1、Background and significance of the project

The current mainstream paradigm of artificial intelligence (AI) mainly relies on deep learning models driven by big data, and has achieved great success in the field of perception and pattern recognition. However, this kind of "data-driven" approach with relevance learning at its core also faces significant limitations. For example, large-scale pretrained models (LLMs) are capable of generating fluent text and complex responses, but their internal inference processes are like a "black box" and lack interpretability and autonomy. Models often lack self-awareness and purpose orientation, and are unable to truly understand semantics, let alone set goals or review their own decisions. As a result, current AI systems have encountered a bottleneck on the road to general intelligence (AGI): on the one hand, it is difficult to ensure that decisions are always in line with human values, and on the other hand, they are prone to hallucinations, biases, or uncontrollable behaviors in an open environment. It can be said that the existing paradigm lacks the description of the higher-order cognitive and consciousness mechanisms of the human brain, and cannot meet the needs of the next generation of intelligent development.

In order to break through the above bottlenecks, the academic community and the industry have begun to realize [the necessity of the development of brain-like intelligence]{.underline}, that is, to reconstruct the AI system by drawing on the cognitive architecture and consciousness mechanism of the human brain, so as to achieve a qualitative change from "being able to perceive" to "being able to know" and "understanding to reflect". This trend has been hailed as a revolutionary shift from traditional artificial intelligence to "intelligent self-knowledge". In this context, the DIKWP model proposed by Professor Duan Yucong points out a new direction for the development of artificial intelligence. DIKWP stands for Data-Information-Knowledge-Wisdom-Purpose, and adds the highest level of "intention/purpose" to the classical DIKW (pyramid) system. This improvement enables the model to form a semantic closed loop from the bottom perception to the high-level purpose, and supports two-way feedback and iterative update of semantics at all levels. The DIKWP model essentially introduces a "purpose-driven" cognitive framework: the data and information layer processes external perception and pattern recognition, the knowledge and intelligence layer realizes understanding and reasoning, and the purpose layer acts as the supreme conductor to provide goal constraints and value guidance. This new cognitive system is an academic milestone, providing an innovative way to solve the "black box" problem of current large models and improve the interpretability and controllability of AI systems.

More importantly, after the DIKWP model introduces "intent", it gives the AI system a common cognitive language, so that every step of the AI decision-making process can be traced and understood by humans. By embedding "purpose" into the model, AI decisions are no longer based solely on data relevance, but always revolve around a clear goal. Not only does this make AI "smarter" and more autonomous to plan and reflect, but it also ensures that it always serves human values and safety needs. It can be considered that the Artificial Consciousness (AC) system is a key support for the realization of AGI through this purpose-driven cognitive architecture that gives AI a human-like ability to self-regulate and reflect. At the theoretical level, consciousness is considered by many scholars to be an important part of general intelligence, and a self-aware system is more likely to have the ability to understand, create, and learn independently. This is reflected in the DIKWP model: the purpose system at the top of the model is equivalent to the "will" and "self" of the AI, providing the internal driving force for the continuous evolution of the system.

The theoretical support role of artificial consciousness system on AGI can be seen from this: by introducing an architecture similar to human consciousness, AI will have the ability of global planning, active learning, self-correction, etc., so that it will evolve from a passive tool to an active agent. Professor Duan Yucong's research shows that an artificial consciousness system can be seen as a combination of "subconscious mind system (LLM) + consciousness system (DIKWP)". Among them, large models such as LLM assume pattern matching and associative functions similar to the human subconscious, and the DIKWP system is like a "waking consciousness" responsible for high-level decision-making and purpose management. These two parts work together to make AI both possessive of a wealth of knowledge and experience (subconscious) and self-reflective and purpose-oriented (conscious). This brain-like dual-system architecture is considered to be an effective way to achieve strong artificial intelligence, which not only theoretically provides a new perspective for simulating human consciousness, but also provides a potential solution to solve the challenges of semantic bias and value alignment in AI systems.

In summary, this project is based on the DIKWP model and artificial consciousness theory, aiming to break through the limitations of the current AI paradigm and explore a new generation of model system of brain-like intelligence. By **organically integrating the five layers of data-information-knowledge-intelligence-purpose and introducing a purpose-driven artificial consciousness architecture, the project hopes to build a brain-like model with a prototype of self-awareness. This model will theoretically enrich the simulation of cognition and consciousness by artificial intelligence, improve the interpretability and autonomy of AI systems in engineering, and lay the foundation for the realization of general artificial intelligence in application. This is not only of great significance for solving the current "big model black box" dilemma, but also will promote AI from "intelligence" to "intelligence", and take a key step towards strong artificial intelligence (**AGI).

