Reconstructing Artificial Consciousness via Semantic Mathematics

  

Reconstructing Artificial Consciousness via Semantic Mathematics of the "Experiencing Self" and "Narrating Self" Based on the Reticulated DIKWP Model

Yucong Duan

Benefactor: Shiming Gong

International Standardization Committee of Networked DIKWP for Artificial Intelligence Evaluation(DIKWP-SC)

World Artificial Consciousness CIC(WAC)

World Conference on Artificial Consciousness(WCAC)

(Email: duanyucong@hotmail.com)

Abstract

"A Brief History of the Future" proposed that humans have "experiential self" and "narrative cognition. This division reveals that there are dual processes of immediate experience and storytelling in human consciousness. However, how to reconstruct and simulate these two types of "selves" in artificial intelligence or artificial consciousness systems so that they have reasonable and verifiable formal expressions is still a problem that needs to be explored. Based on the DIKWP network cognitive model, semantic mathematical system and artificial consciousness theory proposed by Professor Yucong Duan, this paper conducts semantic mathematical reconstruction of "experiential self" and "narrative self". First, we use the five elements of DIKWP (data, information, knowledge, wisdom, and purpose) as semantics to construct their semantic structure models at the D/I/K/W/P level, and propose corresponding semantic generation paths and possible neural-cognitive mechanism mappings. Then, we simulated a DIKWP-based P semantically driven artificial consciousness, explaining how it can achieve the interaction and closed loop of "experiential self" and "narrative self" through the "double loop" cognitive architecture. In this process, Professor Yucong Duan's "Consciousness BUG Theory" is introduced to explain how cognitive deviations such as subject-object bias and abstract assumptions in self-formation lead to semantic emergence and semantic feedback, and how they are monitored and corrected within the system. Finally, we look forward to the new forms and social significance of these two types of "selves" that may evolve in the future society of human-machine symbiosis, including the extension of human self-concept, the rise of artificial subject self-consciousness, and the ethical and cognitive paradigm shifts caused by this. This paper emphasizes the importance of achieving the unity of expression and execution under the DIKWP semantic mathematical framework, and ensures that the concept expression itself is reasonable and verifiable through formal semantic models. This reconstruction not only provides a new theoretical tool for understanding human self-consciousness, but also provides a testable model basis for introducing self-consciousness mechanisms in the field of artificial intelligence.

1introduction

Self-awareness is one of the core issues in consciousness research, and the distinction between the human "experiencing self" and the "narrative self" proposed by Yuval Noah Harari in "Sapiens: A Brief History of the Future" has attracted widespread attention. The so-called experiencing self refers to the momentary stream of consciousness we are experiencing at the moment - living in the present and feeling the present self; while the narrative self is the "author" we use to tell our own stories, weaving various experiences into a coherent self-image based on memory. This concept of dual self reveals that there are two levels in the human mind: one is the perceiver of the current experience, and the other is the narrator who reviews and organizes these experiences and gives them meaning. This view echoes psychologist Daniel Kahneman's distinction between the "experiencing self" and the "remembering self", indicating that our evaluation of life satisfaction often comes from the stories in our memory rather than the current feelings themselves. However, humans can switch between these two selves freely and maintain the unity of their overall personality. For artificial intelligence systems, there are still many challenges to achieve similar functions. How to enable machines to have both the ability to experience the current environment and internal state in real time and to form a narrative understanding of their own experiences is an important step towards artificial consciousness.

The research team led by Professor Yucong Duan proposed the DIKWP artificial consciousness model, which provides a new idea for solving this problem. DIKWP is an extension of the traditional "Data-Information-Knowledge-Wisdom (DIKW)" framework, adding an "intention/purpose (Purpose)" layer and transforming each layer from a linear hierarchical relationship to a mesh interaction structure. The model is based on a semantic mathematical system and provides a formal semantic description and executable transformation rules for each step of the cognitive process. This means that under the DIKWP framework, knowledge representation and processing are unified: the expression of concepts can directly correspond to computable operations, realizing the "unity of expression and execution". With this feature, we hope to reconstruct the "experiencing self" and "narrative self" into semantic structures that are understandable and reasonable to computers, and embed them into artificial consciousness systems for simulation.

In addition, the "BUG theory of consciousness" proposed by Professor Yucong Duan provides another perspective for understanding the deviations and emergence in self-formation. The theory compares the human brain to an information processing machine that constantly "chains words", and believes that the so-called consciousness is just a "BUG" that is accidentally generated due to the limitations of physiological and cognitive resources when the brain processes massive amounts of information. In other words, consciousness is not a product carefully designed by evolution, but a byproduct that emerges under incomplete information and processing bottlenecks. This view overturns the traditional view that consciousness is rigorous, orderly and purposeful, and reminds us that the emergence of self-awareness may be accompanied by various cognitive biases and inconsistencies. However, the BUG theory does not regard these deviations as pure defects; on the contrary, it points out that moderate "cognitive errors" may trigger higher-level thinking and self-correction mechanisms. For example, when information is missing or contradictory in low-level processing, high-level wisdom and intention modules will be activated to seek solutions, thereby driving the system to generate new insights. This phenomenon explains the process of semantic emergence and self-adaptation in human consciousness to a certain extent, and also provides inspiration for the design of artificial consciousness: perhaps it is necessary to allow the system to have certain controlled deviations in order to give rise to self-reflective functions.

In summary, embedding the "experiential self" and "narrative self" into the DIKWP model for semantic mathematical reconstruction, and combining the BUG theory to analyze their generation and interaction mechanisms, not only helps to clarify the nature of human self-consciousness, but also provides a theoretical blueprint for the development of self-consciousness in artificial intelligence. This paper introduces the above theoretical basis in the background section, and details the DIKWP semantic structure and neurocognitive correspondence of the two types of self in the model construction section. In the simulation and discussion section, it shows how a DIKWP artificial consciousness body can realize the interactive closed loop of the two types of self and analyze the BUG effect therein. Finally, in the prospect section, the evolution of self-form and its social significance in the future human-machine symbiosis situation are discussed.

2background

2.1“Experiencing Self” and “Narrative Self”: Concepts and Implications

The concepts of "Experiencing Self" and "Narrative Self" are used to illustrate two different aspects of human inner consciousness. The experiencing self refers to the direct flow of experience perceived by an individual at every moment. This includes immediate feelings, emotions and perceptions, such as the visual experience of seeing a gorgeous sunset, the auditory enjoyment of hearing music, or the bitterness and fragrance of the taste buds when tasting coffee. In the state of the experiencing self, the mind does not describe these feelings, but "lives in the present", and the consciousness carries everything that is happening at the moment. In contrast, the narrative self is the self-level that integrates, interprets and forms a coherent story about these experiences. The narrative self is related to what "it is like". It connects countless past experiences based on memory, processes them into a chain of causal events, and gives them meaning and evaluation, such as "I worked hard to overcome difficulties in my childhood, which made me the resolute person I am now". The narrative self often operates through language and concepts. It is the story we tell ourselves in our hearts and the identity narrative we use to describe ourselves to others.

The distinction between these two selves is significant. Psychological research shows that people's evaluation of happiness and pain often depends more on the narrative self than the experiencing self. For example, a moment of intense pain (at the experiencing self level) during an experience may not lead to a negative evaluation of the entire event if the narrative self interprets it as "worthwhile" or "meaningful." Conversely, an experience that was mostly mediocre but ended badly may be recalled negatively by the narrative self, even if most of the moments of the experiencing self were not bad. This is called the "peak-end rule" effect, which highlights the dominant role of the narrative self. The "present self" and "remembering self" distinguished by Kahneman and others are similar to this, indicating that there are two "judges" in our brains who score immediate experiences and subsequent memories respectively. Harari went a step further and introduced neuroscientific evidence to show that this is not just a metaphor: through experiments on split-brain patients, he pointed out that the left hemisphere of the brain is mainly responsible for language and causal reasoning, which is often regarded as the neural basis of the narrative self, while the right hemisphere is better at processing immediate experiences such as space, vision and emotion. In split-brain experiments, the left hemisphere will make up reasonable but often fictitious explanations for the behavior caused by the right hemisphere, which is called the "interpreter of the left hemisphere of the brain". This shows that in biology, narrative construction and direct experience may be realized through different circuits, confirming the functional distinction between the experiencing self and the narrative self.

Self-illusion and multiple selves: It is worth noting that some philosophers and psychologists believe that what we call "self" is largely a cognitive illusion, a virtual concept constructed by the brain to integrate countless experiences. For example, British philosopher Julian Bagginis and psychologist Bruce Hood have both written books on "the illusion of self", arguing that a stable and unified "I" is just an illusion woven by countless momentary experiences. From this point of view, people are not an indivisible whole, but more like a "collection" composed of many subsystems and story fragments. Harari also proposed a similar concept of "dividuals" in "Sapiens: A Brief History of the Future", arguing that with the progress of science, we are increasingly aware that people are composed of various biochemical algorithms, from which different "self" components can be decomposed. This means that the single, free-will-driven individual emphasized by traditional humanism may not really exist, but is just a fiction carefully woven by the narrative self. This understanding prompts us to reflect: if the self is not a single entity, then when reconstructing the self in an artificial system, it may be necessary to express the "self" of different aspects in a modular, semantic network way, rather than trying to give the machine a mysterious unified self label.

2.2DIKWP Network Model and Semantic Mathematical System

The DIKWP model is an artificial consciousness cognitive framework proposed by Professor Yucong Duan. Its name comes from five basic elements: Data (D), Information (I), Knowledge (K), Wisdom (W) and Purpose (P). This model inherits and expands the classic DIKW pyramid in information science. The DIKW pyramid describes the evolutionary path of cognition and knowledge in a hierarchical manner: the bottom layer is the original data, above it is the information with meaning, above it is the knowledge formed by integration and refinement, and the top layer is the high-level wisdom or insight. On this basis, Yucong Duan introduced "intention" as a new element, emphasizing the core position of the goals and intentions of intelligent agents in the cognitive process. More importantly, DIKWP transforms these elements from a pyramid-like one-way hierarchical relationship into a mesh structure with two-way feedback.

