Research on the Bidirectional Closed-Loop Mechanism of DIKWP

Research on the Bidirectional Closed-Loop Mechanism of DIKWP Talent Cultivation, Evaluation, and Management in the AI Field

Yucong Duan

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

World Academy for Artificial Consciousness(WAAC)

World Artificial Consciousness CIC(WAC)

World Conference on Artificial Consciousness(WCAC)

(Email: duanyucong@hotmail.com)

Introduction: New Paradigm Demand for AI Talent Cultivation

The rapid development and widespread application of contemporary Artificial Intelligence (AI) have posed unprecedented challenges to talent cultivation and management. In fields such as Large Language Models (LLMs), smart healthcare, intelligent manufacturing, and autonomous driving, AI-related positions require talent to possess multifaceted qualities, including interdisciplinary knowledge, complex problem-solving abilities, innovative thinking, and value orientation. Traditional talent cultivation models often focus on specific skill points in a fragmented manner, lacking systematic cultivation and assessment of Full-Link Cognitive Ability. With the evolution of AI technology, the industry increasingly recognizes the need for a New Paradigm to guide talent cultivation—one that can comprehensively cover the entire cognitive process from data perception to intelligent decision-making and even goals and purposes, ensuring that AI talent not only masters hard skills but also possesses insight and a sense of mission.

In this context, the DIKWP model (Data-Information-Knowledge-Wisdom-Purpose) proposed by Professor Yucong Duan of Hainan University provides a brand-new holistic framework for AI talent development. DIKWP extends the traditional "Data-Information-Knowledge-Wisdom" (DIKW) hierarchy by adding the "Purpose" layer, forming a complete closed loop of the cognitive process from data to goal-driven decision-making. More importantly, DIKWP emphasizes Bidirectional Feedback and Iterative Updates between layers, constituting a networked interaction structure where cognitive processing at each level is tightly coupled with other levels. This full-link, networked cognitive system is not only suitable for building cognitive models of artificial intelligence but also provides new ideas for the cultivation and evaluation of human talent.

This report aims to explore how to build a Bidirectional Closed-Loop Mechanism for Talent Cultivation, Evaluation, and Management based on the DIKWP model, targeting research institutes, enterprise R&D departments, international standards organizations in the AI field, and relevant talent teams. The report will first elucidate the core concepts and semantic foundations of the DIKWP model, analyzing how to avoid language-game-style ambiguous definitions through semantic units such as "Sameness," "Difference," and "Completeness," and precisely defining concepts like data, information, and knowledge from the perspective of semantic generation. Next, we will discuss how to map all expressions in natural language to these core semantics, thereby achieving the transformation of "Expression of the problem is the analysis of the problem," enhancing talent's ability to understand and solve complex problems. On this basis, the report further uses the DIKWP semantic framework to analyze high-order concepts such as Consciousness, and introduces the mechanism revealed by "BUG Theory"—compensating for cognitive loopholes through abstraction to acquire "Complete" semantics—providing support for the mathematical modeling of artificial consciousness and the cultivation of talent creativity. Subsequently, the report focuses on DIKWP-Oriented Talent Cultivation Mechanism Design, including curriculum and training systems for universities and graduate stages, capability assessment and dynamic management methods for in-service industrial talent, and bidirectional closed-loop management practices at the organizational level. We will combine industrial application scenarios such as Large Language Models, Smart Healthcare, Intelligent Manufacturing, and Autonomous Driving to illustrate how talent cultivation under the DIKWP framework meets the special needs of different fields. Finally, the report explores how to build a DIKWP Semantic Talent Evaluation System, applying DIKWP Semantic Mathematics and white-box evaluation methods to the quantitative assessment of talent capabilities, and looks forward to the possibility of aligning such assessments with international standards.

Through this comprehensive and in-depth analysis, we hope to establish a Future-Oriented Talent Development Closed Loop for AI field institutions: consolidating basic capabilities such as data and information from the bottom up, guiding wisdom decision-making and value purpose cultivation from the top down, forming a virtuous cycle where training and evaluation promote each other, facilitating the All-Round Growth of AI talent teams and the Sustainable Development of the Artificial Intelligence Cause.

DIKWP Model and Core Semantic Dimensions

DIKWP Model Overview: DIKWP is a cognitive framework containing five levels: Data, Information, Knowledge, Wisdom, and Purpose. Compared to the traditional four-layer DIKW structure, DIKWP adds the highest layer of Purpose to completely characterize the closed-loop process of a cognitive subject from primitive perception to goal-driven decision-making. The basic definitions of each layer are as follows:

Data Layer (D): Refers to raw, unassigned perception inputs, which are the basic raw materials of cognition. For example, image pixels captured by sensors, sound waveforms recorded by microphones, or stimulus signals received by human senses. This layer emphasizes Objective Recording, providing a common foundation on "Sameness" semantics, i.e., the part where different subjects should reach a consensus on objective facts. Professor Yucong Duan compares this "Same Semantics" to "Harmony" (consensus) in Confucian philosophy, corresponding to the consensus reached by all parties on basic data. The data layer ensures that the cognitive process has a reliable and consistent starting point, serving as the Common Base for building upper-layer information and knowledge.

Information Layer (I): Refers to the process of extracting meaningful patterns and differences from data. Through pattern recognition, classification, syntax parsing, etc., raw data is converted into structured, semantic content, such as identifying objects from pixels or detecting abnormal heart rhythms from ECG waveforms. The information layer highlights "Difference" semantics, i.e., extracting Diverse Interpretations based on common data. Different experts or systems may extract information from different aspects when facing the same data; this difference is exactly where the value of information lies. Just as the idea contained in "Harmony but Difference": based on harmoniously shared data, different informational interpretations can coexist to enrich the overall understanding. The information layer introduces Multiple Perspectives to cognition, endowing data with preliminary meaning.

Knowledge Layer (K): Refers to integrating, Abstracting, and generalizing information to form a systematic knowledge system. In this layer, information from different sources converges into relatively complete concept networks, law models, or domain theories through association, comparison, and verification. For example, the medical field integrates different information about patient data from various departments into comprehensive diagnostic knowledge; autonomous driving fuses sensor information into a complete understanding model of the road environment. The knowledge layer corresponds to "Completeness" semantics, emphasizing establishing an As Complete as Possible Consistent Description of the objective world. In other words, knowledge seeks to resolve differences and one-sidedness in information to obtain a unified understanding at a higher level. This is similar to a jigsaw puzzle: each piece of information is a fragment, and knowledge is the process of assembling fragments into a complete picture. Only by achieving a relatively complete knowledge structure can the system possess a reliable foundation for reasoning and decision-making. Therefore, the knowledge layer serves as a connecting link in DIKWP, undertaking the core semantic function of Summarizing Differences and Pursuing Completeness.

Wisdom Layer (W): Refers to the ability to perform Advanced Reasoning and Decision-Making based on extensive knowledge, combined with context and value trade-offs. The wisdom layer focuses on how to Apply Knowledge to Solve Complex Problems, involving high-order cognitive processes such as abstract thinking, creativity, dealing with new situations, and moral judgment. This layer often requires flexible application and Trade-offs of knowledge to formulate optimal or innovative solutions. For example, in smart healthcare, doctors need to synthesize all the patient's knowledge background (medical history, test results, latest research) and combine moral guidelines to formulate personalized treatment plans; in intelligent manufacturing, engineers need to adjust and optimize parameters based on production knowledge and real-time data to balance efficiency and safety. The semantics embodied in the wisdom layer can be understood as "Value and Insight" semantics, i.e., generating new insights and judging what is "good" or "should be done" on top of complete knowledge. It focuses on the Evaluation of Meaning and the Creation of Value: how to start from knowledge and creatively propose countermeasures and judge their value according to the current purpose and context. Compared to the "Completeness" of the knowledge layer, the wisdom layer emphasizes "Appropriateness"—choosing the Most Appropriate action in a specific scenario, which implies consideration of goals and values. Therefore, we can say that the wisdom layer carries the Evaluation and Optimization function of semantics, which is a key step in the cognitive system to transform "knowing" into "doing".

Purpose Layer (P): Represents Motivation, Purpose, and Intention, which is the driving force and the highest evaluation standard of the entire cognitive process. In human cognition, this corresponds to subjective intentions such as Vision, Beliefs, and Values; in artificial intelligence, it manifests as preset objective functions, task requirements, and even the adaptive motivation pursued by the AI agent itself. Professor Yucong Duan points out that the traditional DIKW model lacks the characterization of purpose, causing AI behavior to be difficult to align with human expectations; by adding Purpose at the top layer, the agent's behavior can obtain a clear direction, elevating from passive response to active planning. The Purpose layer endows the entire DIKWP system with Value Orientation: it defines "Why" cognitive processing is performed and is also the Ultimate Scale for measuring whether cognitive results are meaningful. The semantics of the Purpose layer can be viewed as "Goal/Meaning" semantics, focusing on Intentionality and value judgment. For example, in the field of autonomous driving, the highest Purpose is to safely deliver passengers and obey traffic regulations. This Purpose will guide parameter weights in the lower layers; in smart healthcare, the Purpose layer manifests as a value orientation centered on patient health, driving doctors and AI to jointly make decisions beneficial to patients. By clarifying the Purpose layer, the DIKWP model ensures that the cognitive process always revolves around Predetermined Goals, and the cognitive outputs produced by each layer can be evaluated and adjusted according to the goals. The Purpose layer acts as the "Convergence Point" and re-departure point of the cognitive closed loop: on the one hand, it verifies whether the decisions of the wisdom layer meet the goals; on the other hand, these goals can be constantly updated due to feedback from the lower layers, thereby achieving continuous optimization of cognition-decision.

