Comparison and Realistic Positioning

Table 2 provides a horizontal comparison of the two pathways across six stages. The non-invasive pathway trades partial guidance authority for safety, and its experience can be summarized as "the brain as chief creator, AI as accompanist"; the invasive pathway trades safety and ethical risk for control, and its experience is "system-driven, brain-responsive." In the foreseeable future, the realistic development focus of artificial dreaming should be firmly placed on the non-invasive pathway; the legitimacy of the invasive pathway lies primarily in medical treatment scenarios (such as clinical intervention for nightmare disorders and post-traumatic stress disorder), rather than entertainment applications for healthy populations.

7. Two Paradigms: The Divergence Between Artistic Dreaming and Ludic Dreaming

Artificial dreaming is not a single form. Based on differences in core objectives, it can be differentiated into at least two fundamentally distinct types: artistic dreaming, centered on aesthetic expression and inner exploration, and ludic dreaming, centered on entertainment immersion and interactive experience. The two differ entirely in value orientation, technical pathway, intervention intensity, and evaluation criteria, and should not be subjected to the same aesthetic and ethical yardstick. The "dreamness" aesthetic discussed below may apply only to the artistic dreaming paradigm, and not to ludic dreaming's pursuit of extreme realism.

7.1. Artistic Dreaming: An Aesthetic Paradigm Centered on "Dreamness"

Artistic dreaming takes aesthetic experience, emotional expression, and inner exploration as its core objectives. What it pursues is not "realism," but "dreamness"—a unique aesthetic material constituted jointly by blurriness, jumpiness, modal imbalance, and instantaneous dissipation.

7.1.1. The Four Dimensions of Dreamness

First, blurriness: details in dreams are forever hazy; the openness left by blurriness is precisely the space for imagination to unfold, rather than a defect awaiting repair. Second, jumpiness: dream narrative follows emotional association rather than waking causality; scene transitions require no segue, and this jumpiness produces a poetry unattainable in the waking state.Third, modal imbalance: vision is clear while audition is suppressed; the imbalance instead makes visual experience deeper and more cohesive. Fourth, instantaneous dissipation: dreams vanish upon awakening, cannot be possessed, cannot be replayed; this impermanence itself constitutes the fundamental appeal of dreams.

Only the superposition of all four constitutes the aesthetic essence of dreams; aiming at full realistic replication is precisely the eradication of dreamness. If further stratified, modal imbalance belongs to the perceptual layer, blurriness and jumpiness to the structural layer, and instantaneous dissipation to the temporal layer—together the three layers constitute the complete aesthetic coordinate of "dreamness."

7.1.2. The Aesthetic Coordinate of Dreamness: Resonance with Yūgen

Dreamness shares a deep resonance with "Yūgen" in classical Japanese aesthetics. The beauty of Yūgen lies between light and dark, between clarity and obscurity, pursuing lingering sentiment and aftertaste—beauty that leaves a resonance after the scene has faded [23]. This provides a clear aesthetic coordinate for dream-making design: the design direction should not be high-definition realism, but the Yūgen-style restraint, blank-leaving, and sense of evanescence; the technical goal is not to make the dream clearer, but to make the dream deeper. This concept is also consistent with the position of "evocation over replication" in From Alignment to Evocation: aesthetic value comes from the active completion by the subject's consciousness, not from the saturated infusion of external signals [1].

7.1.3. Reverse Prescriptions for Technical Design

Dreamness directly prescribes three technical constraints. First, do not complete missing modalities: the silent-film property is the source of dreamness rather than a technical defect; audition should be used with restraint. Second, preserve blurriness and jumpiness: do not aim at high-definition rendering and logical self-consistency; preserve generative space for the subject.Third, design a gentle dissipation process: exiting the dream should fade as natural awakening does, rather than being abruptly cut off; "dissipation" itself is a constituent part of the experience.

Artistic dreaming naturally adapts to mild-to-moderate non-invasive intervention, achieving aesthetic experience without disrupting sleep architecture, and is the form of artificial dreaming this paper considers to have greater developmental value.