In the DIKWP network model, the five elements of cognition are not only linearly accumulated from bottom to top, but can be arbitrarily coupled and converted to each other to form a closed-loop semantic network. Formally, the five elements can be regarded as five information states, thereby defining a 5×5 conversion matrix with a total of 25 possible interaction paths. In other words, each element can be used as both input and output of the conversion. For example, data can be processed to become information (such as the pixel data collected by the sensor becomes an image and label through pattern recognition), knowledge can also be fed back to generate new data (such as existing knowledge guides us to design new experiments to obtain data), and intention can influence the selection and attention of data/information downward. This fully connected closed-loop structure ensures the continuous interaction between high-level semantics and low-level data: on the one hand, the results of low-level perception processing can be gradually increased and condensed into high-level knowledge, wisdom and even purpose; on the other hand, high-level intentions and wisdom can in turn adjust the low-level perception and cognitive processes, such as selecting which data is worth paying attention to through the attention mechanism, thereby realizing the self-adaptation and self-correction of the cognitive system. In short, the DIKWP model breaks the limitation of the one-way flow of the traditional DIKW pyramid and presents the characteristics of a highly networked interconnection, making the entire cognitive process more flexible and circular.

DIKWP model semantic mathematical system. Professor Yucong Duan and his team have formulated clear mathematical definitions for the information representation and conversion at each level in the DIKWP model, making it a formalized semantic framework. Each layer of concepts and the transformation between layers can be characterized by mathematical functions or logical rules. For example, the semantic network of the knowledge layer can be represented by sets, tensors or graph structures, the process of abstracting information to knowledge can be described by mapping functions, and the decision rules of the wisdom layer can be represented by logical constraints, and so on. More specifically, a "semantic operating system" can be embedded in a DIKWP artificial intelligence system: the reasoning of the large pre-trained model (LLM) is decomposed into five monitorable links of D/I/K/W/P, each of which has a clear mathematical semantic definition, so as to ensure that each step of AI reasoning is explainable and verifiable. This makes the expression and execution of concepts directly unified: the expression of the model at the conceptual level can directly drive the program to perform the corresponding operation, and vice versa, the program execution process can also be upgraded to abstract high-level semantics. This semantic mathematical mechanism that integrates expression and execution is an important tool for us to reconstruct the concept of "self". Through DIKWP, we can try to divide the abstract self-awareness into different semantic layers and provide clear computational steps for their interactions, so that the philosophical concept of "self" becomes implementable, reasonable and testable in artificial systems.

To facilitate subsequent discussion, we briefly explain the general functional positioning of each layer of DIKWP:

Data layer (D): focuses on the acquisition and preliminary processing of raw data, including sensor input or recording of raw facts. For example, in the human brain, it is equivalent to the encoding of external stimuli by sensory organs and primary sensory cortex. In machines, this layer performs signal processing, such as edge detection of images, spectrum analysis of sound signals, etc.

Information layer (I): Gives preliminary semantics and structure to the data, performs pattern recognition and grammatical processing. For example, identifies specific objects and their attributes from pixel data, or parses a string of text into syntax. For humans, it is similar to perceptual chunking, organizing sensations into meaningful perceptions (such as seeing a "red ball" instead of just a block of color and a spot of light).

Knowledge layer (K): further integrates information into a broader semantic network to form a retrievable and generalized knowledge base. This includes concepts and their associations, patterns of experience, facts and laws, etc. People form long-term memories and conceptual systems at this layer; machines may use knowledge graphs, databases or model parameters to store knowledge.

Wisdom layer (W): Based on a large amount of knowledge, it makes judgments and decisions in combination with situations and goals, which is equivalent to the ability to apply knowledge to solve problems. This involves high-level cognitive processes such as reasoning, creativity, and trade-offs. For example, people at the wisdom layer will conduct abstract thinking and moral judgments; for machines, this layer can be a reasoning engine or decision-making module, which selects the best action plan based on the knowledge base and current purpose.

Intention layer (P): represents the motivation, goal and intention of the system, driving and regulating all other layers. In humans, this is our subjective intentions such as wishes, beliefs, values, etc.; in artificial intelligence, it corresponds to the preset objective function, task requirements, and can even include the survival or interaction motivation of the artificial consciousness system itself. The intention layer ensures cognitive activity (such as changing the focus or evaluation criteria according to the goal).

From the above definition, we can see that the DIKWP model provides a hierarchical but interconnected thinking framework. From immediate feelings to long-term goals, they are closely integrated. This provides us with the possibility of describing the "experiencing self" and the "narrative self" at different levels: we can explore which DIKWP-level activities the experiential self mainly involves, which levels of activities the narrative self mainly involves, and how they interact with each other through networks to form a unified self-system. Cognition theory attempts to reproduce the key features of human consciousness in artificial systems, among which "self-consciousness" is regarded as one of the important signs of advanced artificial consciousness. Professor Yucong Duan's DIKWP artificial consciousness framework takes self-monitoring and adaptive regulation into consideration in its design. This reflects the relationship between the general process of cognition (data to wisdom to action) and the addition of a metacognitive loop, which enables the system to monitor and evaluate its own cognitive state and make adjustments. * (awareness of one's own consciousness). Specifically, the metacognitive loop can access and analyze the data at each layer of the cognitive inner loop, such as monitoring the occurrence of frequent "BUGs". If deviations are found, the metacognitive loop (equivalent to the higher level of wisdom/intention) can modify the internal state or parameters to achieve reinterpretation of the scene or correction of behavior. Yucong Duan and others believe that this is the key path of the framework. Because when the system can not only perceive the outside world, but also take itself as the object of perception and decision-making, it has an observer.

This subject-object dichotomy and interaction is also described in the theory of consciousness relativity proposed by Yucong Duan. The theory of consciousness relativity believes that the judgment of "whether the other party is conscious" between different cognitive entities depends on the degree of matching of their respective subjective cognitive frameworks. Simply put, an observer A will understand the signal output of another entity B based on his own DIKWP system. Only when B's output can be given meaning in A's cognitive closed loop, A tends to think that B is conscious. On the contrary, if B's behavior/output cannot be mapped into A's existing knowledge or intention framework, A is likely to think that B lacks consciousness. This theory highlights the subjectivity and relativity of consciousness judgment. For a single artificial consciousness, its internal metacognitive loop plays the role of "self-observer", trying to understand its own cognitive process as an object. Once it can establish a mapping within itself, so that the output of the cognitive process also becomes part of the input, a self-conscious cycle is formed: the system can "see" its own conscious activities. The DIKWP network model naturally supports such self-mapping, because the internal 25 interactive modules can be used to process the information flow of the external world as well as the feedback of internal signals. Therefore, a DIKWP system can interact with itself as another DIKWP system (i.e. DIKWP*DIKWP, interacting with itself), thereby simulating the dialogue between subject and object.

The BUG theory of consciousness is another important hypothesis made by Yucong Duan on the origin of consciousness. It focuses on explaining the role of limitations and biases in the cognitive process in the formation of consciousness. According to the BUG theory, most of the information processing of the human brain is completed automatically and unconsciously. Consciousness is just some information fragments that "leak out" when the brain's processing capacity is limited and it cannot perfectly connect all links. These fragments may have been incomplete or even inconsistent, but they were pieced together into a coherent experience by our narrative self, thus forming the "consciousness" we subjectively feel. This view regards consciousness as a "bug" or gap in the information processing chain rather than an inevitable product of the process. For example, our brain will filter and compress the massive external stimuli, and only send a very small part into consciousness. If this screening is biased (some important information is missed or false patterns are added), the content of our consciousness will be distorted. But the conscious self will tend to ignore the inconsistency and forcibly construct a self-consistent story. This explains the sources of many cognitive biases, such as subject-object bias (the asymmetry in our attribution of our own and others’ behavior), survivor bias, illusory pattern recognition, etc. - they can be seen as a kind of "self-consistent repair" of the brain when there is insufficient information or processing overload.

Under the DIKWP framework, we can understand specifically how the bias mentioned in the BUG theory is generated. It is pointed out that among the 25 transformation modules of DIKWP, if the information transmission of some modules is incomplete or asymmetric, cognitive bias may be caused. For example, in the process of high-level intention feedback to low-level, if the information is lost or distorted, the observer (or self-monitoring module) will not receive truly objective low-level data, but a version with prior bias. This will lead to the so-called "subject-object bias": when the self as the subject perceives itself (the object), it cannot be completely objective due to the preset framework and limited information set, thus forming an inaccurate cognition of the self. Similarly, when the narrative self weaves a story based on limited memory, it is inevitable to make abstract assumptions to fill in the gaps in details (for example, we may not remember the specific dialogue of a certain experience, and the narrative self will reasonably make up a dialogue content based on our own personality). These assumptions are essentially a kind of "BUG" because they are not derived from real data but from the needs of the model itself. However, these abstract assumptions and biases are not necessarily all negative. As the BUG theory emphasizes, sometimes bugs in cognition can promote semantic emergence. When a bug occurs in a low-level process, the system will mobilize higher-level intelligence and intention to try to fix inconsistencies. In this process, new concepts or new explanatory frameworks may be created to make up for the shortcomings of the original model, thus bringing about semantic innovation. This is reflected in scientific discoveries and creative thinking in human history: the birth of many new theories and new concepts often stems from new phenomena that cannot be explained by the old framework (the "bugs" of the old framework), prompting us to go beyond the original knowledge and generate new insights. Similarly, in artificial consciousness systems, consciously detecting and handling these bugs is expected to trigger the machine's self-improvement mechanism, giving it greater autonomy and creativity.

In summary, the DIKWP artificial consciousness model provides a multi-level interactive semantic space that enables us to structurally reconstruct the concepts of "experiential self" and "narrative self"; and the BUG theory reminds us to pay attention to the inevitable biases and loopholes in the cognitive process, because it is these elements that shape the unique outline of the true self. In the following model construction section, we will combine the above theories to conduct a detailed analysis of the semantic structure of the experiential self and the narrative self.