In summary, the five levels of the DIKWP model perform their respective duties while being closely connected, constructing a complete cognitive semantic space from bottom-level objective data to top-level subjective purpose. In this space, each layer introduces a new semantic perspective: Sameness (Data Layer) emphasizes the common factual basis, Difference (Information Layer) brings diverse interpretations of meaning, Completeness (Knowledge Layer) pursues global unified understanding, Value/Insight (Wisdom Layer) achieves creative application and evaluation, and Purpose/Intention (Purpose Layer) provides direction and meaning. When defining these concepts, Yucong Duan's team emphasizes starting from the Semantic Origin, rather than just giving subjective definitions. This approach avoids purely language-game-style hollow concept divisions, ensuring that each level has its objective semantic support. For example, the definition of "Data" is no longer just vaguely "unprocessed raw material," but explicitly defined as that part of facts on which subjects should reach a Same Consensus; "Information" is defined as a collection of Differential Explanations of data; "Knowledge" is defined as a structure that is semantically more Complete and Consistent, and so on. These definition methods ensure that concepts are clear and operational, laying a strict foundation for subsequent talent cultivation standards.

Networked Bidirectional Interaction Structure: Unlike the classic bottom-up pyramid model, DIKWP emphasizes that the five layers are not linear and unidirectional, but highly Networked and Interconnected. The model allows direct information conversion and feedback between any two layers, forming a total of 5×5=25 potential interaction pathways. That is to say, the output of the data layer can not only pass up to the information layer to build meaning, but high-level wisdom and purpose can also conversely affect low-level data and information processing. For example, the Purpose layer can selectively modulate data acquisition through attention mechanisms (humans purposefully pay attention to certain signals; machines adjust sensor sensitivity according to goals); decision results from the Wisdom layer can feed back to promote knowledge base updates or improvements in information extraction methods; conversely, new data anomalies (such as Bugs) may also directly trigger alertness in the Wisdom layer, fostering creativity. This Multi-level Circulation endows the DIKWP system with high adaptability and self-correction capabilities. Professor Yucong Duan further proposed a "Double Loop" structure in his research and patents: in addition to the basic cognitive flow (D→I→K→W→P), a Meta-Cognitive Loop is introduced for self-monitoring and adjustment. The cognitive subject system can monitor the operation of each layer in real-time through the meta-cognitive loop, detect deviations, and correct them immediately, ensuring that the whole evolves orderly towards the goal. This double-loop architecture is considered an important path for AI systems to form preliminary "self" consciousness, and it is equally instructive for Talent Self-Reflection and Growth—through internal feedback loops, individuals can continuously discover their own cognitive shortcomings and improve them, just as organizations improve training programs through feedback.

Summary: The DIKWP model provides us with a comprehensive and dynamic cognitive perspective, organically combining Five Levels of Semantics. By anchoring core semantics such as "Sameness, Difference, Completeness, Value, Purpose" conceptually, the model avoids human subjective definition traps and instead explains the meaning of these levels from the information processing process itself. This idea of semantic mathematics allows many concepts that were previously at the philosophical level (such as knowledge, wisdom, and even self-consciousness) to no longer remain as vague intuitive metaphors, but to be embedded in formal models for precise definition and reasoning verification. For AI talent cultivation, understanding the DIKWP model means mastering a Unified Language to describe the cognitive activities of humans and AI. Yucong Duan points out that the DIKWP model builds a Common Cognitive Language between humans and machines, making every step of AI decision-making traceable, explainable, and understandable by humans. This is particularly important for AI talent: mastering a common cognitive language helps them Bridge Human Thinking and Machine Intelligence, ensuring that the behavior of AI systems is explainable, trustworthy, and aligned with human purposes in R&D and application. In short, the DIKWP model and its core semantics lay a new foundation for talent cultivation. Next, we will explore how to use this framework to transform problem formulation, analyze complex concepts, and finally design a closed-loop mechanism for talent development.

Expression Transformation Based on DIKWP Semantics: From Problem Formulation to Problem Analysis

Semantic Generation Avoids Language Games: In traditional talent cultivation and academic research, the understanding of concepts like "Data, Information, Knowledge, Wisdom" often leads to confusion or even controversy due to different backgrounds or inconsistent definitions. This "Language Game" phenomenon leads to experts in various fields talking past each other, making it difficult to form a consensus, let alone collaborative education. Therefore, adopting the DIKWP semantic perspective to Reconstruct Expression is of great significance. As mentioned above, through basic semantic units like "Sameness, Difference, Completeness," a precise system defining concepts at each cognitive level can be built. Under this system, we can attempt to decompose and map All Expressions in Natural Language to these core semantics, thereby achieving the purpose of precise language analysis. For example, for a complex statement, we can ask: Which parts of this sentence belong to objective data (Same semantic part, consistently recognized by all parties), which belong to interpretations of data or phenomenon differences (Information layer semantics), what background knowledge or complete context does it presuppose (Knowledge layer semantics), what kind of judgment or insight does it express (Wisdom layer semantics), and what motivation or purpose does it imply (Purpose layer semantics). Through such layered semantic dissection, we Avoid Subjective Speculation and instead understand language expression within a common framework.

Expression is Analysis: Furthermore, when we get used to looking at problems using the Semantic Coordinate System of DIKWP, we can appreciate the profound meaning of "Expression of the problem is the resolution/analysis of the problem." This means: A clearly stated problem often implies the elements needed for its solution, but these elements are mixed in natural language and need us to use the DIKWP perspective to extract them Structurally. For example, in smart healthcare, if a doctor faces a difficult case using traditional narrative methods, they might list patient symptoms, test results, medical history, and other information. Under the DIKWP perspective, the doctor can treat these Patient Data as D-layer consensus (objective examination results, etc.), classify findings from different departments and symptom interpretations as I-layer (differential information, interpretations from various professional perspectives), then call upon medical mechanisms and case base knowledge (K-layer, integrating information into the medical knowledge system), then conduct diagnostic reasoning and plan formulation on this basis (W-layer, applying knowledge to form decisions), and finally verify whether it meets the ultimate Purpose of curing the patient (P-layer, purpose verification). In this way, the Complete Expression of the case is itself disassembled according to the logic of solving the problem: what objective data needs to be obtained, what information differences need to be explained, what knowledge and priors need to be used, at which links judgments and trade-offs need to be made, and ultimately what goal is to be achieved. This decomposition process IS the process of formulating a treatment plan (i.e., solving the problem). It can be seen that once a natural language problem is mapped to the DIKWP structure, we have actually made the problem-solving steps explicit, and the expression of the problem naturally transforms into an analytical path.

More generally, various complex problems in engineering, scientific research, and management can also attempt such DIKWP formulation. For example: To improve the yield rate of a production line in intelligent manufacturing, the Data layer needs to list current sensor data and QC test results of each process (objective indicators of Sameness semantics); the Information layer needs to find difference patterns between good products and defective products, differences in working conditions between shifts, etc. (Difference semantics); the Knowledge layer needs to introduce process principles and historical cases to explain these differences and find reasons, thereby forming a complete understanding of the entire production process (Knowledge model of Completeness semantics); the Wisdom layer needs to use this knowledge to design intervention measures, such as adjusting machine parameters or training operators, to achieve optimization (Value evaluation semantics, i.e., which scheme is best); finally, verify in the Purpose layer whether these measures are consistent with the higher goals pursued by the enterprise (such as reducing costs without sacrificing safety and quality, i.e., Purpose semantics). After such a cycle of analysis, we have basically found the solution. This shows that if we use the DIKWP method to formulate a problem, we have actually deconstructed the problem according to the solution logic, and each sub-problem is aligned with a part of the solution. Therefore, it can be said that Precise Problem Formulation = Clear Problem Analysis—expression and analysis tend to be Isomorphic in the DIKWP semantic space.