2、A review of the current research status at home and abroad

With the continuous revelation of the working mechanism of the human brain, brain-like intelligence has become one of the cutting-edge hotspots of global artificial intelligence research. Abroad, as early as the 2010s, IBM and other companies launched brain-like chips and cognitive computing programs, such as IBM TrueNorth synaptic chips trying to simulate human brain neuron connections from the hardware level to achieve low-energy intelligent processing. Academia has also developed concepts such as Global Workspace Theory ([GWT]{.underline}) and Integrated Information Theory ([IIT]{.underline}). ) and other theories of consciousness to explore the possible ways to achieve machine consciousness. These theoretical models explain consciousness from the perspectives of cognitive psychology and neural information integration, respectively: GWT sees consciousness as a broadcast "blackboard" in the brain, where key information is broadcast in the global workspace for different modules to access; IIT, on the other hand, attempts to quantitatively measure the degree of consciousness of a system using information entropy and causality. However, most of these theories of consciousness belong to the category of cognitive science and philosophy, and they are still preliminary in terms of engineering implementation. A large number of foreign teams have tried to integrate ideas such as GWT into AI architectures, such as the LIDA cognitive architecture proposed by Stan Franklin et al., which simulates the cycle of consciousness and attention, and [institutions such as DeepMind]{.underline} have also begun to pay attention to the Free Energy Principle ([FEP]{.underline}). ) in cognitive intelligence. However, in general, AI systems with real autonomous consciousness have not yet emerged, and artificial consciousness is still in the exploration stage from theory to engineering.

At the same time, large model (pre-trained model) technology is developing rapidly around the world, such as [OpenAI]{.underline}GPT series, Google's BERT/PaLM and DeepMind's Gato, etc. These models have demonstrated unprecedented performance in general-purpose language understanding and multimodal processing, and are seen in part as a promise for AGI. However, large models also expose problems such as insufficient knowledge and reasoning ability, and lack of goal drive. On the one hand, the traditional large model mainly obtains correlation mapping through massive data training, lacks explicit knowledge representation and logical reasoning mechanism, and is prone to hallucinations that are inconsistent with facts。 On the other hand, they are not embedded in long-term goals or value systems of their own, and need to rely on training goals provided by humans or later artificial feedback adjustments (e.g., RLHF). As a result, large models have limited capabilities in complex decision-making and long-term planning, as well as hidden dangers in value alignment and security. In recent years, some explorations have begun to emerge internationally to enhance the cognitive ability and controllability of large models. For example, Microsoft et al. have tried to integrate symbolic logic and knowledge graphs into pre-trained models to improve reasoning and common sense skills. OpenAI and others have introduced the use of tools and chain-of-thought prompts to enable the model to simulate the step-by-step reasoning process when solving problems. However, these improvements are still peripheral enhancements, and the model has not been refactored from the architectural level, which cannot fundamentally solve the inherent limitations of the "lack of self-awareness" of the large model.

In this context, domestic scientific research forces are also actively exploring new paths for artificial consciousness and brain-like models. In general, the research layout of brain-inspired intelligence in China mainly includes: brain-inspired chips and computing architectures for simulating neural computing (such as the brain simulation engineering of the Institute of Brain Research of the Chinese Academy of Sciences), cognitive and neural computing models (such as the work of Tsinghua Brain-inspired Computing Center), and cognitive robots and autonomous systems. However, compared with Europe and the United States, China started late in the theory of "artificial consciousness" and its practical exploration. However, what is remarkable is that the DIKWP artificial consciousness model proposed by Professor Duan Yucong's team provides a unique innovative starting point for related research in China. Professor Duan Yucong took the lead in extending the classical DIKW model into the DIKWP framework and applied it to the artificial consciousness system to construct a network cognitive structure. This work has enabled China to form a theoretical system of independent innovation in the field of artificial consciousness, and has laid a first-mover advantage through the accumulation of a series of patents. According to reports, Professor Duan Yucong is the first inventor of 114 authorized invention patents, covering many cutting-edge directions from large model training, artificial consciousness construction, cognitive operating system to AI governance and privacy security. Although these patented technologies have not yet been industrialized on a large scale, they have attracted extensive attention in the research on AI interpretability, security and value alignment, and are regarded as important "underlying codes" to lead the future.