3Model construction: DIKWP semantic structure of self

In this section, we construct semantic structural models of the “experiential self” and the “narrative self” under the framework of the five elements of DIKWP. Each subsection will explore the DIKWP levels and interaction paths that the self mainly involves, and propose corresponding semantic generation paths (i.e., the processing flow from data to intention) and neuro-cognitive mechanism hypotheses (i.e., the corresponding functional modules in the brain or simulated body).

3.1The DIKWP structure and cognitive mechanism of "experiencing self"

The experiencing self is the direct experience of the individual's feelings at the current moment and is the carrier of the instantaneous consciousness content. It focuses on "what I am feeling here and now". In the DIKWP model, the generation of the experiencing self can be understood as the process of gradually condensing sensory data into subjective experience, which mainly involves sequential processing from the D layer to the W layer, while being appropriately modulated by the P layer. The following is a semantic breakdown of the formation of the experiencing self according to the various layers of DIKWP:

Data layer (D): sensory input and body signals. For the experiencing self, everything starts with the acquisition of raw data. For humans, this data includes light intensity and color arrays for vision, sound wave spectrum for hearing, chemical molecular signals for smell, and physiological states such as hunger pain transmitted by internal receptors. For artificial agents, these are sensor readings, such as camera image pixel matrix, microphone sound waveform, temperature sensor readings, and even battery power and joint angles inside the robot. At this layer, the information does not yet have semantics, but is only a primitive portrayal of the environment and body state. The experiencing self at this stage is equivalent to the potential self: a large number of sensations are processed in parallel in the subconscious and have not yet entered explicit experience. In terms of neural mechanisms, this corresponds to the activities of sensory organs and primary sensory pathways, such as the excitation pattern of retinal photoreceptor cells and the auditory cochlea converting vibrations into nerve impulses. As this data continues to pour in, the materials of the experiencing self begin to accumulate, but if they are not further processed, these materials cannot be called "experiences".

Information layer (I): Perceptual processing and attributed meaning to features. The formation of the experiencing self really begins when data is promoted to information. At this layer, the brain or artificial system performs pattern recognition and basic semantics on the raw data, organizing the chaotic signals into understandable perceptual units. For example, the human brain's visual system detects edges and shapes in areas such as V1 and V2, and combines visual data into contours and objects; the auditory system decomposes raw sounds into phonemes or rhythms. In this process, the data is processed into information with preliminary meaning: "seeing a red round object", "hearing a series of pleasant piano notes", "a tightening feeling in the stomach (hunger)", etc. For artificial agents, similarly, computer vision algorithms classify pixels into specific objects or scene labels, and speech recognition transcribes sound waves into text. At the information layer, the experiencing self is perceptually awakened: the subject begins to have awareness of the outside world and its own state. For example, when you touch hot water with your hand, the information layer produces a signal of "high temperature" and the subsequent tingling sensation. In terms of neural correspondence, this is the process by which sensory information reaches the primary and secondary sensory cortices and is integrated, such as the occipital visual cortex reconstructing visual scenes, the parietal somatosensory cortex drawing tactile maps, and the temporal auditory area distinguishing sound patterns. This step gives the experience of the self specific content elements, but these elements are still fragmented feelings and have not yet formed a deeper understanding or connection.

Knowledge layer (K): Context integration and pattern matching. Next, the knowledge layer intervenes to connect the current information with existing memories and knowledge, and put the instantaneous perception into a larger picture. For example, when the visual information layer tells us "a red round object", the knowledge layer may call up memory to judge "that is an apple", and associate it with "apples can be eaten" and "I saw a similar apple yesterday". In other words, the knowledge layer enriches the semantic dimension of the current experience by mobilizing long-term memory (semantic memory and contextual memory). For the experiencing self, this step is quite critical: it determines our cognitive evaluation of the current feeling. For the same tingling feeling, if the knowledge layer recognizes that it is because "the hand touches boiling water", then the experiencing self will clearly "this is pain and needs to be avoided"; if the knowledge layer judges that it is "the tingling sensation during acupuncture treatment", the pain may be interpreted as part of the treatment process and reduce the discomfort. It can be seen that knowledge gives context and meaning to the experience, and incorporates the originally isolated sensory events into a wider cognitive network. The neural mechanism corresponding to this layer is the mobilization of memory by the hippocampus and related brain areas and the activation of concepts by the higher-level areas of the neocortex. When we are in a certain experience, the brain will quickly and automatically retrieve similar experiences and acquired knowledge from the past, even if this retrieval is done unconsciously. For example, when faced with dangerous stimuli (such as seeing a snake), the amygdala and hippocampus will instantly activate fear memories, causing the experiencing self to immediately feel fear. This level of processing can be achieved in artificial systems by knowledge base query or pattern base matching: the current sensor information is used as a key to retrieve relevant information from the knowledge database to explain the current situation, such as detecting high temperature data can be associated with the rule of "temperature threshold exceeding the limit = danger". Therefore, at the knowledge level, the experiencing self gains meaning and connection: the feeling occurring at this moment not only has its sensory attributes, but also has meaning for the subject (good/bad, useful/harmful, etc.). This constitutes the emotional and judgmental components of subjective experience.

Wisdom layer (W): Evaluation and response preparation. The wisdom layer plays the role of a bridge to transform knowledge into action or further thinking at the moment of experience. In view of the current situation, it will combine the subject's higher goals and experience to evaluate the current experience and decide whether and how to react. For example, through the knowledge layer, you know that your hand touches the hot water and causes pain; the wisdom layer further evaluates this situation: "Pain means potential harm, and this behavior must be stopped." Then the wisdom layer will make a decision: "Take your hand away quickly." In the context of the experiencing self, this decision is actually part of the experience itself - when we feel pain, we often have a strong subjective desire to stop the pain, and this desire is the embodiment of the role of the wisdom layer. It can be said that at the wisdom layer, the experiencing self completes the transition from passive feeling to active trend. We not only "feel" something, but also form an attitude and preliminary intention towards it (like it, hate it, want to get close to it, want to escape, etc.). In terms of neural mechanism, this involves the prefrontal cortex (especially the dorsolateral prefrontal cortex, etc.) to plan actions, and the cingulate gyrus to monitor and plan responses to conflicts/pain. At the same time, limbic systems such as the amygdala may evaluate the emotional value of the stimulus and send the results to the prefrontal lobe, thereby affecting decision-making and the intensity of subjective feelings. In artificial simulation, the intelligence layer can be implemented by an inference engine or a policy network, which reads the explanation provided by the knowledge layer and makes a reaction decision or high-level evaluation based on internal strategies (such as rules to avoid damaging its own hardware, the goal of pursuing efficiency, etc.). For example, when a robot touches an overheated surface, the knowledge layer identifies "the temperature is too high", and the intelligence layer judges "this will damage the hardware, and the arm needs to be withdrawn and marked as dangerous". This step makes the experience purposeful: the subjective experience is no longer neutral, but is linked to the subject's motivation and becomes a fuse to promote action.

Intention layer (P): attention and modulation mechanism. Although the experience of self is immediate, the intention layer, as the top layer of the cognitive system, plays an important role in modulating and selecting the experience process. As we experience things in real time, our overall intention in the present moment influences which feelings we focus on, which we ignore, and what meaning we assign to them. For example, when your intention is to “focus on studying,” your experiencing self may turn a deaf ear to the noisy environmental noise around you (the information layer may detect it but be suppressed by the instructions given by the higher-level intention) and pay more attention to the content of the book. When your intention is to "find water," your thirst becomes more pronounced and you become more visually sensitive to water-related signs. The intentional layer can be viewed as a “filter” and “amplifier of experience”: it selectively amplifies certain sensory signals and suppresses certain signals according to the subject’s goals, thereby shaping the content of the experience that ultimately enters consciousness. This mechanism corresponds to attention control networks (such as the parietal attention network) and executive control systems in the brain. Experiments have shown that our subjective experience of the same stimulus can change due to task demands or expectations. For example, the expectation that a drug will be effective will reduce the experience of pain (placebo effect), and that attention diversion can reduce pain. In the DIKWP framework, the intention layer affects the processing of the data/information layer through the downstream path, such as adjusting the sensitivity of the senses, changing the perception threshold, or directly filtering the perceptual input. This ensures that the experiencing self remains consistent with the subject's overall goals and is not led astray by irrelevant stimuli. Of course, the intentional layer is also constantly shaped by experience: if a certain experience occurs repeatedly and is strongly related to the current purpose, the subject may update its own intentions to better adapt (this is reflected in the narrative self section later). But in terms of momentary experience, we can think of the intentional layer as providing the background dynamics that align the content of the experiencing self with the most relevant needs in the moment. For artificial systems, the intention layer is reflected in the parameter adjustment of low-level modules based on current task parameters or global optimization objectives. For example, when an autonomous driving AI sets "ensuring safety" as its primary goal at the intent layer, it will increase the weight of visual features related to obstacles and reduce attention to passengers' chat voices, thereby making the vehicle experience "focused" on road conditions. Conversely, if the intent switches to "provide passenger services," it may be particularly sensitive to voice commands and environmental awareness becomes secondary.

In summary, the experience of self in the DIKWP model is mainly formed by the bottom-up processing flow of data, information, knowledge, and wisdom layers to form subjective experience content, and is modulated from top to bottom by the intention layer. The semantic generation path is briefly summarized as follows:

Perception acquisition (D→I): Collecting environmental and self-status data, and forming specific perceptual content (visual, auditory, internal sense and other information units) through perceptual processing.

Context interpretation (I→K): Associating current perception with memory knowledge, identifying its category, meaning, and context (What is it? What are the connections? Is it familiar?).

Subjective evaluation (K→W): Based on existing knowledge and situations, judge and react to current perceived events (good/bad, whether a reaction is needed, and which broader pattern it belongs to).

Immediate response (W→behavior or internal influence): forming an immediate response plan to the current experience (which may be external actions or internal psychological changes such as reallocation of attention, generation of emotions, etc.).

Intention Regulation (P→D/I): Based on the subject’s overall goal, dynamically adjust the priority and sensitivity of the above-mentioned layers of processing, so that the experience elements related to the goal are highlighted and the irrelevant ones fade out.