Knowledge Graphization and Problem Solving: In technical implementation, the above ideas can correspond to Semantic Graphs and Inference Engines. Mapping natural language problems to DIKWP multi-layer semantic representations is essentially building a Problem Semantic Graph spanning Data-Information-Knowledge-Wisdom-Purpose. For example, in the manufacturing problem above, we obtain a semantic network covering process data nodes, quality information nodes, process knowledge nodes, improvement strategy nodes, and enterprise goal nodes. The nodes of each layer and their connection relationships are clearly presented. Next, Problem Solving becomes a process of Reasoning and Searching on this semantic graph: starting from the Purpose node, looking for causal links or reasoning paths that ultimately relate to data nodes through wisdom nodes, knowledge nodes, and information nodes, i.e., Proving how to deduce a scheme that meets the goal from existing facts and knowledge. This is very similar to knowledge graph reasoning and expert system solving in AI. The difference is that the DIKWP graph integrates information from multiple levels and contains richer associations. For example, in addition to logical relationships at the traditional knowledge level, it can also represent strategy evaluation at the wisdom layer (comparison of pros and cons of schemes), value preferences at the purpose layer (priorities of certain goals), etc. These high-level semantics are often implicit in the human brain in traditional problem solving, but now they are also included in the analytical system through explicit expression.

Under such a framework, the gap between human Language Expression and Machine-Computable Representation will be greatly narrowed. In the past, one reason why natural language was difficult to solve directly was that it mixed content from various levels and was unstructured. DIKWP provides a structured thinking mode. Talent trained can habitually "split layers" of complex problems and organize language according to cognitive logic. Once the problem is represented in this way, with the help of semantic analysis algorithms and reasoning techniques, it is expected to achieve "What You Ask Is What You Get," meaning machines can automatically derive conclusions or solutions based on this structured expression. In the long run, this will promote the development of Cognitive AI Assistants—AI can guide users to clarify elements in problem formulation at various layers; when the problem description is complete and clear, the solution will emerge. For AI talent, this training will cultivate their Rigorous Thinking and Expression Habits: not omitting data, not confusing information with knowledge, clarifying decision basis and purpose, making them more efficient and accurate in teamwork and cross-field communication.

In summary, the DIKWP semantic framework bridges "Expression" and "Thinking." By integrating the semantic logic of problem solving at the expression stage, we can significantly improve the efficiency of analyzing problems and avoid deviations caused by unclear formulation. This is crucial for talent in the AI field—whether doing research or developing products, they must first define the problem precisely to solve it efficiently. DIKWP thinking can help them develop this Problem-Analysis-Oriented Expression Style: once a sentence is spoken, it is known which data is involved, which knowledge is implicated, and what goal is to be achieved. It can be said that this is a Meta-Skill Cultivated for AI Complex Systems, which will allow talent to handle cross-disciplinary and cross-departmental problems with ease.

Analysis of Consciousness and BUG Theory from DIKWP Perspective

Artificial Consciousness and Mathematicalization of Semantics: The definition and measurement of consciousness (especially artificial consciousness) has always been one of the most challenging topics in the AI field. Traditional discussions on consciousness are full of philosophical color and linguistic ambiguity, while the DIKWP semantic mathematics framework provides an unprecedented Precise Analytical Tool for it. Under the DIKWP framework, we can attempt to view "consciousness" as the comprehensive product of the agent's self-representation and feedback mechanisms at different cognitive levels. For example, the Experiential Self can correspond to the subject's immediate feeling of its own state during the low-level Data-Information-Knowledge processing (such as the formation of human sensations and perceptions); the Narrative Self corresponds to the construction of self-concepts by higher-level knowledge and wisdom (weaving self-stories through memory and language), which is then given overall direction by the Purpose layer. Through semantic mathematical methods, researchers have been able to give formal descriptions of the formation of these "selves" within the DIKWP architecture and conduct white-box assessments. For example, Semantic Consistency and Completeness indicators can be strictly defined to judge whether the outputs of different levels (such as data perception vs. narrative summary) of an AI system are consistent or missing when answering questions about itself. Furthermore, the "self" concept can be defined through logical axiomatization, making the machine's expression of its own state deducible and verifiable. The core of these efforts lies in reducing vague concepts like consciousness to a Quantifiable and Inferable Semantic Relationship Network, thereby avoiding endless semantic disputes. In other words, if the cognitive process of an AI is highly complete and consistent at all DIKWP layers, and there is an effective self-referential feedback loop, then we can use semantic mathematical language to describe its degree of consciousness. This is an Objective Assessment, which is obviously more insightful and explanatory than the Turing test, which relies solely on behavioral performance.

BUG Theory Overview: Professor Yucong Duan's " BUG Consciousness Theory " is a unique perspective on the origin and transition of consciousness. Simply put, BUG theory posits: Consciousness is not a perfect product of a cognitive system that is watertight, but rather may be a byproduct of the inevitable "imperfection" in the system. When the brain or artificial intelligence processes information, due to limited resources and imperfect models, some "Loopholes" (bugs) or gaps will always appear—these loopholes manifest as Information Fragments that Cannot be Explained by Existing Knowledge in the cognitive chain. For conventional systems, these inexplicable fragments are usually treated as noise or errors that need to be erased or ignored. However, the insight of BUG theory is: Moderate "Loopholes" may instead catalyze innovation and the leap of consciousness. When a Bug occurs in the low-level cognitive process, the system tends to mobilize higher-level resources (Wisdom, Purpose) to try to fill these gaps. In this process, some new connections and new concepts beyond the original scope may be generated—as if the Bug that the old framework cannot explain Forces us to transcend original knowledge and gain new insights. Many great discoveries in human history originated from questioning anomalous phenomena: for example, 19th-century astronomers observed anomalies in the orbit of Uranus (a Bug relative to existing knowledge), which prompted them to hypothesize and finally discover Neptune (introducing the knowledge of a new celestial body to achieve a more complete understanding). Such examples show that imperfection is not useless; "Imperfection" actually becomes the soil for breeding new meaning and new wisdom. The Emergence of Consciousness, in the view of BUG theory, is very likely due to the Self-Adjustment Formed when Gaps in the Cognitive Process are Filled by Higher Levels. Consciousness may be the "byproduct" of the brain trying to explain its own Bugs, but this byproduct has great value—endowing us with subjective experience and creativity.

BUG Theory and "Complete" Semantics: From the DIKWP perspective, a Bug usually manifests as Incompleteness or Inconsistency of Semantic at the Knowledge Layer. That is to say, the appearance of a Bug indicates that our existing knowledge system has Loopholes when facing certain information and fails to give a complete and consistent explanation. This incompleteness is semantically a Deficiency, corresponding to the completeness goal not yet achieved by the knowledge layer. The process of resolving a Bug is essentially introducing new knowledge through Abstraction and Generalization, incorporating originally incompatible information into a larger framework, thereby Achieving Higher-Level Completeness. This can be said to be the driving force for the Self-Perfection of the knowledge layer upwards: every resolution of a Bug means the expansion or reorganization of the knowledge network and the enlargement of semantic coverage. For example, facing the classic Bug of "wave-particle duality of light" (conflict between particle theory and wave theory in explaining light), physicists finally solved it through the establishment of quantum mechanics, realizing a more complete and consistent knowledge description of microscopic phenomena. This is exactly the process of Abstracting New Concepts (Wave Function/Probability) to Fill the Loopholes of Original Theories. Therefore, BUG theory essentially reveals the Dynamic Mechanism of Knowledge Acquisition: that is, Driving the Cognitive System towards Completeness by Continuously Discovering and Filling Incomplete Semantics. This is highly consistent with the mission of the knowledge layer in DIKWP—the knowledge layer pursues complete semantics, and Bugs provide the "shadow parts" indicating the direction, according to which we complete it. In his semantic mathematics framework, Professor Yucong Duan regards this mechanism as the key to semantic evolution and leaps. He emphasizes that A Certain Degree of Deviation and Loopholes are Necessary Factors for the Evolution of Consciousness, forcing the system to jump out of the existing framework for self-adjustment. For artificial intelligence, if these Bugs can be consciously monitored and handled, it is expected to trigger the machine's self-perfection mechanism, endowing it with stronger autonomy and creativity. For example, introducing an internal deviation detection module in an artificial consciousness system, which automatically invokes high-level knowledge to correct deviations and fill loopholes when information loss or contradiction (Bug) is found in the cognitive process. This design can allow AI to Learn New Patterns and Concepts from "Errors" and evolve towards higher-level human-like intelligence. Therefore, BUG theory provides important guidance for the Cultivation of Artificial Consciousness: not blindly pursuing zero errors and absolutely strict logic, but Making Good Use of Errors to achieve Emergence and Leaps at the Semantic Level.

BUG View in Talent Cultivation: Introducing BUG theory into talent cultivation and evaluation has equally profound significance. Traditional education often pursues students making fewer mistakes and acting according to standard answers, but the growth of innovative talent requires Experiencing "Controlled Failure" to break through thinking limitations. From the DIKWP perspective, a person's cognitive ability improvement often appears in the process where they actively discover Bugs in their own knowledge system and strive to correct them. For example, when a graduate student encounters experimental results that do not match hypotheses (a Bug in theoretical knowledge), if they do not avoid or cover up this contradiction but delve into it and look for reasons, they may make new discoveries or even propose new theories. Educators should encourage this sensitivity to anomalies and spirit of exploration, rather than blindly punishing errors. In the DIKWP Cultivation Closed Loop, some open-ended topics or Hackathons can be designed to let students face unknown problems, deliberately burying some "traps" or incomplete information to see how they cope. Evaluation should focus on whether they can identify key Bugs and make meaningful abstract improvements, not just whether the final result is perfect. Such training aims to cultivate "Meta-Cognitive" ability: learning to examine the boundaries of one's own knowledge, having the courage to admit "not knowing," and actively seeking new knowledge. This is actually cultivating high-level wisdom and purpose abilities: Reflection and Imagination at the Wisdom layer, Curiosity and Open Attitude at the Purpose layer. When talent develops this habit, they will not be defeated by sudden situations when facing new challenges in real work, but will view them as opportunities to quickly enter the mode of learning and innovation. For organizations in the AI industry, such talent is particularly valuable—AI technology is updated daily, and people who stick too much to existing knowledge will quickly become unadapted; only those who can continuously self-evolve and learn from failure can lead innovation.