Specifically, Duan Yucong's team has made a number of exploratory achievements in the engineering implementation of the artificial consciousness model. For example, they proposed the concept of DIKWP ×DIKWP dual-loop architecture, that is, introducing a metacognitive loop in addition to the basic cognitive process for self-monitoring and self-reflection. This architecture is similar to the human thinking process of thinking before thinking (i.e., "conscious consciousness"), which is considered to be the key path to give AI initial self-awareness, and represents a new direction for AI to move towards autonomous consciousness. In addition, the team embedded the DIKWP model into the large language model and proposed a plan to build a "semantic operating system", decomposing the LLM inference process into five monitorable links: data, information, knowledge, wisdom, and intent. Each link has a clear mathematical definition and semantic representation to ensure that the AI decision-making process can be checked step by step, thereby greatly improving the security and controllability of the system. This innovation is also unique in the world: in contrast, OpenAI has a small number of patents, while Google and Huawei, although they have rich patent reserves in other fields of AI, they have a relatively fragmented layout in cognitive structure innovation, and DIKWP-related patent portfolios are known for being systematic and innovative, and have a high voice in the world.

The current status of international research shows that artificial consciousness and brain-like intelligence are in the stage of moving from concept to prototype, and various theoretical frameworks (such as GWT, IIT, FEP, DIKWP, etc.) have their own emphasis. By introducing an intent-driven mesh cognitive architecture, the DIKWP model in China theoretically integrates data-knowledge processing and purpose-oriented, and preliminarily realizes the artificial consciousness prototype system DIKWP-AC with small model and low computing power in practice. According to reports, in 2023, Duan Yucong's team at Hainan University developed the world's first small-model, low-computing power, and explainable artificial consciousness software prototype system DIKWP-AC, which is divided into "mathematical subsystem" and "physiological subsystem".Two parts. The mathematical subsystem is responsible for internal semantic reasoning, and the physiological subsystem is responsible for interacting with the external environment, so as to simulate the combination of cognition and body. The team collaborated with Hainan Provincial People's Hospital to apply DIKWP-AC to the diagnosis and treatment of gout, lupus and other diseases to verify the effectiveness of the artificial awareness model in complex decision-making tasks. In general, the research on the combination of large models and cognitive architecture at home and abroad is in the ascendant, and the DIKWP model, as an artificial consciousness framework independently proposed by China, has been at the forefront of exploration, providing valuable experience and theoretical reserves for the construction of a new generation of brain-like intelligence.

3、Research objectives and content

The overall goal of this project is to build a brain-like large model system with preliminary artificial consciousness characteristics based on the DIKWP model, and break through the limitations of current AI in terms of cognitive ability and autonomous consciousness. Specifically, we will design the corresponding five core subsystems of data collection, information fusion, knowledge modeling, intelligence evolution, and purpose control around the five-layer mapping relationship of DIKWP, and make them organically integrated to form a hierarchical and progressive cognitive system with network interaction. The design objectives and contents of each subsystem are summarized as follows:

Data acquisition system: Establish a perception layer that simulates biological sensation to realize the acquisition and preprocessing of multi-source heterogeneous data. The system will integrate multi-modal sensing and data filtering mechanisms, including the collection, annotation and preliminary processing of multi-modal data such as vision, hearing, and text. The function of the human primary sensory cortex is simulated by neural computerization, and the key features in the original data are extracted to provide input for the information layer. The goal of this system is to efficiently extract useful information and ensure data quality in massive real-time data, and measurable indicators include multi-modal data alignment accuracy, noise filtering effect, and real-time data processing throughput.

Information fusion system: realize multi-modal fusion and semantic extraction of perceptual data, and simulate the unified representation process of sensory information by the human brain. The system will use the Transformer architecture combined with the explicit semantic coding of the DIKWP information layer to perform pattern recognition, classification and preliminary semantic understanding of the features from the data layer. The core research contents include cross-modal feature fusion, context-sensitive information association, and structured representation. The goal is to establish associations between information from different modalities and sources to form a preliminary cognitive representation of the environment and tasks. Metrics can include consistency of multimodal semantic representations, accuracy of information extraction, and improved performance in downstream cognitive tasks.