7.2. Ludic Dreaming: An Entertainment Paradigm Centered on "Realism"

Ludic dreaming takes entertainment immersion, interactive experience, and open-world exploration as its core objectives. What it pursues is neither "dreamness" nor "reality" itself, but a kind of "playability" within consciousness—an immersive state in which one retains freedom of action, predictable rules, and explorable space within a dream. Its aesthetic core can be summarized as "controllable surprise": rules stable enough to support action, yet retaining sufficient randomness to sustain exploratory desire. Its demand for "realism" is higher—high-resolution visuals, continuous space, physical rules, multimodal feedback, controllable interaction—closely matching the user needs of GTA-like open-world games. The core demand of this paradigm is "playable and explorable like the real world"; the previously discussed "dreamness" aesthetic not only fails to meet its needs, but would actually weaken its entertainment experience.

7.2.1. Core Needs of Ludic Dreaming

Corresponding to the experiential demands of open-world game players, the core needs of ludic dreaming can be summarized in three points: First, spatial immersion need: the sense of place of "I have lived here, I know this place inside out" requires continuous, explorable, memorable open-world space, rather than fragmented dream scenes. Second, interactive feedback need: actions have corresponding results, objects are interactable, rules are predictable; it possesses a game-like physical system and interactive logic, rather than the illogical rule jumps of dreams. Third, modal completeness need: environmental sound effects, music, voice, tactile feedback, and other multimodal synchronization, rather than the silent-film-style visual single channel.

These needs are essentially the transplantation of the screen game experience framework into the interior of consciousness, pursuing an "intracranial open-world game" rather than the aesthetic experience of native dreams.

7.2.2. Technical Pathway and Intervention Intensity

Ludic dreaming's high demands for fidelity and controllability determine its natural tendency toward stronger intervention intensity:

• Mild intervention cannot satisfy the needs at all, and can only serve as a supplementary embellishment for ludic dreaming;
• Moderate lucid dream mode can achieve basic free exploration and interaction, but with insufficient fidelity, unstable rules, and large individual differences, it can only satisfy light entertainment needs;
• Severe invasive intervention is the ultimate technical pathway for ludic dreaming; only by directly driving multi-brain-region collaborative generation is it possible to achieve a stable, high-fidelity open-world experience.

This also means that ludic dreaming inherently faces higher sleep health risks and ethical controversies, and its technical development and application boundaries must be more cautious than those of artistic dreaming.

7.2.3. Essential Differences from Screen Games

Even if ludic dreaming pursues a GTA-like experience, it still differs essentially from screen games: First, different depth of embodiment: ludic dreaming is endogenous embodied experience; the bodily sense is endogenously generated by the brain, far stronger than the viewing-style immersion before a screen, and the depth of sense-of-place construction far exceeds that of screen games. Second, different consumption logic: screen games come at the cost of sedentary behavior and sleep deprivation; ludic dreaming occurs within the sleep process and theoretically can achieve "experiencing while sleeping," but high-intensity intervention would offset the sleep benefits. Third, different nature of risk: the risks of screen games are mainly sedentary physiological effects and content-related psychological effects; the risks of ludic dreaming penetrate to the neural level, including sleep architecture disruption, reality dissociation, memory confusion, and other deep risks.

7.3. Paradigm Divergence and Value Complementarity

Artistic dreaming and ludic dreaming are not opposed to each other, but two different pathways differentiated from artificial dreaming technology, corresponding to different needs and scenarios.

• Artistic dreaming turns inward, exploring the deep aesthetic value of consciousness and emotion; it is lightweight, safe, and suitable for daily application—the more universally applicable form of artificial dreaming;
• Ludic dreaming turns outward, pursuing extreme immersion and interactive experience; it is heavy, high-risk, and suitable for entertainment applications in specific scenarios—the more imaginative frontier exploration of artificial dreaming.

The two should not negate each other, nor can they be evaluated by the same standard: "dreamness" cannot be used to demand the entertainment quality of ludic dreaming, nor can "fidelity" be used to deny the aesthetic value of artistic dreaming. The core position of this paper—"prioritizing the protection of sleep function and restraining strong-stimulus modalities"—is mainly directed at universally applicable artistic dreaming; for frontier exploration forms such as ludic dreaming, what matters more is to establish clear risk disclosure and ethical boundaries, rather than simply denying their value.