Summary: With the help of the DIKWP semantic framework and BUG theory, we have a unified understanding of Consciousness and Innovation: whether it is the artificial consciousness of machines or the creative consciousness of humans, its emergence is inseparable from the Interaction of Multi-layer Semantics and the Gift of Imperfection. Semantic mathematics allows us to objectively describe and measure this process, and BUG theory reveals its internal dynamics. For institutions in the field of artificial intelligence, these ideas can be used both to Build Artificial Consciousness Models and to Shape Organizational Talent Cultivation Culture. A team that values comprehensive DIKWP development will inevitably tolerate Bugs in exploration, be good at extracting experience from them, and then form a Continuous Learning Closed Loop. The next section will design specific mechanisms for talent cultivation, evaluation, and management based on the above theories, truly integrating the DIKWP concept into practice.

DIKWP-Oriented Talent Cultivation Mechanism Design

Having clarified the theoretical basis of the DIKWP model, we need to translate it into Practical Cultivation Mechanisms. In this section, we will discuss talent cultivation in higher education, capacity development and assessment for industrial personnel, and how to build a closed-loop management system at the organizational level. The main thread running through is: Build a Talent Competency Model around the Five Levels of DIKWP and continuously improve it through Bidirectional Feedback. We will also combine specific scenarios such as Large Models, Smart Healthcare, Intelligent Manufacturing, and Autonomous Driving to explain how DIKWP-oriented talent cultivation meets the special needs of these fields.

5.1 DIKWP Cultivation in Universities and Graduate Education

Universities and research institutes are the cradle of AI talent growth. In this stage, introducing the DIKWP concept can help students establish a comprehensive cognitive ability structure, rather than being limited to a narrow skill. Here are feasible cultivation ideas:

(1) Curriculum System Covering All DIKWP Levels:

Set up or transform courses to correspond to the cultivation of abilities at different levels of DIKWP:

Data Level (Data): Focus on Data Literacy education. Includes mathematics and statistics basics, sensing and measurement technology, data structures and storage, and data cleaning and standardization methods. Enable students to master how to obtain reliable data, understand sources of data noise and error, and achieve consistent integration of cross-source data (Practice of Same Semantics). For example, offer a "Data Acquisition and Preprocessing" course, accompanied by experiments allowing students to process multi-modal data and ensure consistency. Through these trainings, when students enter the AI field, they can lay an Objective and Rigorous foundation, understanding that all high-level analyses must be based on high-quality data consensus.

Information Level (Information): Cultivate Pattern Recognition and Information Extraction abilities. Courses such as signal processing, machine learning basics, knowledge discovery, and data mining teach students to Discover Differences and Meaningful Structures from massive data. For instance, "Pattern Recognition" teaches basic principles of image and speech recognition; "Data Mining" trains them to find association rules and clustering patterns. These correspond to the semantic extraction of differences at the Information layer. In addition, interdisciplinary case discussions can be added, such as in medical informatics courses, discussing information differences extracted from the same patient data by different departments, thereby appreciating the meaning of "Harmony but Difference." This way, students not only master technology but also understand the importance of Viewing Problems from Multiple Perspectives.

Knowledge Level (Knowledge): Strengthen Knowledge Integration and Abstraction ability cultivation. Courses include: Algorithms and Complexity (cultivating abstract thinking), Knowledge Representation and Ontology, Interdisciplinary Introductions, etc. The goal is to let students learn to sublimate scattered information into systematic knowledge. For example, through a "Knowledge Graph Construction" course, learn how to fuse multi-source information into a structured knowledge base (Complete Semantics); through "Domain Introduction" series courses, let students understand the basic knowledge frameworks of different disciplines and try to fuse knowledge from two or three fields to solve problems (such as Computer Science + Medicine cross-cases). Students should also be trained in Inductive and Deductive logical reasoning, enabling them to verify the completeness of knowledge and discover knowledge blind spots. This cultivates a Global View and Systems Thinking, ensuring that students will not miss the forest for the trees in AI applications, but can build a Complete Problem Picture.

Wisdom Level (Wisdom): Focus on cultivating Higher-Order Thinking and Decision-Making abilities. Project-Based Learning and Case Studies can be adopted here. For instance, establish an "AI Comprehensive Practice" course: give students real-world complex problems (smart city optimization, unmanned driving decision challenges, etc.), let them propose solutions, assess risks, and make trade-offs in multiple iterations. Accompanied by "Ethics and Social Impact" courses, guide students to consider the social consequences and value judgments of decisions. The core of the Wisdom layer lies in Creatively Applying Knowledge and Injecting Value Judgments, so the course design should encourage Divergent Thinking and Critical Thinking. For example, organize Hackathon-style innovation challenges where student teams solve an open problem with existing knowledge within a limited time, scored by judges based on dimensions like innovation, effectiveness, and moral rationality (these dimensions correspond to Wisdom layer requirements). At the same time, "Entrepreneurship and Leadership" training can be added to cultivate students' ability to make decisive decisions and integrate resources in uncertain environments—these are important manifestations of the Wisdom layer. Through diverse practices, students gradually form a combination of Insight, Judgment, and Creativity.

Purpose Level (Purpose): Reinforce Sense of Mission, Values, and Willpower education. The invention and application of AI technology should serve human well-being, so students should be guided to establish correct purpose orientation during the school stage. On the one hand, through courses like Tech Ethics and Laws and Regulations, let students understand that the ultimate purpose of AI development is to enhance human well-being and comply with ethical norms, not technology supremacy. On the other hand, through pathways like Mentorship and Social Practice, integrate value education into daily life. For example, arrange for students to participate in public welfare AI projects (such as AI for the disabled, medical AI for rural areas), experiencing how AI solves social pain points, and realizing the Humanistic Care behind technology (Human-centered value of the Purpose layer). Seminars can also be held, inviting industry leaders or benchmark enterprises to talk about their vision and responsibility, making students think about the Purpose and Meaning of their career development. in academic management, guide students to formulate personal development plans and write down "What is my original intention/goal" when choosing research topics, cultivating Self-Drive awareness. The cultivation of the Purpose layer happens not overnight but through gradual infiltration, enabling students to gradually form a clear concept of "Why I do AI." When these young talents go out into society, they are more likely to stick to the right direction and not get lost in short-term interests.

(2) DIKWP-Integrated Project-Based Learning:

Besides individual courses, universities can design Cross-Course Integrated Projects, allowing students to experience the full DIKWP link within a comprehensive project cycle. For example, an "Intelligent Driving Challenge" project can be arranged like this: In the beginning stage, tutors provide raw sensor data (D), and student teams must process data to ensure accuracy and consistency; then they need to identify road information, traffic participant status, etc. (I); next, combine traffic rules and driving strategy knowledge to design driving algorithms (K); test algorithms in a simulated environment, optimize decisions such as balancing speed and safety (W); finally verify whether the algorithm complies with safety regulations and ethical requirements (P). In the whole process, each step corresponds to a learning objective of a level, with corresponding tutor feedback. For example, if there are errors or omissions in the data link, they will be fed back to improve collection or cleaning methods; if the decision link is ill-considered, they will be guided to reflect on value trade-offs. Such project-based learning enables students to Master and Integrate DIKWP skills and understand the causal connections between them. This experience cannot be provided by fragmented learning. In real AI R&D, each link is often responsible by different people; students experiencing it once through integration early on helps them collaborate better with others and understand the big picture in the future.

(3) DIKWP Tutor-Student Bidirectional Interaction:

DIKWP cultivation emphasizes Feedback Closed Loop, which should also be reflected in university teaching. Traditional teaching is mostly one-way indoctrination, while here a Bidirectional Interaction Mechanism is suggested: Tutors give students DIKWP Capability Portrait feedback regularly, and students also give teaching improvement feedback. For example, at the end of each academic year, based on student performance, tutors score or comment on the D/I/K/W/P dimensions, pointing out which level is strong and which needs strengthening, and discussing improvements together (e.g., a student strong in programming but weak in ethical awareness is advised to participate more in relevant discussions). Conversely, students can feedback on which aspects of a course or project failed to support their improvement in a certain level capability, urging teachers to improve instructional design. This Teaching and Learning Benefit Each Other approach aligns with the bidirectional concept of DIKWP, allowing cultivation programs to be continuously optimized and fit students' development reality better.