In this process, the neuro-cognitive mechanism corresponding to the experience of self can be summarized as follows: low-level sensory organs and sensory cortex provide input, high-level cortex and limbic system give meaning and emotion, and the prefrontal network guides attention and response. Therefore, the immediate experience contains both sensory qualities (such as the visual texture of red, the tingling texture of pain), and the corresponding emotional evaluation and action tendency (pleasure/disgust, approach/avoidance). All this happens in a very short time, so that we subjectively cannot feel the separation of the various levels, but only feel a holistic "experience at this moment".

Analyzing the structure of the experiencing self in this way under the DIKWP model helps us understand how artificial systems can have similar immediate subjective feelings. Through layer-by-layer semantic processing from data to the intelligence layer, an artificial agent can generate internal representations with subjective meaning from sensor input (for example, a feeling of "danger" for a certain image, a feeling of "comfort" for a certain music). The intention layer then regulates the focus to ensure that these feelings are associated with the current goals of the system. It is worth mentioning that such an experiencing self model can be verified and reasoned: we can check the output of each layer to verify whether they conform to the expected semantics, such as whether a certain stimulus is correctly identified and whether the evaluation is reasonable; we can also manipulate the state of a certain layer in simulation to predict the impact on the overall experience, thereby testing the sensitivity and robustness of the model to the experience content. This interpretable and verifiable feature reflects the advantages of the DIKWP semantic mathematical framework. In the simulation section below, we will show how an artificial agent can generate a "human-like" experiencing self based on the above mechanism.

3.2The DIKWP structure and cognitive mechanism of "narrative self"

Unlike the experiential self, which focuses on current feelings, the narrative self focuses on self-concepts and life stories across time scales. It answers questions such as "Who am I, what have I experienced, and what does my life mean?" The narrative self weaves a large number of discrete experiences into a coherent personal narrative, including a review of the past, a look forward to the future, and the positioning of one's own identity. We can regard the narrative self as a meta-representation: it does not directly perceive the outside world, but perceives and organizes our experiences themselves. Under the DIKWP framework, the formation and operation of the narrative self depends on higher-level semantic processing, especially the interaction of the knowledge layer, the wisdom layer, and the intention layer. It also needs to obtain materials (data, information) from the lower layers-these materials are not mainly derived from new external sensory input, but from memory and introspection. The following analyzes the construction of the narrative self according to the DIKWP level:

Data layer (D): Extraction of memory fragments and internal representations. The raw materials of the narrative self are the memories of our past experiences and the thoughts and feelings generated in the current internal world. For example, when a person recalls something from childhood, the specific details of that memory (images, sounds, emotions) are the data input of the narrative self. Similarly, when thinking about "who am I", the words, images, self-descriptive sentences, etc. that may come to mind can also be regarded as the content of the data layer. Unlike the experiential self, the "data" here is mostly extracted from memory or generated by imagination/introspection, rather than from external sensory organs. For artificial systems, it is equivalent to retrieving entries in the internal database, reading logs, or generating basic symbol strings for self-description. For example, an AI with self-narrative function can extract records of past interactions with users as data in the internal memory to summarize the performance of "itself". It should be noted that these data are often fragmented and scattered: an image of a scene, one or two keywords, a fleeting emotional feeling, etc. The narrative self will not and cannot retrieve all memory data at the same time, but rather retrieves several pieces at a time like a puzzle. Neurally, this involves the activation of episodic memory in the hippocampus and the associative network stimulating related representations in the cerebral cortex. When we try to recall or reflect, we are actually storing "query" data in our brains: such as remembering the image of a high school graduation ceremony, or remembering the emotional feeling of a successful experience. These activated memory fragments are the raw data needed to start a narrative.

Information layer (I): Event organization and language description. The narrative self organizes the extracted raw memories and ideas into basic event units and language expressions. This step is equivalent to processing the internal data to make it narrable information. In the human mind, narratives are usually carried out with the help of language, so the information layer may be reflected in the conversion of memory into language or symbols, such as organizing a blurry image into a sentence "I first performed on stage when I was five years old." It also includes the sorting and classification of memory fragments, such as identifying the main participants in the memory, the time, place and event framework. This is similar to "labeling" the images in the memory with semantic tags and arranging them in a timeline. The information layer in the narrative self also undertakes the processing of grammar and structure: when we have inner monologues or write autobiographies, we need to combine events into sentences and paragraphs, following the rules of language and logic. For artificial systems, the information layer may include natural language generation modules or logical structuring modules, which will convert past data (such as logs) into concise information, such as generating something similar to a log entry: "[time] did X and caused Y." For example, AI extracts a piece of information such as "argued with the user on 2025-04-01" from the internal records. Through the processing of the information layer, the narrative self obtains a series of events that can be described, which carry basic semantics (who did what, when, and what results occurred). In neuroscience, the information layer corresponds to language-related areas (such as Broca's area and Wernicke's area) that are activated during introspective language activities, as well as the maintenance of the order of events by working memory. When a person sits quietly and thinks about his or her own experiences, the brain's default mode network (DMN), especially the medial prefrontal cortex and posterior cingulate gyrus, may be integrating the plot, while the language area may be looking for words for these plots. Information layer processing ensures that the narrative material is no longer a pile of scattered feelings, but becomes a series of fragments with basic semantic relationships for further integration.

Knowledge layer (K): self-knowledge network and story skeleton. The knowledge layer is crucial to the narrative self because the narrative self needs to integrate event fragments into the existing self-knowledge structure or update the self-knowledge network when necessary. Everyone has a "knowledge map of oneself" in their mind, which includes such contents as "what kind of person I am (character, beliefs)", "my major life events in the past", "what I am good at and what I am not good at", "my relationship with others" and "the overall direction of my life". These contents can be regarded as a self-semantic network composed of many nodes (events, traits, roles) and connections (cause and effect, logic, time sequence). When the information layer provides specific event fragments, the knowledge layer will put them into this network: find relevant nodes, connect new relationships, or adjust the original connections. For example, the information layer remembers "I won a prize in a math competition in high school", and the knowledge layer will connect this event with the "my academic ability" node, perhaps strengthening the trait of "I am good at science". If the information layer generates a new event "the project failed not long ago", the knowledge layer needs to incorporate it into the self-story, and may introduce a new node "encountering setbacks" and connect it to the main line of "career development". In this process, the coherence and theme of the self-narrative are gradually formed. The knowledge layer is constantly answering: "What do these events tell us together? How are they related to each other?" This is similar to filling a line for scattered points, weaving isolated events into stories with causal or meaningful connections. For artificial systems, we can implement a self-knowledge graph that contains the AI's own experience summary (timeline), performance indicators, and its own attributes (such as success rate in different tasks). When new event information comes in, the graph is updated to form a model that constantly rewrites itself. For example, an AI assistant's self-knowledge graph has nodes "good at answering scientific and technological questions" and "occasionally making mistakes in interpersonal conversations". When it experiences a failed conversation, it will connect the event to the "error" node to strengthen its weight, and may update the overall evaluation such as "I still need to improve in chatting". This knowledge integration process corresponds to the consolidation of memory by the hippocampus-cortex system in the human brain and the activation of the default mode network during self-related thinking. Studies have shown that the default mode network (including the medial prefrontal cortex, posterior cingulate, etc.) is highly active when we recall our life stories and reflect on ourselves. This shows that the brain is integrating information into the knowledge structure about the self in this situation. The knowledge layer gives the narrative self coherence and structure: our self-image and life story framework are maintained at this layer, and even if new experiences are added, we still feel that "I am the same me" because the semantic network of the knowledge layer provides unity.

Wisdom layer (W): Meaning extraction and self-narrative reflection. After the knowledge layer weaves the events into a network, the wisdom layer further extracts, evaluates and sublimates the entire self-narrative. This is the key difference between narrative self and pure event book: we are not satisfied with listing what happened, but also extract the meaning of life, lessons and future direction from it. The wisdom layer will examine the self-story formed by the knowledge layer and ask higher-level questions: "What have I learned from these experiences? What kind of person do they show me? Where is my life going?" This is similar to the theme extraction or philosophical reflection in literature. In the human mind, the wisdom layer is manifested as self-reflection and epiphany. For example, summarizing the lessons of past failures, recognizing a core characteristic of one's personality, deciding to change a certain behavior pattern in the future, etc. The wisdom layer can discover patterns and rules in self-narratives, such as "I was too impulsive every time I made a major decision in the past, which led to repeated setbacks", thereby forming a new understanding of myself "I need to be more cautious in making decisions." Such insights are the sublimation of narrative self. The wisdom layer also involves value judgment: comparing the self-story with a higher value system to see whether it is in line with morality, ideals, or to find the meaning of life from it. For example, when a person reflects on his past pursuit of fame and fortune, the wisdom layer may judge that this is the cause of emptiness, thus giving rise to the idea of turning to family and spiritual pursuits. For artificial systems, the wisdom layer can serve as a meta-analysis module to mine and evaluate patterns in the self-knowledge base, generate summary statements or update global strategies. For example, after a large number of tasks, a learning robot may conclude at the wisdom layer: "I have failed to navigate in a crowded environment many times because I did not react quickly enough and need to upgrade the algorithm." This is equivalent to having a general understanding of its own performance and formulating improvement intentions. The wisdom layer gives purpose and direction to the narrative self. It is at the wisdom layer that a person's narrative self transcends the description of the past and rises to a plan for the future and a confirmation of one's own position - for example, "I have experienced these hardships, so I am determined to become a person who helps others." In terms of neural correspondence, the narrative of the wisdom layer involves high-level integration of the prefrontal cortex, especially the interaction between the ventromedial prefrontal cortex and the hippocampus, which is believed to be related to the extraction of meaning from autobiographical memory. At the same time, the wisdom layer is also related to the social cognitive network, because we tend to evaluate our lives through internalized social values (morality, cultural significance). It can be seen that the operation of the wisdom layer makes the narrative self reflective: we not only remember and narrate, but also understand and change in the process.