Knowledge modeling system: Construct a knowledge representation and reasoning mechanism based on the DIKWP knowledge layer, and endow the model with long-term memory and logical reasoning ability. We will integrate knowledge graph, symbolic logic, and deep learning to achieve interpretable knowledge storage and deduction. The system is responsible for mapping the semantic data output by the information layer into structured knowledge representations (such as graph nodes and relationships), and generating new knowledge or decision-making basis through inference algorithms (deductive reasoning, inductive reasoning, etc.). His research interests include knowledge acquisition and updating (automatic extraction of new knowledge from data and contradiction resolution), knowledge representation learning (combination of vector representation and symbolic representation), and inference engine (combination of rule reasoning and probabilistic reasoning). The indicators of the system can be evaluated by the coverage of the knowledge base, the accuracy of reasoning, the length of the reasoning chain, etc., and the goal is to achieve the in-depth understanding and application of domain knowledge by the machine.

Intelligent Evolution System: Realizing high-order cognition and adaptive evolution capabilities in the DIKWP intelligence layer is equivalent to simulating the formation process of human intelligence. The system will comprehensively use the information provided by the knowledge layer to carry out decision planning, strategy optimization, and experience learning, so that AI can continuously improve its performance in a dynamic environment. His research focuses on complex decision-making (decision-making algorithms that take into account multiple factors and uncertainties), continuous learning (task drills and feedback reinforcement to optimize strategies), ethics, and value judgments (Introduce value constraints and security considerations in decision-making). A key feature of intelligent evolutionary systems is the ability to be self-reflective—the ability to assess the consequences of one's own decisions and adjust cognitive strategies when needed. The performance of the system can be measured by the rate of task completion in complex environments, the assessment of the quality of decisions (e.g., compliance with expected goals and safety specifications), and the speed of adaptation under changing conditions. The goal is for AI to show human-like flexibility and intelligence, and to be able to draw inferences from unknown tasks and continuously improve.

Purpose control system: As the implementation of the DIKWP purpose layer, it is the top-level control center and power source of the whole model. The system is responsible for generating, maintaining, and adjusting the AI's internal goals (intentions), and regulating the lower layers through two-way feedback. Research topics include: purpose expression and understanding (designing an internal goal representation language for AI so that it can represent short-term task goals and long-term goals), purpose generation mechanisms (dynamically generating or adjusting intentions according to external needs and internal states, simulating human characteristics such as curiosity-driven and autonomous goal setting), purpose assessment, and conflict management(Ability to assess priorities and make trade-offs when multiple intents coexist or conflict with environmental constraints). The purpose control system will also implement metacognitive functions, monitor the state of AI in the process of achieving goals, and trigger a self-correction process once it deviates from the goal or has an abnormality (which corresponds to the metacognitive cycle in the aforementioned dual-loop architecture). Metrics to measure the system include the rationality and diversity of purpose generation, the goal achievement rate, and the system's ability to correct purpose biases. The ultimate goal is for AI to be continuously self-motivated, able to autonomously adjust behavior according to changes in the internal and external environment, so as to better complete the overall task.

In order to ensure that the above five subsystems work together, a clear technical roadmap will be developed: first, the formal semantics and mutual mapping of the various levels will be defined theoretically, and then the engineering implementation will gradually integrate the systems using a modular architecture. In the initial stage, we will focus on the integration of the data/information layer and the modeling of the knowledge layer, so as to lay a good foundation for high-level wisdom and intention. In the medium term, the wisdom evolution mechanism is introduced to realize the leap from knowledge to wisdom; In the later stage, we will focus on the emergence of purpose regulation and self-consciousness, and build an artificial consciousness structure that completes the dual cycle. Test scenarios will be developed simultaneously at each stage of development to verify the functions of the system in terms of perception and understanding, knowledge reasoning, decision evolution, and goal management. Through the step-by-step technical path design, we will build a hierarchical and interconnected prototype of the DIKWP brain-like model, and draw a detailed roadmap from basic research to application to provide guidance for the smooth implementation of the project.

4、Technical route and innovation point

The technical route of this project follows the five-layer logic architecture of DIKWP, and on this basis, it integrates the network artificial consciousness design of perception-cognition-reasoning-reflection. The core idea is to realize the unity of bottom-up data-driven and top-down purpose regulation through dual-loop interaction, so that AI systems can have a human-like cognitive-introspection cycle. The traditional computing system proposed by Turing is a linear input-output process, while our DIKWP mesh architecture is closer to the parallel distributed processing of the human brain: each layer is divided according to its function, and it is connected into a network through feedback, so as to realize the repeated iteration of information between different levels of abstraction. The technical route can be summarized as follows: first, the low-level perception and information processing provide the original materials; The middle-level knowledge and wisdom module integrates and sublimates the material; The purpose module of the upper level imposes guidance and correction on the lower level according to the overall goal; At the same time, the upper management also obtains environmental feedback from the lower level to form a closed-loop regulation.