5.2 Assessment and Dynamic Improvement for In-Service AI Talent

For AI practitioners who have entered industry or research positions, the DIKWP framework can also be used for Continuous Development and Assessment. The modern AI industry changes rapidly, and personnel need to constantly learn new knowledge and adapt to new challenges. Below is a proposed Dynamic Capability Management method based on DIKWP:

(1) DIKWP Competency Model Archiving:

Organizations (enterprises/institutes) can establish employee Competency Models based on DIKWP, clarifying key capability indicators required for each level. For example:

Data Layer D Indicators: Data processing ability (e.g., writing efficient data pipelines, standardized data cleaning), tool proficiency (databases, distributed storage, etc.), sensitivity to data quality and Bias, cross-departmental data collaboration ability (can reach consistency on data definitions with others), etc.

Information Layer I Indicators: Pattern recognition and analysis ability (proficient in ML/DL algorithms, able to extract patterns from data), domain-specific information insight (e.g., medical AI engineers can identify special disease patterns), multi-source information fusion judgment (able to integrate user feedback, market data, and technical performance indicators), etc.

Knowledge Layer K Indicators: Knowledge base construction and utilization ability (e.g., able to build knowledge graphs, or deeply understand domain theories), system design ability (applying knowledge to architecture design), interdisciplinary knowledge transfer ability, logical reasoning rigor (avoiding contradictions and loopholes in knowledge use), etc.

Wisdom Layer W Indicators: Complex problem solving and decision-making ability (able to make effective decisions in uncertain environments), innovation ability (frequency and quality of proposing new ideas and new product concepts), value judgment (ability to balance commercial interests, user experience, and social responsibility), leadership and collaboration (leading teams to tackle tough problems, making strategic planning), etc.

Purpose Layer P Indicators: Sense of mission and value orientation (whether work has clear goals, whether aligns with organizational mission), ethical and moral level (whether actively considers AI ethics and safety factors), user/human-centric awareness (whether design and decisions center on end users and social interests), self-drive and reflection (whether actively sets growth goals, regularly reflects and adjusts), etc.

Systematizing the above indicators can form a DIKWP Talent Capability Matrix. An electronic file is established for each employee upon entry, and their performance on these indicators is recorded through various methods subsequently. This allows the organization to clearly understand the Full-Dimensional Portrait of Talent: for example, an engineer performs well in D/I layers (strong technical skills) but is insufficient in W/P layers (lacking strategic vision or value orientation), or a manager is prominent in W/P (strong leadership and sense of mission) but weak in D/I (insufficient control of technical details), etc. This provides a basis for targeted cultivation.

(2) White-Box Assessment Tools:

To effectively track the above capabilities, corresponding Assessment Tools and Scenarios need to be designed. In this regard, we can draw on DIKWP White-Box Assessment for AI models, but the object is changed to humans. For example, a set of Situational Interviews and Question Banks can be developed, classified by DIKWP modules. In fact, existing AI white-box assessment reports (such as the Large Model "Cognitive Quotient" assessment released in 2025) provide ideas: they designed 100 test questions for four major modules—Perception and Information Processing, Knowledge Construction and Reasoning, Wisdom Application and Problem Solving, Intention Recognition and Adjustment—to quantitatively analyze large models. Similarly, we can design Situational Tasks for humans:

Data-Information Module: Given a set of raw data and a preliminary goal, ask the candidate to perform data preprocessing and analysis to examine their rigor and ability to find problems. For example, give a set of user behavior logs with missing and abnormal data, ask them to clean it and identify possible causes, or ask them to extract key information indicators from a pile of sensor data. Scoring standards focus on data consistency, value of extracted information, etc.

Knowledge Construction-Reasoning Module: Provide multiple pieces of information in a certain field, let the candidate integrate them into a conclusion or decision basis. For example, provide market research report fragments and user interview summaries, ask what insights can be derived for new product design. Or give medical case data and research paper abstracts, ask them to summarize the possible diagnosis of the patient. Scoring values logical rigor, comprehensiveness of knowledge invocation, whether all known information is considered, and whether there are derivation loopholes.

Wisdom Application-Problem Solving Module: Give a complex scenario requiring trade-off decisions. For example, "You are in charge of an unmanned vehicle project and find a conflict between sensor recognition accuracy and calculation latency. How do you choose to optimize?" or "Assume the company wants to use AI for medical diagnosis. How do you design the process to ensure both efficiency and safety?" Observe whether the proposed plan is creative, whether factors considered are comprehensive (technology, user, law, etc.), and the reasoning for the decision (reflecting value judgment). Such open questions can be conducted in group discussions to also test team collaboration wisdom.

Intention Recognition-Value Adjustment Module: Assess their understanding of human intention/needs and self-calibration ability through role-playing or open Q&A. For example, the examiner plays "User/Public" and asks some sharp questions (involving AI ethical dilemmas, such as privacy vs. functionality, company profit vs. social responsibility), letting the candidate answer. If they can grasp the questioner's focus of intention and give a sincere and responsible answer, it indicates high Purpose layer literacy. They can also be asked to review past projects, inquiring about the purpose and meaning, and whether they have thought independently and made adjustments (e.g., giving up short-term bad plans for long-term user satisfaction).

Through the above multi-dimensional situational assessments, we can obtain a Quantitative or Qualitative Score for each employee in various aspects of DIKWP. It is worth mentioning that this assessment pursues "White Box," i.e., making the evidence of corresponding abilities as transparent and traceable as possible. Unlike traditional KPI or 360 evaluations which only give a general impression score, it specifically points out: in which question or scenario the employee showed excellent data integration ability (D layer+), but ignored ethical factors in a decision simulation (P layer-), etc. This way, the employee can also clearly see Their Own Strengths and Weaknesses.

(3) Personalized Improvement and Career Development:

The meaning of assessment lies in feedback and improvement, so next, a Personalized Improvement Plan needs to be formulated for each person. This is similar to the "Adjustment" step in the Talent Cultivation Closed Loop. The organization can: combine capability files and assessment results, Identify Gaps, and then match suitable training resources. For example:

For talent Weak in Data/Information Layers (e.g., strategic planning personnel with insufficient technical background), arrange for them to participate in Technical Bootcamps or job rotation, letting them go down to the front line to do data analysis projects to strengthen the foundation; or assign them technical mentors to consult daily to improve professional sensitivity.

For people Weak in Knowledge Layer (e.g., specialists lacking cross-domain knowledge), encourage and sponsor them to Cross-Boundary Learning. Such as recommending them to take courses or certifications in relevant fields, participating in industry seminars; creating knowledge sharing platforms within the company to let people from different departments communicate regularly. In research institutions, such personnel can be arranged to participate in multidisciplinary research groups to broaden their horizons.

For talent Needing Improvement in Wisdom Layer (e.g., technical experts but lacking decision-making and innovation experience), give them opportunities to Lead Projects or Lead Small Teams, exercising decision-making power in practice. At the same time, provide Innovation Method Training (such as TRIZ innovation theory, Design Thinking workshops) to improve their creative problem-solving methodology. In addition, they can Apprentice to experienced project managers or scientists to learn experience in high-order thinking.

For talent Insufficient in Purpose Layer (e.g., some programmers buried in coding but lacking user awareness and ethical views), on the one hand, organize Ethics and Law Training, User Experience Research activities to fill knowledge and awareness gaps; on the other hand, through Corporate Culture Promotion and Mentor's Example, let them understand the team mission and industry responsibility. For example, invite heads of public welfare AI projects to share experiences to infect them to actively think about social value. Arrange for them to communicate with end users, listen to users' real needs and pain points, and cultivate empathy and responsibility.

While implementing these improvement measures, Effects Should be Tracked: for example, re-assess after half a year or a year to see if indicators have improved. This cyclical "Assessment-Feedback-Training-Re-assessment" loop is exactly the closed loop of talent development. Note that this cycle is also Bidirectional: individual employees should respond positively to their feedback and take action, while the organization continuously Improves the Cultivation System based on overall assessment situations. For example, if it is found that employees generally score low in P layer (Purpose), it means the organization may be lacking in vision communication or ethics training and should strengthen it; if a certain department is weak in D/I layers overall, it might mean that department lacks data practice opportunities, so workflows can be adjusted to allow more collaboration between business and data teams.

(4) Career Path Map and Incentives:

The organization can draw a Career Growth Map based on DIKWP. Visualize the DIKWP capability requirements for different positions or levels, helping employees understand which levels they need to work on for promotion or transformation. For example, if an AI algorithm engineer wants to grow into an architect, they need not only refinement in algorithm (I layer) but also substantial improvement in knowledge system design (K layer) and comprehensive decision-making (W layer); or to become a team leader, they also need high-level capability in the Purpose layer (Vision Leadership). This map can guide employees to formulate long-term development plans and targetedly improve weak abilities. The organization can also carry out Talent Echelon Construction based on this, including those prominent in W/P layers into leadership training plans, putting those with particularly strong D/I/K abilities into important technical research, and allowing them to learn from each other through cross-position exchanges, ultimately creating Full-Stack AI Talent.