8. Consciousness Ecology: The Four Boundaries of Artificial Dreaming

Technological optimism assumes by default that "the more real, the better; the more immersive, the better." But the evolutionary logic of biological dreams and the physiological functions of sleep suggest that artificial dreaming has four insurmountable boundaries, ultimately pointing to a consciousness-ecology problem that transcends technology itself.

8.1. Neural Boundary: The Firewall of Reality Sense

One of the core functions of thalamic gating is to establish and maintain the cognitive distinction between "internal generation" and "external reality"—the foundation of self-awareness and reality sense. Biological dreams actively preserve modal differences from reality (silence versus sound, blurriness versus clarity); this is itself a protective design. If artificial dreaming completely replicates the signal characteristics of real senses, long-term repeated stimulation may weaken the thalamus's signal-distinguishing capacity, increasing the risk of reality dissociation and depersonalization [24]. Erasing the modal differences between dreams and reality is equivalent to dismantling consciousness's natural firewall.

8.2. Aesthetic Boundary: "Dreamness" Comes from Imperfection

As discussed in Chapter 7, the artistic appeal of dreams stems from imperfection: blurry details, jumpy narrative, silent visual tension, and the quality of dissipation upon awakening. Full-sensory high-definition, surround-sound, logically airtight dream-making would only become an "intracranial movie," losing the dream's unique metaphoricity and hazy beauty. The aesthetic boundary and the neural boundary converge here: protecting dreamness is simultaneously protecting reality sense.

8.3. Health Boundary: Must Not Sacrifice Sleep Restoration

Sleep is a core physiological process sustaining life, and its restorative function has priority over entertainment and artistic experience. Three health bottom lines must be placed in the foreground: first, prioritize the development of mild-to-moderate non-invasive technologies, and remain cautious toward severe invasive technologies; second, no dream-making technology may come at the cost of disrupting sleep cycle architecture or sacrificing sleep restoration functions; third, establish clear safety thresholds and exit mechanisms, immediately terminating intervention once impaired sleep quality is detected.

For the artistic dreaming paradigm, the safety boundary of mild-to-moderate intervention should be strictly observed; for the ludic dreaming paradigm, experimental applications must be clearly distinguished from daily applications, and severe invasive intervention for the purpose of daily entertainment should be prohibited.

Beyond physiological health, the psychological impact of long-term artificial dreaming on self-identity, reality perception, and emotional patterns must also not be overlooked; a corresponding psychological assessment paradigm must be established before commercial application; because ludic dreaming involves higher intervention intensity, its psychological risks are more prominent, requiring stricter long-term monitoring standards.

8.4. Ethical Boundary: Exitability and the Right of Offline Consciousness

With conventional AI art's harmful output, users can close their eyes and turn off the screen, possessing absolute exit rights; artificial dreaming acts directly on consciousness, and a subject in an immersive state, once encountering system failure or malicious injection, lacks effective means of exit. Therefore, three ethical principles should become preconditions: first, restraint—not pursuing extreme sensory and emotional stimulation; second, exitability—retaining a safe exit channel at any depth of immersion; third, informed consent—the subject fully knowing the design framework and potential risks of the experience.

Going further, in the natural state, sleep is the only period each day when a person need not respond to the world. If dreams are also brought within the scope of technological intervention, humans will lose their last offline time of consciousness. The ultimate ethical question of artificial dreaming is therefore not "will it harm people," but: should humans retain a kind of consciousness time that no technology can access? This paper calls it the right of offline consciousness—it is not a specific safety indicator, but a reserve in the sense of consciousness ecology: the consciousness system needs offline periods that are not generated, not recorded, and not guided, in order to maintain its capacity for self-renewal and internal-external distinction. The inalienability of the right of offline consciousness does not stem from individual preference, but from the structural needs of the consciousness system: once consciousness is continuously guided, generated, and recorded, its capacity for self-renewal and internal-external distinction will systematically degrade—and this is not a right that can be voluntarily relinquished.

8.5. Research Limitations

This paper has six limitations.

First, the core sample is a phenomenological self-observation of a single subject, used only for posing questions and illustrating mechanisms, and does not constitute statistical evidence; the generalizability of each segment of the sample awaits testing by large-sample sleep laboratory research.