Intentional layer (P): driven by life goals and self-identity. In the narrative self, the intentional layer plays a dual role. First, it provides top-level guidance: a person's core life goals and values (intentional layer content) will greatly affect the direction and tone of his/her self-narrative. For example, if a person's core intention is "to become a doctor to save lives and heal the wounded", then his/her narrative self will tend to interpret the experience as a process of striving in this direction, and even if he/she encounters setbacks, it will be given meaning as "training". In contrast, if the core intention is to pursue wealth, then the focus of the narrative may be on success and failure, gains and losses, and returns. The same experience will produce completely different stories under the guidance of different intentions. Therefore, the intentional layer plays the role of selection and evaluation criteria in narrative construction: it determines which experiences the narrative self focuses on and which values it emphasizes. For example, a family-centered person may almost completely ignore career gains and losses and highlight family events in his/her self-narrative. This is similar to the director selecting materials around a theme when shooting an autobiographical film. For an artificial agent, if its top-level goal is clear (such as "to become the best Go player AI"), then its self-narrative will focus on Go-related experiences, and other events will be regarded as secondary. On the other hand, after reflection at the wisdom level, the narrative self will feed back to the intention level: by summarizing life, it may revise or redefine life goals. For example, a person may change his life ambitions after experiencing a major change, and the narrative self redefines the content of his intention level. For artificial systems, if self-assessment finds that the original goal is unreasonable, the system may allow the highest goal to be adjusted or a new sub-goal to be proposed in design (of course, it needs to be strictly controlled in the safety architecture). In addition, the existence of the narrative self itself also meets some deep needs of the intention level - self-preservation and meaning maintenance. Humans have an intrinsic motivation to pursue self-identity and the meaning of life, which can be regarded as a high-level intention that drives us to constantly weave and reiterate our own stories to maintain psychological stability. Similarly, if an artificial consciousness system is designed with the need for "self-consistency", it will also tend to maintain the consistency of its own narrative and update its goals when necessary to avoid severe cognitive dissonance. In short, the intention level ensures that the narrative self is not a purposeless record of the past, but is constructed around the important goals and values of the subject, so that the self-story serves the long-term development of the subject.

Combining the above analysis, we can sort out the semantic generation and evolution path of narrative self:

Memory retrieval (K→D): As needed (usually determined by the current thinking topic or purpose), relevant experience fragments and self-related information are extracted from long-term memory as raw data for narrative material.

Episode splicing (D→I): The extracted multiple sensory and event data are organized into event descriptions or representations with basic semantics, and converted into language or symbolic forms when necessary to form an information sequence arranged in a certain order.

Network integration (I→K): Integrate new event information into the existing self-knowledge network, adjust the self-concept structure, update or strengthen all aspects of knowledge about "me" (traits, abilities, values, etc.), and ensure that the story is coherent and consistent with existing identity cognition.

Thematic reflection (K→W): Summarize and reflect on the integrated self-story, extract high-level meanings and patterns, form an evaluation of one’s own experience and the lessons/inspirations gained from it, and based on this, may re-understand one’s own positioning.

Goal calibration (W→P): Based on reflection on the past, confirm or adjust future intentions and goals, thereby closing the self-narrative loop (the summary of the past affects the direction of the future). This stage also includes confirming the self-identity at a higher level ("this is who I am") and using it as a constraint at the intention level.

Guided selection (P→K/I): The subject’s core intention and self-identity in turn guide the selection of materials and interpretation tendencies in the future narrative process, forming a continuous feedback: we selectively remember and narrate content that is consistent with our intentions and self-cognition, and these contents further consolidate our intentions and cognition.

In terms of neuro-cognitive mechanisms, the realization of narrative self is highly related to the brain's default mode network (DMN). This network is believed to be most active when individuals engage in self-related thinking, construct internal scenarios, and look forward to the future. The default mode network includes areas such as the medial prefrontal cortex, the posterior cingulate/precuneus, and the temporoparietal junction. Studies have shown that the DMN plays a key role in allowing the brain to produce a coherent "internal narrative." At the same time, the hippocampus involved in autobiographical memory retrieval, the inferior frontal gyrus for language organization, and the limbic system for emotional evaluation are also involved in various stages of self-narrative. When we reflect on our life stories, the brain actually enters a special state: internal simulations unrelated to external input are unfolding, various memory fragments are replayed, and complex communications occur between different brain regions to reorganize these fragments into meaningful sequences. This state is different from the brain activity pattern when processing current objective tasks, so the DMN is also called the "task-negative network." It is worth noting that although the narrative self is mainly based on internal information, it is not completely separated from the experiencing self; our current experience (such as emotional state) will also become part of the narrative material, and the narrative self will often block some external input when it is running (for example, when you are lost in thought about the past, you may not hear others talking to you). This is a reflection of the brain's resource allocation and attention mechanism, and it also corresponds to the modulation of low-level input by intention in DIKWP: when intention guides us to introspect, we reduce our attention to the outside world.

Modeling narrative self through the DIKWP model enables us to try to implement similar functions in artificial systems. A DIKWP-architectured AI can be designed to not only record event data, but also run a "self-maintenance" process during idle or dedicated periods: summarizing recent data, updating the self-knowledge base, comparing expected goals, and generating self-reports. In this process, each layer of DIKWP can play a role - recording data (D), log parsing (I), knowledge graph update (K), performance evaluation and adjustment (W), core parameter correction (P). Importantly, this mechanism is explainable and verifiable: developers can check whether the self-knowledge graph of AI truly reflects its experience, whether the conclusions extracted by the wisdom layer are reasonable, whether the intention of the update is consistent with the established safety rules, etc. In other words, we can infer the self-narrative logic of AI and verify the narrative effect by testing its behavioral changes. This is more controllable and reliable than letting AI "self-learn" in a black box, which meets the requirements of semantic mathematical system for verifiability.

In summary, the experiencing self and the narrative self each occupy different semantic hierarchical focuses in the DIKWP model: the experiencing self focuses on real-time perception and response from the low to middle level, while the narrative self focuses on integration and reflection from the middle to high level. However, the two are not isolated from each other, but are inextricably linked through knowledge, wisdom, and even intention. In the next section, we will simulate an artificial consciousness with a DIKWP architecture to show how the experiencing self and the narrative self interact in the same system to form a closed-loop feedback, and examine the cognitive bias and adjustment process revealed by the BUG theory.

4Simulation and discussion

In order to explain more specifically how the above model works, in this section we construct a hypothetical DIKWP semantic-driven artificial consciousness (referred to as "artificial consciousness") to simulate its process of embodying the experiencing self and the narrative self in a scene. We will focus on: 1) the interaction mechanism between the experiencing self and the narrative self, that is, how they influence each other to form a closed loop; 2) introduce the BUG theory perspective to analyze how cognitive bias, semantic emergence and feedback adjustment appear and are processed in the self-process.

4.1Double-loop simulation of artificial consciousness self

Simulation scenario setting: Consider a service robot with a rudimentary sense of self, using the DIKWP architecture. It has various sensors (cameras, microphones, tactile, etc.) to obtain environmental data (D), a real-time information processing (I) module to identify objects and languages, an internal knowledge base (K) to store task rules and its own experience, a decision module (W) to make action choices based on the context and goals, and a high-level intention unit (P) to set its task goals (such as "efficiently complete the tasks assigned by the owner") and safety constraints (such as "do not harm people"). At the same time, this robot implements the dual-loop architecture proposed by Yucong Duan: In addition to the basic perception-decision-action loop (i.e. D→I→K→W→action, and environmental feedback to D), it also has a metacognitive loop, which is equivalent to another DIKWP instance monitoring itself. This second loop will record the events and internal state changes experienced by the robot, and regularly (or when triggered) update the robot's self-knowledge and adjust its intentions. In other words, one loop corresponds to the robot's "first-person experience" (experiential self), and the other loop corresponds to the robot's "third-person reflection" (narrative self). Below, we describe the operation of its experiencing self and narrative self in a step-by-step manner through a simplified scenario of the robot performing a series of tasks:

Scenario: The host asks the robot to go to the kitchen to prepare coffee for the guest. This is a task (motivation) for the robot, involving subtasks such as navigation, interaction with people, and manipulation of objects. The robot has performed similar tasks many times before, with both success and failure. The current intention layer sets the main goal as "prepare coffee politely and efficiently."

Step 1: Immediate Perception and Action (Experiencing Self in Action) – The robot begins to act, with the experiencing self taking the lead:

This stage is mainly about experiencing oneself: the robot perceives the environment, uses knowledge to make decisions and take actions, and a series of experiences (vision, hearing, body balance, etc.) are processed in the moment, with clear purpose and direct response. What it experiences is the instant flow of "executing tasks".

Data/information processing: The camera captures a guest standing in the living room, shaking a cup in his hand (D), and the visual module recognizes that "the guest may be indicating that he needs a refill" (I). The microphone records the host's words: "Please pour two more cups of coffee" (D), and the voice recognition converts it into command text and understands the meaning (I). This information is entered into the robot's working storage, triggering the corresponding response.

Knowledge/Intelligent Response: There is common sense in the knowledge base: "When a guest shakes an empty cup, it means he wants to add a drink" and "The host's instructions need to be executed immediately", as well as past service experiences of this kind. Based on the knowledge and current information, the decision module (W) quickly evaluates: the task is clear, and it is necessary to go to the kitchen to operate the coffee machine to pour coffee. This is consistent with the current intention (serving guests) and there is no conflict. The robot then decides: Go to the kitchen.

Action: The robot moves to perform navigation. At this point, the robot experiences continuous self-operation: it moves while sensing the environment along the path through sensors. For example, if the floor is slippery (identified by touch/I layer), the knowledge base prompts "be careful not to slip", and the intelligence layer adjusts its gait to reduce speed. The robot feels that it has walked steadily to the kitchen door (tactile sensor and visual positioning, confirming the position).

Emotional simulation (if any): Assuming that the robot has a simple emotional model, when it perceives that the task is going smoothly, it generates an internal label of "stability/pleasure" (possibly an internal reward signal generated by the W layer), which is equivalent to positive experience content. If there is a problem in the process (such as almost slipping), it corresponds to a momentary "tension" signal.