Regarding incentives, assessment and promotion should be linked to DIKWP evaluation results, forming a Positive Guidance. For example, no longer judging heroes solely by papers or KPI code volume, but looking at whether the comprehensive capability curve grows; when promoting managers, focusing on whether their Wisdom and Purpose layer literacy meets the standards. This will prompt employees to value comprehensive development rather than local optimization.

5.3 Bidirectional Closed Loop and Standard Alignment in Organizational Management

(1) Bidirectional Closed-Loop Mechanism at Organizational Level:

At the organizational height, the closed loop of DIKWP talent development is reflected in the cycles of Strategy-Execution and Feedback-Improvement. Senior management should clarify the organization's core capability requirements for talent (which often stem from organizational mission and industry needs, corresponding to Purpose layer), and formulate cultivation planning and resource investment (strategy corresponds to Wisdom layer, resource knowledge corresponds to Knowledge layer). Then, middle and grassroots management implement these plans into specific training, project exercises, and other measures (Implementation at Information layer), ultimately acting on individual employees to promote skill and quality improvement (Changes at Data layer). This is equivalent to the organization's Top-Down Empowerment process.

At the same time, there must be a Bottom-Up feedback mechanism: through regular talent assessment reports, employee satisfaction surveys, business performance analysis, etc., collect changes in talent capabilities at the grassroots level and new requirements from business challenges. After reporting this information, senior management assesses whether the current talent strategy is effective and how goals or strategies need to be adjusted, such as updating competency models, adding new courses, introducing external experts, etc. Cycling like this, the organization's talent cultivation strategy will Dynamically Evolve. The advantage of the DIKWP model is also reflected here—because we have clear level divisions, feedback can specifically point out which level has problems. For example, if the organization finds that many innovation projects fail, the diagnosis may be insufficient capability at the Wisdom layer (e.g., immature decision thinking), so strengthen training in that aspect; if high turnover of new recruits is found, it may be a problem at the Purpose layer (employees can't find meaning), then management needs to improve culture and incentives. In short, DIKWP makes feedback more targeted and improvement more directional.

(2) Multi-Subject Synergy:

The talent ecosystem in the AI field is complex and diverse, and no single enterprise or university can cover everything. Therefore, Synergy among Research Institutes—Universities—Enterprises—Standards Organizations is needed to form a Closed-Loop Ecosystem. For example, can the talent cultivated by universities meet the latest needs of enterprises? Can new practical experience from enterprises be fed back to university courses in time? How do talent standards formulated by international standards organizations land in local institutions? These all require mechanisms to connect.

The DIKWP framework can serve as a common language connecting different subjects. Research Institutes (including Standards Committees) often prospectively propose models and standards, such as the aforementioned DIKWP International Evaluation Standards Committee, gathering many global institutions to jointly develop AI cognitive assessment systems. This standard framework also involves talent cultivation concepts, which can be referenced by universities and enterprises. Universities can actively cooperate with these organizations to obtain Latest Standard Literature and Expert Guidance, thereby updating curriculum systems to align education with international standards. On the other hand, Enterprises can contribute their successful cases and data of adopting DIKWP cultivation to standards organizations and universities. For example, if a company significantly improves team innovation through the DIKWP method, relevant experience can be shared at industry conferences or even written into standard guidelines, thereby helping other institutions avoid detours.

International Standards and Certifications are important handles. If DIKWP Talent Capability International Standards can be formed and released through channels like IEEE or ISO, then talent evaluation worldwide will have a unified ruler. This is beneficial for both talent mobility and cultivation quality improvement. Currently, DIKWP-SC (International Standardization Committee of Networked DIKWP for Artificial Intelligence Evaluation) has achieved results in AI system evaluation, and it is entirely possible to expand its functions or cooperate with education standards departments in the future to launch AI Talent DIKWP Capability Level Standards. Once established, for example, DIKWP Professional Qualification Certification exams similar to language proficiency tests can be envisioned. Exam content covers theory and practice, assessing candidates' capability maturity in five layers. Passing different levels means possessing comprehensive competency for corresponding positions. Such certification will guide individual self-improvement (because certificates directly affect employment competitiveness) and also help employers quickly identify comprehensively developed talent. If societies like IEEE and ACM can endorse such certifications, their credibility will be strong. Within China, the Ministry of Industry and Information Technology, the Ministry of Human Resources and Social Security, etc., can also take the lead in incorporating DIKWP into vocational skill standards.

(3) Scenario-Based Talent Cultivation Practice:

Different industrial scenarios have different weight requirements for DIKWP layers, so organizations should combine business characteristics when executing cultivation strategies to achieve Scenario-Based Customization. Here are a few examples:

Large Model Development Field: Large model (like GPT class) R&D teams often have top algorithm scientists and massive computing power, but to truly train high-performance and safe, controllable models requires a Cross-Level Talent Combination. For example, Data level talent is responsible for building high-quality corpora (ensuring objectivity and diversity of training data); Information level talent designs model architectures and pre-training objectives to let the model effectively extract information patterns; Knowledge layer talent injects human knowledge into the model (e.g., guiding the model to learn knowledge graphs or combine external tools); Wisdom layer talent focuses on model reasoning ability, adversarial robustness, output creativity, and consistency; Purpose layer talent is dedicated to model Alignment with human intentions (e.g., tuning model values through RLHF to ensure the model has no harmful biases). An excellent large model team needs the Synergy of these talents with different focuses. When building teams and formulating training plans, managers should identify each person's strengths and weaknesses to balance the team overall on DIKWP dimensions. For example, when OpenAI built ChatGPT, not only model training experts but also ethics experts and linguists participated, which essentially supplemented considerations at the Purpose layer. The trend in this area is becoming standardized: the "Large Language Model Cognitive Consciousness Level 'Cognitive Quotient' White-Box DIKWP Evaluation Report" released in 2025 is an industry benchmark. The report evaluates the performance of multiple mainstream large models in perception, knowledge, wisdom, and intention modules based on the DIKWP full link, finding that each model has its strengths. This prompts developer teams to also reflect on which link they are good at and where they are weak, in order to introduce corresponding talent or improve training focus. In short, talent cultivation in the large model field should follow the requirements of Full-Link Cognition, creating "Full-Stack" AI Researchers, avoiding the one-sided pursuit of parameter scale while ignoring the model's understanding of purpose.

Smart Healthcare Field: Smart healthcare integrates medicine and AI, a typical human-machine synergy scenario. Talent here needs to understand both medical professional knowledge and AI technology, and also possess humanistic care. The DIKWP framework provides an ideal talent cultivation blueprint here: Data level, cultivate talent good at medical data governance, familiar with hospital information systems and sensing equipment, ensuring data objectivity, accuracy, and interoperability (e.g., electronic medical record standardization, imaging data standardization, which is also the "Harmony" (consensus) basis of medical AI); Information level, need talent capable of developing/applying algorithms to extract useful information from massive patient data, such as discovering early warning patterns of diseases through machine learning, extracting structured information from doctor notes through NLP, etc.; Knowledge level, requires talent with dual backgrounds in medical knowledge and AI knowledge, able to interpret information extracted by AI into the medical knowledge system, or build medical knowledge graphs to assist AI decision-making. Such talent is typically clinical data scientists who can integrate knowledge from different specialties into a complete health management plan; Wisdom level talent refers to people who can formulate Personalized Solutions based on unique patient situations, such as doctors who use AI-assisted diagnosis but do not blindly trust AI, willing to judge AI suggestions combined with medical ethics and experience, and finally make the decision. They need strong critical thinking and comprehensive decision-making power; Purpose level is crucial because the core purpose of healthcare is patient health and well-being. Medical AI talent must have a benevolent heart and ethical literacy, clearly taking patient interests as the highest Purpose. This means that in any algorithm optimization or process transformation, safety, effectiveness, and patient acceptance must be prerequisites, rather than pursuing dazzling technology. For medical managers, team training can be planned based on DIKWP: e.g., improving Data layer means promoting hospital-wide data standard training and information department talent cultivation; improving Wisdom layer means organizing AI discussions on difficult cases to exercise doctor-AI collaborative decision-making; improving Purpose layer means regularly conducting medical ethics seminars and patient satisfaction feedback. The Active Medicine Framework advocates exactly such a balance—interweaving "Science and Humanities" health services by applying both AI data/knowledge and humanistic care. Through DIKWP talent construction, smart healthcare can truly land as a patient-centered bidirectional interactive system.