Second, the graded assessment in Chapter 4 is a theoretical inference based on existing sleep physiology literature; head-to-head empirical comparisons of the boundaries of the three levels of intervention are currently lacking, and in particular, data on the long-term impact of severe invasive intervention on human sleep function are almost entirely absent. The relevant risk judgments in the text are all expressed as "potential risks" rather than "confirmed harms." Overall, this paper's discussion of the physiological impact of artificial dreaming is a framework hypothesis and still requires verification by subsequent controlled experiments.

Third, "dreamness" and "consciousness ecology" are theoretical concepts proposed in this paper, and their explanatory power and design value await testing by subsequent empirical research and design practice.

Fourth, scope limitation: this paper primarily discusses the pathways and boundaries of artificial dreaming from the levels of neurophysiology, sleep function, and aesthetic mechanisms, and does not yet deeply address the cross-cultural symbolic connotations of dreams, deep psychological dynamics, the collective unconscious, and other humanistic and clinical-psychological dimensions; The psychological effects, cultural significance, and mechanisms of action of dream-related practices—such as dream divination, dream interpretation, and dream therapy—require dedicated discussion in separate papers. These practices have long histories in various cultural traditions.

Fifth, application boundary reminder: the current discussion is entirely at the theoretical research stage. If artificial dreaming enters the stage of commercial practical application in the future, focusing only on technical experience and physiological safety will be insufficient; a dual monitoring and assessment system for both physiological and psychological health must be established, long-term tracking of intervention effects on emotional patterns, self-cognition, reality sense, and social function, and cautious assessment of cumulative effects and potential risks—only this constitutes a responsible attitude toward technological application for users.

Sixth, this paper focuses its argumentation on the artistic dreaming paradigm, and its discussion of the technical details, engineering pathways, and risk quantification of ludic dreaming is relatively general; the relevant content awaits further deepening by subsequent dedicated research. In addition, differences in how different cultural traditions position dreams—such as the association with "dream omens" in traditional Chinese culture, the aesthetics of "dream fantasy" in Japanese culture, and the framework of the "unconscious" in Western culture—will profoundly influence the design orientation and ethical boundaries of artificial dreaming, and this dimension also awaits supplementation by cross-cultural research.

Pointing out these limitations is precisely to keep the theoretical claims within the range that the evidence can support, and also to provide certain boundaries and directions for subsequent research.

9. Conclusions

Artificial dreaming is not a technological replication of reality, but an extension of humanity's own consciousness-generative capacity through artificial intelligence. As an advanced form of AI art, its reasonable goal is not to manufacture a "more realistic hallucination," but to create endogenous experience that conforms to neural safety principles while possessing unique aesthetic value, on the basis of understanding dream mechanisms and respecting sleep's physiological functions.

This paper takes a single sleep-boundary consciousness event as its entry point, revealing four layers of constraint in the human dream system—anatomical visual bias, functional auditory suppression, emotional safety thresholds, and boundary leakage effects; these evolutionarily formed rules are not obstacles that technology needs to break through, but design blueprints that artificial dreaming should follow. On this basis, the paper systematically assesses the impact of three levels of intervention on sleep's four major restoration functions, clarifies the correspondence between intervention intensity and dream fidelity, and delineates the realistic boundaries: mild intervention can only guide dream themes, while severe intervention theoretically may approach a high-fidelity open-world experience.

At the paradigm level, the paper distinguishes two pathways—artistic dreaming and ludic dreaming: the former centers on "dreamness," follows a mild-to-moderate non-invasive route, balances aesthetic value and sleep health, and is the most universally applicable form of artificial dreaming; the latter targets "realism," relies on severe invasive technology, pursues extreme immersive experience, and is the frontier exploration direction of artificial dreaming, accompanied by higher health and ethical risks. The two are not opposed, but complementary paradigms corresponding to different needs.

From the single-image AI art creation discussed in From Alignment to Evocation to the continuous consciousness generation explored in this paper, the core proposition of neuroaesthetics remains valid: the strongest evocation channel often requires the strictest restraint; the stable operation of the system is always more important than single-point impact. Brain-computer interfaces will eventually push artificial dreaming into reality, and the faster technology advances, the more necessary it is to hold the bottom lines given by biological evolution—the ecology of consciousness, the essence of aesthetics, the foundation of health, and the measure of ethics. Only within these boundaries can artificial dreaming truly become a new art form that expands human experience.