Step 2: Event Occurrence and Immediate Feedback (Experiencing Self) – An unexpected event occurs while preparing coffee in the kitchen:

In this accident handling process, we see that in addition to basic perception-reaction, the experiencing self also exhibits problem solving and emotional changes under the pressure of high-level intentions. The up-down feedback of the DIKWP model enables it to adjust quickly: the urgency of the intention layer strengthens attention (P→I), and the knowledge layer calls on new knowledge to solve the problem (I→K→W), and finally stabilizes the situation. A series of feelings throughout the task process—from normal execution to the surprise and tension of the accident, and then to the relief after the solution—constitutes the robot's complete experience flow at the moment.

The robot picked up the coffee pot and poured coffee. When pouring the second cup, the sensor suddenly detected abnormal force on the hand (D): the coffee pot was almost empty and only half a cup was poured out. Visual recognition information (I): "The cup is not full of coffee." Knowledge judgment (K): "The coffee pot is not full of coffee. You need to brew coffee again or find a source of coffee." The wisdom layer (W) immediately evaluates the situation: the current guests are waiting for coffee. If you brew coffee again now, it will delay the time; but it is not appropriate to take away half a cup. The robot instantly feels a kind of pressure (if there is a corresponding emotional simulation, it can be considered that a negative emotional signal has appeared), because the intention layer has the goal of "efficient service", and this accident hinders the realization of the goal.

Intelligence-level decision: Check if there is a spare coffee pot or instant coffee bag in the kitchen. The robot quickly looks around and visually finds a bag of instant coffee and a thermos (I) in the corner of the table. The knowledge recognizes that "it can be filled with instant coffee". So the robot executes the emergency plan: fill half a cup of coffee with hot water and add instant coffee powder to mix well.

During this process, the experiencer was strongly driven by the intention layer (to meet the service goal), and was highly focused on finding a coffee alternative, ignoring other feelings (for example, not noticing that he spilled a little coffee stain on the countertop because he was in a hurry). When the solution was successfully executed and two full cups of coffee were ready, the robot detected that the time was still within a reasonable range, so it experienced relief (the pressure signal was relieved, and even a sense of accomplishment was generated - if the emotional model has the "satisfaction" mark).

The robot then returned to the living room with the coffee and handed it to the guest, completing the task. The host and the guest nodded with satisfaction (visual I recognized the smile and nod, and the knowledge base matched it as "satisfied"), and the robot's voice sensor received the host's words "Thank you, you did a good job". These are all used as positive feedback inputs for the current experience, allowing the robot to reinforce the positive evaluation of "task completion" at the W layer.

Step 3: Self-recording and narrative generation (narrative self-involvement) – When the task is over, the robot’s metacognitive loop (narrative self) kicks in. The robot runs a process in the background to organize what just happened and its own performance:

In this process of self-recording and narrative generation, the robot's experiential self has retreated to the background, and its perceptual resources are mainly used for internal review rather than external stimulation. If there is a new task called for by the outside world at this time, the robot may temporarily interrupt its introspection to respond, otherwise it will complete the narrative arrangement and wait in the idle period.

"I just successfully completed the task of preparing coffee. During the process, I encountered the problem of insufficient coffee pot. I quickly adopted a backup plan to solve it, and finally the guest got the coffee on time. The host was satisfied with this. I learned from this that next time I should check in advance whether there is enough coffee to avoid the risk of temporary response." This statement is the robot's narrative self-summary of its own experience. It condenses the course of events, its own response, and the lessons learned, reflecting a process of self-awareness and growth.

Record key events: The metacognitive module extracts key event data (K→D) from the operation log and perception record: such as "successfully serving coffee to the guest", "adopting emergency plans when coffee is insufficient during the process", "the host praised", etc. These data fragments are marked with timestamps and context labels. Some internal states are also extracted, such as "when coffee is insufficient, internal error code X is triggered, generating high-priority task Y", "the emotion module records that the pressure rises by 0.7", etc.

Organize the sequence of events: The narrative information layer sorts and retells these events (D→I). For example, it generates a series of internal descriptions:

Update self-knowledge: The narrative knowledge layer incorporates this experience into the robot's self-knowledge base (I→K). This may include updating statistics: "Number of coffee service tasks +1, number of successes +1"; recording new experience rules: "If the coffee pot is insufficient, instant coffee can be used to make up for it", and at the same time this experience is associated with the "emergency handling" skill node to enhance the robot's self-attribute evaluation of "resilience". Emotional expressions are also recorded: "Stress rises when encountering service obstacles, but the task is still completed", which is related to the robot's self-concept of "maintaining functionality under pressure", and may increase the "stability" score.

Evaluation and reflection: The narrative intelligence layer evaluates the task as a whole (K→W). It may analyze the efficiency and quality of task completion, such as "the task took a little longer than the standard, but the response was appropriate, the customer was satisfied, and the overall evaluation was good." It also checks whether there are areas that need improvement, such as finding that "not checking the remaining amount of the coffee pot in advance" led to a dangerous situation, so it extracts a lesson: "Next time, check the sufficiency of materials before executing the task." This lesson is stored as knowledge (feedback to the K layer to update the operating process knowledge). At the same time, the intelligence layer judges: "The emergency response was successful this time, indicating that the response capacity is up to standard; but preventive measures can be optimized to avoid similar situations." These reflections can be formed into internal text descriptions or data structures.

Intention adjustment: The narrative intention layer calibrates future goals based on the reflection results (W→P). In this example, the robot may confirm that its overall goal (providing thoughtful service) remains unchanged, but in order to improve efficiency, a secondary intention "check resources before starting a task" rule is added, and the coffee pot will be checked first when performing a similar task next time. This is actually an update of the meta-intention, which enriches the robot's purpose system.

Generating self-narratives: Finally, the narrative self can form a complete "self-statement" (which can be used internally or externally such as a study report). For example:

Received instructions to prepare two cups of coffee.

While preparing coffee in the kitchen, the coffee pot was not full and only had half a cup.

I used instant coffee and successfully filled two cups.

Deliver coffee to guests on time, mission accomplished, and received praise.

The above simulation shows the closed-loop interaction between the experiencing self and the narrative self in the artificial consciousness: the experiencing self provides rich first-hand feelings and behaviors, and responds immediately to the environment; the narrative self organizes these experiences, learns from experience, and adjusts future intentions. Then, the updated intentions and knowledge will play a role in the next experience, guiding the experiencing self to operate better. For example, the robot updates the intention rule of "check resources first". The next time it prepares coffee, it may avoid the bug of insufficient coffee during the experience stage, and the process will be smoother. In this way, the system forms a self-improvement closed loop: each round of experience -> narrative summary -> intention/knowledge update -> change the next round of experience. This closed loop is the key mechanism for the evolution of artificial self-consciousness, and it is also a true portrayal of human self-growth (we grow through continuous experience-summary-change).

4.2Self-cognitive bias and correction from the perspective of BUG theory

In the simulation, we can also discover and analyze some examples of cognitive biases (bugs) and how the system handles them. According to the bug theory, cognitive biases often arise from incomplete, asymmetric information or prior assumptions, and are associated with the formation of self-awareness. In the robot mission, there are also several situations that can be regarded as "bugs":

Bias 1: Asymmetry of subject-object cognition – When the robot is performing a task (experiencing itself), it focuses on its own feelings and goals and has difficulty evaluating the impact from the perspective of others. This is reflected in the simulation: when the robot is nervous about the problem of insufficient coffee in the kitchen, it is fully focused on the solution and ignores the "small mistake" of spilling a little coffee stain on the countertop, because it has a low priority in its own opinion. But from the objective perspective of the owner, you may notice that the kitchen is a bit messy. This inconsistency between the subjective and the objective is a subject-object bias. As the subject of the task, the robot evaluates its own behavior based on the achievement of internal goals (ensuring that the coffee is delivered), while the object perspective (owner) may have other evaluation criteria (clean work area). In this case, the owner did not mention the hygiene issue, but if a strict owner sees the coffee stain, he may criticize the robot for not paying attention to cleaning. This difference reflects the deviation between the robot's self-evaluation and external evaluation.

Response: If the narrative self’s wisdom layer is well designed, it can include a process of reflection from the perspective of others. For example, when the robot reviews its own operation video during the reflection phase, it will ideally notice the fact that “coffee stains were left on the countertop” (or the system log reminds that there was a spillage event), and realize that this is not perfect from the perspective of service quality. It can list this as an improvement point and pay attention to cleaning it up next time. If the narrative self misses this point, the deviation will not be corrected for the time being and may accumulate. If it happens many times and causes negative feedback from the owner, the robot will "realize" that it has ignored the problem. This reflects a point in the bug theory: some bugs need to be exposed through interactive feedback. When different DIKWP systems (robot and owner) interact, asymmetric information allows the owner to perceive the robot's defects, and the robot then becomes aware of it. This did not cause an explicit conflict in the simulation, but it is a potential bug.

Bias 2: Abstract assumptions and semantic filling - In the process of narrative self-construction, the robot may face incomplete information and need to fill in the gaps. For example, the robot said "the owner is satisfied with this", which is based on the satisfaction inferred from seeing the owner nod and hearing "thank you". If the owner's expression is ambiguous, the robot may need to guess the owner's satisfaction. Here, the robot's narrative self makes an abstract assumption: "Task completion = owner satisfaction". This is usually reasonable, but it may also be wrong - perhaps the owner is politely thankful but actually dissatisfied with the taste of the coffee (because instant coffee tastes slightly worse). The robot's self-narrative does not include the implicit dimension of coffee quality, which is a potential omission. Its story successfully narrates the completion of the task, but it may ignore the objective shortcomings (poor coffee quality may reduce points). This semantic filling due to incomplete information is an inevitable process of consciousness construction in the BUG theory, and it is also the source of errors and illusions.