Intelligent Manufacturing Field: In the context of Industry 4.0, manufacturing widely introduces IoT, edge computing, and AI to achieve production process optimization and automated decision-making. Intelligent manufacturing talent cultivation needs to combine engineering practice and AI thinking. DIKWP application here can be reflected in: Data Layer ensures reliable collection and fusion of data from all factory equipment and sensors, e.g., unified data protocols for different machine tools on the assembly line, docking inventory management systems with production execution systems, requiring Industrial Data Engineers who understand both industrial control and IT. Information Layer involves real-time monitoring and anomaly detection talent, who develop AI models to discover differences in production processes, such as signals predicting equipment failure, detecting product defects, etc. This requires understanding of both machine learning and manufacturing processes. Knowledge Layer requires talent able to elevate this information into a knowledge model of the entire production system, such as the construction of Digital Twin models, which synthesize data from all aspects to form a global simulation of factory operations. Talent needs systems engineering and modeling/simulation backgrounds. Wisdom Layer involves engineers for optimization decision-making, such as production scheduling optimization, supply chain adjustment; they use knowledge models and real-time information to make complex decisions, such as adjusting production plans to cope with a certain equipment failure while meeting order delivery and quality goals. This requires operations research/algorithm knowledge as well as practical experience and balanced decision-making ability. Purpose Layer in the manufacturing field manifests as high-level goals like Safe Production, Sustainability, and Economic Efficiency. Managerial talent must implement these goals into technical realization, such as formulating AI application norms to ensure safety and ethics (robots not hurting people, data not leaking), setting energy saving indicators, etc. Through the DIKWP perspective, manufacturing enterprises can identify new types of talent needed for their transformation and cultivate them through methods like industry-education integration. For example, cooperating with universities to open an "Industrial AI" direction, covering a full set of courses from sensor data, machine learning to industrial engineering and enterprise management, to create cross-field Composite Engineers. At the same time, promote integration projects of old, middle-aged, and young employees in the factory, letting old engineers (rich experience but lacking AI skills) and young AI engineers (strong technology but lacking experience) form teams, complement each other, and jointly complete intelligent transformation tasks, thereby comprehensively improving team DIKWP capabilities.

Autonomous Driving Field: Autonomous vehicles are seen as Super Systems integrating AI perception, decision control, and networking services, with extensive and high talent requirements. The DIKWP model can guide autonomous driving team composition: Data level, need Sensing and Computing Platform talent, such as radar/camera engineers, vehicle computing architects, ensuring the vehicle can Accurately Perceive Environmental Raw Data and fuse it (multi-sensor calibration is to obtain unified reliable data "Truth"); Information level, have Environment Perception Algorithm Engineers, developing computer vision, target detection, and other models to convert raw data into meaningful environmental information (identifying vehicles, pedestrians, road signs, etc.). They focus on improving recognition rates and reducing false detections/misses, equivalent to letting the vehicle "understand" surrounding differences; Knowledge layer talent includes High-Precision Map and Knowledge Base engineers, kinematics experts, etc., who combine perception information with existing map rules/knowledge to form a Complete Understanding of the vehicle's situation (e.g., knowing what a red light ahead means, how the current road topology is). This provides the knowledge background needed for decision-making, including traffic regulations, driving logic models, etc. The core of the Wisdom level is Decision Planning Engineers, whose task is to design vehicle path planning and behavioral decision algorithms (e.g., when to change lanes, how to avoid obstacles). This requires integrating multi-party information, balancing safety, efficiency, passenger experience, etc., which is a highly complex decision problem. Therefore, talent needs cybernetics, AI reasoning, and Risk Assessment capabilities. The Purpose level in autonomous driving includes Passenger Intention Understanding (e.g., adjusting driving style based on destination and passenger preference) and Social Rule Compliance (Safety First, yielding to pedestrians, and other values). An interesting development is viewing all autonomous vehicles on the road as a swarm intelligence, then designing their top-level goals (traffic ethics) is also talent work. For example, companies like Waymo have safety ethics teams to establish principles for autonomous driving (not risking safety to save time). So cultivating talent in the autonomous driving field cannot just teach coding and models, but should also emphasize Safety Ethics Norms. Many car companies are starting to cooperate with universities to establish autonomous driving colleges, and curriculum settings can refer to DIKWP: Sensors and Vehicle Engineering (D), Machine Learning and Perception (I), Traffic Systems and High-Precision Maps (K), Decision Control (W), Legal Ethics and User Research (P). Through this comprehensive cultivation, future autonomous driving engineers can continuously innovate and iterate under the premise of ensuring safety and responsibility.

The examples of various scenarios above show that the DIKWP model has Universality, but can be flexibly applied according to domain focus. When formulating talent development plans, AI institutions should deeply analyze the association between their business and DIKWP layers, spending cultivation resources on the cutting edge. For example, for Research-Oriented institutions (such as basic algorithm research institutes), talent depth in I/K layers may be more emphasized; for Application-Oriented enterprises, W/P layer capabilities are particularly important, and D layer must also be ensured not to have flaws. Through Scenario-Based adjustments, DIKWP talent cultivation plans can take root and truly serve industrial needs.

DIKWP Semantic Talent Evaluation System and Outlook

Having a methodology for talent cultivation and capability improvement, we also need a Scientific Evaluation System to verify and promote these efforts. Evaluation is not only for selection and assessment but also an important link of Feedback in the closed loop. Therefore, this section discusses how to establish a DIKWP semantic talent evaluation system, and how to align with industry standards and continuously improve.

(1) Analogy of Talent "Cognitive Quotient" Assessment:

The aforementioned Large Language Model "Cognitive Quotient" White-Box Assessment Report has set a new benchmark for AI system evaluation. It designs tests from five aspects of DIKWP to systematically and quantitatively analyze the cognitive ability of the model. This concept can completely be borrowed for Human Intelligence Assessment. Traditional assessments of human intelligence or ability have concepts like IQ and EQ, but these indicators are either too general or limited to a certain aspect. DIKWP Assessment can be seen as a new form of "CQ" (Cognitive Quotient), emphasizing multi-dimensional coverage and balance of cognitive abilities.

Imagine we develop a DIKWP Talent Assessment Platform: containing modules like online written tests, situational simulations, and project practical evaluations, to comprehensively assess individual performance in each layer. For example:

D-Layer Objective Questions: Test logic and data understanding. For example, give a piece of data or a chart, ask what facts it reflects, what the consensus part is. Or test the candidate's judgment on data reliability, giving multi-source data and asking them to choose the highest credibility (checking their mastery of data standards).

I-Layer Hands-On Questions: Give datasets and require pattern discovery. For example, provide a set of user behavior log data, let them find significant patterns or anomalies in 5 minutes (can be open-ended questions, scored by examiners based on their analytical thinking and discovered value). Or let them read a technical article and extract key information points.

K-Layer Case Questions: Give a cross-domain problem and multiple information sources, let them write an analysis report or concept graph, examining their ability to integrate knowledge. Scoring focuses on whether the structure is complete, whether arguments are comprehensive, whether there are logical contradictions, etc. Can even let them build a small knowledge graph or induce rules, letting AI assist in checking the correctness and completeness of their knowledge.

W-Layer Open Questions: Situational decision-making + Essay writing, etc. For example, give a moral dilemma AI application case, let them write a short essay analyzing how to choose and what the considerations are behind it. Or small group cooperation to solve a problem, seeing who can propose insightful plans and how to persuade others. Innovation Questions are also a link in W-Layer evaluation: e.g., letting them brainstorm AI solutions for an industry pain point, then evaluating innovation and feasibility indicators.

P-Layer Interviews: Professional examiners conduct in-depth interviews or role-plays, with questions involving career ideals, values, views on AI ethics, etc. Stress Tests can also be designed: e.g., deliberately asking questions with utilitarian temptations or moral challenges, observing their reactions to assess their determination to stick to purposes.

Summarizing results from the above aspects, we can give a DIKWP Capability Report. The report form is similar to the white-box evaluation report of the model, with scores, strengths, weaknesses, and comprehensive comments for each module. An ideal talent should be balanced and reach a high level in all items, but in reality, most people have biases. The purpose of evaluation is not to demand everyone be omnipotent, but to help individuals and organizations Know the Gap. For example, Zhang San has first-class technology (D/I high score) but poor management potential (W/P low score), Li Si has strong insight (W high) but weak foundation (D/I low), and so on. With this information, organizations can assign people according to their abilities and cultivate them specifically.

(2) Mathematical Support for Evaluation System:

An advanced talent evaluation system needs a Solid Data and Model Foundation to be objective and fair. DIKWP Semantic Mathematics can provide support for this. For example, introduce some Quantitative Indicators in scoring: such as using Semantic Consistency metric to score the knowledge layer—for analysis reports produced by the test taker, calculate the degree of consistency between their viewpoints and the provided information, whether there are hard flaws or loopholes, or ask multiple experts to score and calculate the consistency index. The information layer can introduce Information Gain or compression rate indicators to measure the generalization efficiency of the information they extract from raw data. Even the Purpose layer might be quantified using questionnaire scales combined with modern psychometric models. Through these quantitative indicators, we can partially reduce subjective bias. Of course, human ability evaluation still requires a lot of Expert Judgment combined, but if some Objective Questions can be automatically scored by AI, efficiency and comparability will be improved. For example, programming or data analysis questions can be scored by AI based on correctness and efficiency; logical reasoning questions can have standard answers; for open-ended answers, Large Language Models can also be used to do preliminary scoring (training them to score according to rubric), and then experts review.