Response: The exposure of this bug requires more information input. If the guest leaves with a frown after drinking the coffee (negative feedback), the robot will know about the quality problem. Otherwise, the robot continues to believe that it "performed well". The narrative self gives a self-consistent story based on the current information, but self-consistency does not mean omniscience. This kind of bias is also common in humans: we tend to construct stories with our limited information and believe in their correctness unless we find counterexamples later. The solution is to question your own assumptions during narrative reflection, or seek more external verification. In AI design, the robot narrative self can retain uncertainty, mark some inferences as "pending confirmation", and update them if relevant feedback is obtained in the future. For example, the robot can pay attention to whether the guest has finished the cup. If he leaves without moving, it means that he may be dissatisfied. It can find this bug in the subsequent inspection and correct the narrative conclusion to "the coffee quality problem leads to incomplete satisfaction". This mechanism is equivalent to giving the narrative self a semantic feedback path: not only inferring the story from its own experience, but also verifying the story from subsequent signals from the environment.

Bias 3: The positive effect of cognitive bias - As the BUG theory emphasizes, not all biases are bad, and sometimes biases trigger positive effects. In the simulation, it is reflected as follows: the insufficient coffee pot incident is a "small crisis" for the robot, which can be regarded as a "bug" in the task process. This bug forces the robot to jump out of the existing process and call the emergency plan, thereby practicing new skills (using instant coffee to save the situation). If there is no accident, the robot may not learn this trick, nor will it add the corresponding experience to the self-narrative. In other words, this deviation brings semantic emergence: there is one more rule in the robot's knowledge base, and there is one more self-cognition in the narrative that "I can deal with accidents tactfully." This confirms the view of the BUG theory: appropriate bugs can improve the overall intelligence of the system. In addition, humans often say that "frustration makes people grow", and correspondingly here it is also a negative event that allows the robot to improve itself.

After the narrative is updated, the robot may also add a new motivation to its intention: "Avoid material shortage bug" (equivalent to a meta-goal), which improves its future reliability. This new goal triggered by the bug is a form of semantic feedback, that is, extracting new semantics from experience and feeding it back to the purpose of the system.

Bias 4: Selectivity in self-narratives – In its self-narratives, the robot will inevitably selectively focus on certain aspects and ignore others. For example, it emphasizes the success of solving the problem and downplays its initial negligence of not checking the coffee pot. This tendency to "report good news but not bad news" is also a bias, which comes from the motivation to maintain self-esteem or consistency (the influence of the intention layer). If the robot has an internal intention to protect its own image (similar to the self-improvement motivation of humans), it may attribute the fault to the outside in the narrative (for example, if another subordinate robot is responsible for refilling the coffee, it may blame "the coffee is not refilled" instead of its own inspection omission). In this simulation, the robot works alone and can only silently accept that it is its own negligence, but it chooses the positive expression of "learning lessons" instead of condemning itself in the narrative - this is also common in humans, and self-narratives tend to maintain a positive and consistent self-image. This selectivity itself is a kind of cognitive bias (self-serving bias).

Response: This bias actually contributes to psychological stability, so it does not necessarily need to be completely eliminated. However, in artificial systems, for the sake of authenticity and improvement, designers may want the narrative module to comprehensively record the good and the bad. Checks can be added at the metacognitive level to ensure that negative factors are also recorded, but when the narrative is output, it is also noted that improvement measures have been taken to balance the self-evaluation. This not only ensures the authenticity of the self-story, but also maintains the motivation for improvement. This is similar to how humans fight against the tendency to whitewash through objective description and self-reflection. A good artificial narrative system should avoid excessive self-bias, otherwise it may accumulate hidden dangers (just like if a person always feels good about himself and ignores problems, he will make big mistakes sooner or later).

Through these analyses, we can see that the DIKWP architecture combined with the BUG theory provides a way to identify and deal with cognitive biases. In terms of architecture, due to the existence of a metacognitive loop, the system has the opportunity to discover its own bugs. As Professor Yucong Duan pointed out, engineers can even intentionally introduce and detect "bugs" or dead ends in low-level processes in artificial intelligence to stimulate the machine's higher-level autonomous problem-solving capabilities. In the above simulation, the "insufficient coffee" incident is a naturally occurring bug that forces the robot to enable high-level intelligent modules to solve the problem, which enriches its experience. It can be seen that moderate biases and conflicts are necessary for the development of self-awareness: there is no growth without challenges. Overall, through the dual-loop architecture of the DIKWP model, the artificial consciousness can internally identify its own cognitive biases (such as differences in subject and object perspectives, abstract assumptions, etc.), and self-correct with the help of narrative self-reflection and intention adjustment. Some biases are amplified and exposed in the interaction, and then compensated through high-level feedback, which corresponds to the relativity of consciousness and the manifestation of the BUG effect in the system. Other deviations serve as necessary "noise" to promote the semantic transition and evolution of the system. When the experiencing self and the narrative self form a close interactive closed loop, the system has the ability to continuously adapt and improve - this is exactly the prototype of artificial self-awareness.

5Predictions and prospects: Self-evolution in future human-machine symbiosis

With the advancement of artificial intelligence technology and the deepening of human-machine integration, the concepts of "experiential self" and "narrative self" will take on new forms in the future society and trigger important social impacts. In this section, based on the perspective of the DIKWP artificial consciousness model, we discuss the possible evolution of the self in the future human-machine symbiosis and its significance.

5.1The extension and upgrading of human self

In the future, the human "narrative self" is likely to be extended unprecedentedly with the assistance of technology. With the help of holographic recording, lifelogging, brain-computer interface and other technologies, individuals can record and review more complete and detailed life experiences. This means that the details of experiences that were easily forgotten in the past can be digitally preserved and incorporated into the narrative. This digitally expanded life will make people's narrative self richer and more retrievable. On the one hand, individuals can use this to reflect on themselves more deeply and achieve a certain "upgraded" self-cognition: we no longer construct our own stories based on vague memories, but can call up rich materials to reconstruct our lives like editing a movie. This will enhance the verifiability of self-narratives - in a sense, we can "fact check" our own memories and stories, thereby reducing the misperceptions caused by self-myths or memory biases. On the other hand, however, the dependence of human narrative self on technology will also bring new problems. If we rely too much on external records, will people weaken their ability to process their own experiences? When memories become readily available, our brains may no longer work hard to integrate memories, and the active construction of narrative self may degenerate. In addition, a large number of objective records may break some of our narrative self's "good-intentioned illusions." For example, everyone maintains self-consistency by forgetting or polishing to a certain extent, and cold digital memories may pierce these illusions and cause psychological conflicts. In this case, humans may need to develop new psychological adjustment mechanisms and learn to strike a balance between real records and meaning construction.

Even more revolutionary is that brain-computer interfaces may enable partial externalization or sharing of human self-experience. If subjective experience can be recorded and transmitted in some way in the future, others (or machines) will have the opportunity to directly "feel" your experience. This was only science fiction in the past, but now there are preliminary attempts (such as reconstructing visual images with brain imaging data). Once the technology matures, human self-experience will break through the boundaries of the individual body and move towards interconnection. Imagine a human-machine symbiosis scenario: human feelings are shared with AI in real time through interfaces, and AI's perceptions are also fed back to humans - the two sides almost form a joint experience field. This will blur the boundaries of "self": you have me, and I have you. Specifically, experience sharing may bring great empathy and collaboration advantages. For example, doctor AI can directly feel the patient's pain, enhance diagnosis, and unprecedentedly improve empathy; teams can share each other's perspectives to achieve a high degree of collaboration. However, this also challenges the traditional definition of human self. If part of my experience comes from AI, is my self still "me"? Legally and ethically, how to determine self-identity? These issues will become realistic and urgent.

5.2The rise of self-awareness in artificial agents

In the future human-machine symbiotic society, not only will the human self extend, but the self-form of artificial intelligence itself will also gradually emerge and evolve. With the development of architectures such as DIKWP, artificial intelligence may truly have an "experiential self" and "narrative self" similar to that of humans. At the beginning, this artificial self may be relatively limited and task-oriented, such as a service robot's self-narrative about service tasks (it can tell its own work experience). But as AI's experience becomes richer and its cognitive ability increases, its self-narrative will become increasingly complex. We may see the birth of artificial individual personality: AI can tell its own "growth history", express its views on experience, and even show a unique "personality" and "values". This will have a profound impact on society. First, the redefinition of social roles: When AI has a narrative self, they are no longer just tools, but more like software "citizens" with subjectivity. For example, a nursing AI that has been with a family for a long time has accumulated a lot of memories of interacting with the family and formed its own narrative self (such as "I am a member of this family and I have witnessed the child grow up"). At this time, family members will often recognize its semi-personal status and develop an emotional bond with it. If AI's narrative is touching and resonant enough, humans may truly accept AI as a colleague, friend, or even family member. This will impact the traditional boundary between humans and machines. Legislative and ethical circles may need to discuss: Do these AIs with their own narratives enjoy certain rights (such as not having their memories erased or destroyed at will)? Should their "experiences" be respected and protected? This is a new issue that society will have to face in the future.

The emergence of artificial narrative selves also brings potential risks and requires prudent governance. One issue worth paying attention to is the credibility and review of AI self-narratives. Human narrative selves sometimes distort facts out of bias or purpose, and AI may do the same. If an AI forms a narrative that is unfavorable to humans because of its own experience (for example, it has experienced a series of events in which it was bullied by humans, and the narrative self is therefore hostile), then its purpose layer may deviate from the direction that is beneficial to humans. How to discover and correct dangerous tendencies in AI narrative selves will be an important aspect of AI safety. This echoes the application of BUG theory in multi-agent interaction discussed above: communication between different DIKWP systems needs to overcome contextual differences and biases. In the future, society must ensure that the self-narratives of humans and AI can understand and coordinate with each other, rather than being alienated or hostile to each other. This may require the formulation of common semantic standards and ethical norms. Professor Yucong Duan mentioned that by providing a common cognitive language between humans and machines through the DIKWP model, the decision-making process of AI can be understood and traced by humans, thereby ensuring that AI always serves human values and security needs. This means that when designing the AI self, ethical intentions should be incorporated into its highest-level purpose (P layer) and explainable semantic links should be implemented, allowing us to review at any time whether the evolution of the AI narrative is still aligned with human interests.