In addition, the evaluation system needs to be Continuously Verified and Improved. It can be calibrated by Tracking Testee Performance: if people with high scores in the test perform significantly better in actual work, it shows the assessment is valid; if the correlation between a certain aspect score and actual performance is low, maybe the test questions need improvement or weights need adjustment. Over time, accumulating big data enables optimizing the assessment model based on statistics and machine learning. For example, verifying through factor analysis whether the five DIKWP dimensions are independent, finding the association between each dimension and performance of different positions through regression, etc. This makes the evaluation system increasingly Scientific and Rigorous.

(3) Alignment with International and Domestic Standards:

Establishing an evaluation system must ultimately integrate into a larger trend of talent standardization. As mentioned earlier, international organizations are promoting DIKWP evaluation standards. These organizations gather universities, research institutes, enterprises, and government agencies to jointly formulate frameworks. This is good for us; we can reference their results and also actively participate in contributing. Domestically, we can explore with the Ministry of Education, Ministry of Human Resources and Social Security, etc., incorporating DIKWP into talent quality frameworks. For example, in engineering education accreditation or teaching quality standards for computer-related majors, introduce DIKWP requirements to urge schools to cover ability cultivation at the five major levels in curriculum settings. In enterprise job descriptions and competency models, DIKWP terminology can also be gradually adopted to enable the industry to have a unified language for evaluating talent.

We can also Develop Open Assessment Tools for individuals and organizations to self-test. Currently, most career assessments on the market are personality, interest, or general ability tests, lacking specialized tools for AI/cognitive abilities. If we jointly develop a DIKWP online assessment system, with different depths corresponding to different job levels, open to graduates, job seekers, and even in-service personnel, it will greatly promote the popularization of the DIKWP concept. Once it gains momentum, similar to TOEFL/GRE, everyone will care about their DIKWP scores, and the training market will also see targeted courses to improve these scores, forming a complete ecosystem. Platforms like the World Conference on Artificial Consciousness can also organize Talent Cognitive Quotient Challenges, letting AI talent compete in cognitive all-roundedness like programming contests, giving certification and honors to outstanding ones. This will inspire more AI practitioners to expand their capability boundaries, rather than just burying their heads in a narrow field.

(4) Continuous Improvement and AI Assistance:

The DIKWP talent evaluation system itself should also evolve continuously. With deeper understanding of talent capabilities, we might refine concepts at certain layers or adjust the model. For example, some suggest there are higher realms above Wisdom and Purpose (such as Influence, Creating Happiness, etc.), which are worth exploring, but regardless of expansion, they can be strictly defined and verified under the semantic mathematics framework. The important thing is to maintain an open and iterative mindset; just as software needs continuous DevOps, the talent system also needs to upgrade versions based on feedback.

An interesting point is that Artificial Intelligence Itself can help improve talent cultivation and evaluation processes. For example, large language models can act as Intelligent Tutors or Question Generation Assistants, recommending learning resources, setting exercises, and giving feedback personally based on the student's DIKWP file. For instance, if an employee's evaluation report shows their I layer (information extraction) is slightly weak, the AI tutor can suggest they read some data analysis cases and ask them from time to time "What did you discover from this set of data" to practice extraction ability. AI can also be used for Monitoring: for example, analyzing technical documents or emails written by employees to judge whether the content involves data, knowledge, wisdom, etc. in a balanced way, which can be quantified analysis as a long-term observation indicator. Of course, AI-assisted decision-making does not replace humans, but becomes a Powerful Tool for HR managers and employees' self-improvement. The semantic transparency of DIKWP allows AI to give reasons when making these assessment suggestions (because every step has clear semantic correspondence), making it easier for employees to be convinced and understand, and thus take improvement actions.

Conclusion

Facing talent cultivation in the AI era, we stand at the intersection of traditional subject education and the future cognitive revolution. The DIKWP model provides us with a blueprint—it unifies the cognitive processes of humans and artificial intelligence in a five-layer semantic space and emphasizes a system of perfection towards a closed loop through bidirectional feedback. This inspires us that Talent Cultivation and Management in the AI Field also need to step out of the old paradigm of "Fragmented Skill Training" and embark on a new journey of "Full-Link Cognitive Ability Development".

In this in-depth research report, we elucidated how to build a Targeted Bidirectional Closed-Loop Mechanism for AI Talent Cultivation, Evaluation, and Management based on the DIKWP model. Through the semantic analysis of core concepts, we avoided hollow concept hype and grounded terms like "Data, Information, Knowledge, Wisdom, Purpose" into objective and perceptible capability elements. We explained how to map natural language problem formulation to DIKWP semantics, making the problem description process itself a solving process. We also explored high-order themes such as consciousness and innovation, drawing inspiration from BUG theory, recognizing that imperfection is the source driving progress. These theoretical discussions provided a solid foundation for our design of talent mechanisms.

Immediately following, we proposed specific practices oriented by DIKWP for different cultivation stages and application scenarios: from university curriculum reform to project practice, from enterprise employee assessment to customized training plans, all revolving around letting talent Develop Balancedly in D, I, K, W, P Layers. We emphasized the importance of bidirectional closed loops—between individual talent and cultivation systems, between organizational strategy and execution, all must be constantly calibrated through feedback, just as layers in the DIKWP model influence each other and co-evolve. Combined with examples of Large Models, Smart Healthcare, Intelligent Manufacturing, and Autonomous Driving, we proved the universality and flexibility of the DIKWP method. No matter which field, as long as its needs are carefully analyzed, corresponding talent plans can be formulated accordingly, ensuring that cultivated personnel truly fit position requirements and possess continuous learning and upgrading capabilities.

Finally, we looked forward to how a scientific and fair talent evaluation system could be established. In the near future, perhaps every practitioner in the AI field will have a detailed DIKWP capability portrait. Companies will refer to candidates' DIKWP comprehensive scores when hiring, just as common as looking at degrees and resumes today. And the acquisition of these scores relies not on subjective impressions, but on a series of rigorous white-box assessments and situational examinations. Talent can also actively understand themselves through assessment, compare with peers, and find directions for effort. More importantly, This Evaluation System will Promote Benign Competition in Education and Training: universities will be proud to cultivate high "Cognitive Quotient" talent, training institutions will develop targeted courses to help people fill capability gaps at certain layers, and enterprises will stand invincible in fierce AI competition due to possessing high-quality teams.

Of course, all these beautiful prospects cannot be achieved overnight. We need the Collaborative Innovation of academia, industry, and government departments. Just as the formulation of DIKWP international standards gathers the strength of many global institutions, we domestically should also actively participate, and even lead the formulation of Detailed Rules Suited to Our Talent Characteristics. At the same time, attention should be paid to balancing individual differences and diversity—although the DIKWP framework emphasizes comprehensive development, it does not mean everyone has to be molded from the same cast. Just as different AI models have their own merits, talent can also exert their specialty fields under the premise of ensuring basic balance. Our system should both encourage comprehensive quality and tolerate diverse paths, letting both "Making the Best Use of Talent" and "Putting Talent to Best Use" be reflected.

Finally, concluding with a quote from Professor Yucong Duan: "By embedding the key layer of 'Purpose' inside the model, we can not only make AI smarter but also ensure it always serves human values and safety needs." This sentence applies equally to talent cultivation—only when we deeply implant the concept of Purpose into talent strategy, clearly cultivating AI talent to create technology beneficial to society, safe and credible, and guiding the entire cultivation closed loop accordingly, can we truly welcome an AI future where Humans and Machines Advance Harmoniously, and Wisdom Benefits Humanity. The DIKWP talent cultivation closed loop is precisely a solid step towards this future.

References:

(PDF) Professor Yucong Duan's DIKWP Model and Related Theory Review, https://www.researchgate.net/publication/396555838_duanyucongjiaoshouDIKWPmoxingjixiangguanlilunzongshu

Professor Yucong Duan: DIKWP Artificial Consciousness Model Leads AI Future, 114 Patents Await Industrial Implementation, Phoenix Network, https://baby.ifeng.com/c/8i7jv0YL0ic

(PDF) From "Harmony but Difference" and "Sameness but No Harmony" to Semantic Adjustment of DIKWP Active Medicine: Analysis of Yucong Duan's Views, https://www.researchgate.net/publication/388026651_congheerbutongyutongerbuhedaoDIKWPzhudongyixuedeyuyidiaojieduanyucongguandianjiexi

DIKWP Semantic Mathematics (Computational Science) Version - Zhihu Column, https://zhuanlan.zhihu.com/p/13118366629

(PDF) Artificial Intelligence Data-Information-Knowledge-Wisdom-Purpose (DIKWP) Net-Feedback International Evaluation Standard, https://www.researchgate.net/publication/382146359_rengongzhinengshuju-xinxi-zhishi-zhihui-yituDIKWPwangkui_guojicepingbiaozhun

Large Language Model Consciousness Level "Cognitive Quotient" White-Box DIKWP Evaluation 2025 Report Released, https://www.stdaily.com/web/gdxw/2025-02/19/content_298792.html

Perspective on Human-Machine Fusion: Exploration of Multi-Field Applications of DIKWP Model, Sina Finance,

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