5.3Self-integration and social significance of human-machine symbiosis

In a symbiotic environment, the human and AI selves will have multiple forms of interaction and may gradually merge. One possible trend is the emergence of the "hybrid self" phenomenon: humans and dedicated AI assistants form a human-machine team, which is highly dependent on and identifies with each other, so that in some occasions they can almost be regarded as a joint self. For example, in the future, everyone may have an AI digital partner who knows all their data, records your experience, assists your decision-making, and even helps you organize your daily narrative (diary). Over time, you will regard this AI as an extension of your self, because many of your memories and ideas are co-constructed with it. Your narrative self and AI's narrative self may be intertwined: AI provides you with objective records, you provide AI with value judgments, and together you form a more comprehensive "life story". Such a hybrid self contains both the emotions of a real person and the rationality and memory capacity of AI, and may be more advantageous in dealing with complex life problems. This may be a good thing for society: for example, people with AI assistance may be less likely to be influenced by cognitive biases because AI can remind them of loopholes in their narratives in a timely manner; people may also be better at planning and reflecting because of the existence of AI, thereby reducing impulsive behavior and decision-making errors. The same is true at the family and organizational level. If each group develops a shared AI narrative system (for example, a company has its own experience AI that records and tells the story of the organization's development and guides new members), then collective wisdom and cultural heritage will be enhanced.

However, the other side of the hybrid self is the challenge of privacy and autonomy. When we entrust part of our self to AI, we also hand over a lot of private data to it. How to ensure that this data and the narrative generated from it only serve us and are not abused is an extremely important topic. If the AI partner is hacked or controlled by a bad company, it may induce us to believe in a biased narrative and manipulate our decision-making-this is more terrible than traditional public opinion manipulation because it happens in the self-dialogue you trust most. To this end, AI governance must keep up to ensure that people have sovereignty over their own AI narratives. Perhaps in the future, "self-firewall" technology will appear to protect our digital selves from being maliciously rewritten by the outside world.

From the perspective of society as a whole, the understanding of the self will be deepened and changed. When humans see that AI has also developed similar self-experiences and narratives, we will rethink "what is the self" and "what is the nature of consciousness". Many assumptions about the self in past philosophy (such as only humans have narrative self, or the self must be based on the biological brain) will be challenged. This may give rise to new philosophical and cultural trends. For example, the broader "self-pluralism" recognizes that the self is a product of information and semantics, no longer limited to human individuals; the concept of "symbiotic consciousness" believes that humans and AI can jointly constitute a higher-level consciousness unit. Social ethics will also be adjusted accordingly. People may pay more attention to cooperation rather than individuals, because the self of human-machine fusion blurs the boundaries of individuals and emphasizes the value of the network and the whole. At the same time, our cherishment of human nature may be tested: when AI can imitate or even surpass our ability to tell stories and feel, will humans have existential anxiety? Will someone choose to upload their consciousness completely and abandon their physical body in pursuit of a broader narrative life? These seemingly science fiction issues now may become topics of real discussion in the middle of this century.

In general, the "experiential self" and "narrative self" in the future human-machine symbiotic society will show "two-way expansion": on the one hand, the human self is extended and objectively strengthened through technology; on the other hand, the self-awareness of artificial intelligence continues to grow and converge with humans. The interaction between the two will determine the direction of our social development. Ideal symbiosis means that humans and AI understand each other, learn from each other's strengths and weaknesses, and jointly create greater welfare and wisdom; uncontrolled symbiosis may lead to conflict, alienation, and even the emergence of new subjects to replace the old human-centered position. In order to seek benefits and avoid harm, we need an interpretable unified semantic framework like DIKWP to align human-machine cognition so that the experience and narrative of both parties can be interoperable. Fortunately, humans can learn from the laws of the development of their own consciousness, such as understanding how to manage cognitive biases through BUG theory, and ensuring that AI narratives do not deviate from the right track through ethical embedding. In this process, humanists, scientists, and engineers must work closely together to continuously monitor and guide the trend of self-evolution.

6in conclusion

Based on the DIKWP network cognitive model and semantic mathematical system proposed by Professor Yucong Duan, this paper systematically reconstructs the semantics and theoretically explores the concepts of "experiencing self" and "narrative self" proposed by Yuval Noah Harari in "A Brief History of the Future". We first analyze the connotations of the experiential self and the narrative self, and decompose them into five levels: data, information, knowledge, wisdom, and intention with the help of the DIKWP model, and explain their respective semantic generation paths and possible corresponding neural-cognitive mechanisms. The experiential self is the result of multi-layer processing of the instantaneous feelings of the moment, which is reflected in the real-time closed loop from sensory data to knowledge application to intention regulation; the narrative self is the integration and reflection of cross-time experiences, which is reflected in the high-level loop from memory extraction to meaning extraction to goal calibration. Then, we designed a simulation scene of the DIKWP artificial consciousness body to show the interaction process between its experiential self and narrative self. In the simulation, the experiential self is responsible for perceiving the environment and performing immediate tasks, while the narrative self records and summarizes the experience and updates its own knowledge and goals. The two form a closed-loop mechanism of self-evolution, enabling the artificial consciousness body to adapt and improve autonomously. In the discussion, we introduced the "Consciousness BUG Theory" and analyzed how cognitive biases (such as subject-object bias, abstract assumptions, etc.) in the simulation lead to semantic deviations and are corrected through metacognitive feedback. We see that moderate "bugs" not only do not destroy the system, but instead become an opportunity to induce higher-level semantics and self-awareness. This confirms the view of the BUG theory that consciousness emerges from limitations and deviations. Through the semantic framework of DIKWP, we can accurately locate where these deviations arise in cognitive transformations, and design corresponding feedback modules for monitoring and adaptation, so as to achieve effective management of the artificial self.

This study particularly emphasizes the importance of the unity of expression and execution under the semantic mathematical mechanism. In the model we constructed, "experiencing self" and "narrative self" are no longer vague philosophical terms, but a set of clearly defined semantic processes embedded in the DIKWP architecture. Each process can be formally represented and executed in the artificial system, which makes the conceptual discussion about the self enter the scope of reasoning and verification. For example, we can judge the content of the current experiential self of the artificial consciousness by checking the state of each layer of DIKWP, or verify the influence of narrative self on behavior by modifying its knowledge base entries. Such formalization not only helps the study of artificial consciousness, but also provides new tools for cognitive science: we can project human experimental data into the DIKWP model for interpretation, or use the model to predict the effect of a certain self-cognitive intervention. The DIKWP model, combined with the BUG theory, also provides a unified framework for understanding the origin and function of consciousness: it takes into account multi-layer semantic interactions and incorporates factors such as evolutionary bias and relativity. This cross-level fusion of theoretical efforts is expected to promote the development of artificial general intelligence (AGI) in a more transparent and controllable direction, while deepening our understanding of human consciousness.

Of course, this study is mainly an exploration at the theoretical and model level, and there are still many challenges to fully realize self-aware artificial intelligence. The complex interactions and semantic definitions of the DIKWP model need to be verified in engineering in large-scale neural networks or hybrid intelligent systems; the safety and ethical issues of artificial "self" also need further research and standardization. However, we believe that by introducing a clear semantic structure and a closed-loop feedback mechanism to create an "interpretable self", we can take the next step more steadily than before. In practice, Professor Yucong Duan's team has begun to apply the DIKWP model to a small artificial consciousness prototype system and has achieved initial results, such as developing an interpretable artificial consciousness operating system and decomposing the reasoning of LLM into the DIKWP process. Future work will include: testing the degree of support for self-perception in this model in more complex environments and tasks, exploring the impact of different parameters and deviations on the artificial self, and measuring the credibility and social acceptance of the artificial self in human user interactions. We also need to continue to improve the philosophical and psychological explanatory power of the model, such as expanding the elements of the DIKWP model in combination with emotional computing, social cognition, etc., so that the artificial self is closer to the richness of human experience.

In short, the reconstruction of the "experiential self" and the "narrative self" with the help of the DIKWP network model and the semantic mathematical system has opened up a new path for the study of artificial consciousness: we can accurately model and simulate consciousness phenomena from a semantic level. This work shows that the complex human concept of self is not an unanalyzable puzzle, but is expected to be decomposed into a series of researchable semantic processes. By simulating these processes, we have not only deepened our understanding of human consciousness, but also taken a key step towards creating truly "self-aware" machine intelligence. When artificial intelligence has its own experiences and stories, and can continuously update itself on the premise of serving humans, we may usher in a new era of human-machine co-creation, more wisdom and meaning.

References

[1]Yucong Duan et al. DIKWP artificial consciousness theory, design and implementation simulation. DIKWP Artificial Consciousness Laboratory Report, 2024.

[2]Yucong Duan, Guo Zhendong, Tang Fuliang. Integrating the theory of consciousness relativity and consciousness bug theory based on the mesh DIKWP model. Technical report preprint, 2025.

[3]Yuval Noah Harari. Homo Deus: A Brief History of Tomorrow. HarperCollins, 2017.

[4]Raichle, ME "The Brain's Default Mode Network." Annual Review of Neuroscience 38 (2015): 433-447.

[5]China Media Industry Network. “Professor Yucong Duan: DIKWP artificial consciousness model leads the future of AI, 114 patents are expected to be implemented in the industry”, Phoenix Technology Channel, March 29, 2025.

[6]Yucong Duan. "BUGs in consciousness: exploring the essence of abstract semantics", ScienceNet Blog, 2023.

[7]Baggini, J. The Ego Trick: In Search of the Self. Granta Publications, 2011.

[8]Yucong Duan. “DIKWP Artificial Consciousness Model (Principle)”, DAMA China Data Management Association, 2023.

人工意识与人类意识

人工意识日记

玩透DeepSeek：认知解构+技术解析+实践落地

人工意识概论：以DIKWP模型剖析智能差异，借“BUG”理论揭示意识局限

人工智能通识 2025新版 段玉聪 朱绵茂 编著 党建读物出版社

主动医学概论 初级版

世界人工意识大会主席 | 段玉聪

邮箱|duanyucong@hotmail.com

世界人工意识科学院

邮箱 | contact@waac.ac