As mentioned above, there are two cycles within the system: one is the basic cognitive cycle (perception→ cognition→ reasoning → decision-making), which completes the closed loop from perception to action; The second is the metacognitive cycle (reflection→ purpose adjustment→ re-cognition), which monitors and regulates the basic cycle. In terms of specific implementation, the purpose control system acts as the core of metacognition, constantly monitoring the state of the knowledge layer and the intelligence layer, and initiating a reflection mechanism to adjust the strategies of the knowledge and intelligence layer if it identifies that the decision is inconsistent with the intended purpose or the model is uncertain. This mesh architecture ensures the efficient propagation of information and control flows within the system: it can quickly feed environmental changes to high-level intents from the bottom up (enabling AI to perceive new situations and change goals), or it can quickly communicate new purpose requirements to lower-level modules from top to bottom (making perception and cognition optimized for new goals). Bi-directional, multi-level feedback makes the system highly flexible and robust, which is a significant advantage that cannot be achieved with traditional one-way pipelined AI architectures.

Based on the above architectural design, this project will achieve key technological innovations at both structural and functional levels:

Structural innovation: The explicit semantic coding and mesh connection mechanism of the DIKWP model are introduced to realize the transparency and modularization of the cognitive process within AI. The traditional DIKW model is a linear pyramid, and the data is processed upwards into wisdom and output, while this project breaks through the linear limit through mesh interaction, and forms a semantic network through two-way coupling of five layers of elements. This structural innovation allows each layer to have clear input and output semantic specifications and transformation rules. For example, we will design a mathematically descriptible state representation space for the five layers of data, information, knowledge, wisdom, and intent, and define the mapping functions between each layer to form a semantic hierarchy similar to that of an operating system. In this "cognitive operating system", the reasoning process that was originally implicit inside the black box model is broken down into explicit steps that can be monitored. This means that every decision in the system can be traced through data processing, information extraction, knowledge reasoning, wisdom evaluation, and purpose selection, each of which has interpretable semantic meaning and measurable mathematical indicators. This explicit coding structure gives AI a "white box" characteristic, which greatly improves the interpretability and controllability of the system. Once an output anomaly is found, the system can locate a specific level for adjustment, which is an innovation capability that is difficult to achieve in the existing end-to-end model.

Functional innovation: Implementing the internal purpose generation and evaluation mechanism of AI, so that the system can autonomously generate, select, and review goals, is a key step towards artificial consciousness. Current AI is generally driven by external instructions and lacks internal motivation and purpose mechanisms. This project will develop an active purpose generation algorithm in the purpose control system: synthesize the external environment needs and internal state to generate new targets that meet the overall value constraints. This is similar to the process of human motivation (e.g., asking new questions out of curiosity). At the same time, for the generated goals, we introduce the purpose evaluation and screening function, that is, the intelligence layer evaluates each potential goal in multiple aspects (expected benefits, resource consumption, ethical impact, etc.), and the purpose layer selects and implements the optimal or most consistent goals with values. The process is equivalent to setting tasks and reflecting on motivation for oneself. Functional innovation also includes self-evaluation and calibration: after the task is completed, the system will call the metacognitive module to review the execution process, summarize the successes and failures, and feed the experience back to the knowledge and wisdom layer to adjust future behavior. This kind of adaptive learning ensures that AI has the ability to continuously evolve, not just the executor of fixed programs. For example, the patent of Professor Duan Yucong's team proposes to reduce the illusion and purpose shift of large models through multi-round dialogue and semantic firewall technology, which is actually a manifestation of adding target constraints and bias correction to the function. Through these innovative features, the system of this project will exhibit the rudiments of a quasi-autonomous agent: capable of driving action according to intrinsic purpose, and constantly revising self-goals in environmental interactions, approximating the expression of human will and consciousness.

To sum up, the uniqueness of the DIKWP artificial consciousness architecture of this project in terms of technical route is that it uses a network dual-cycle structure to open up perception and reflection, and endows the system with autonomy with a purpose-driven mechanism。 This architecture organically integrates perception, cognition, reasoning, and reflection, which is significantly different from the traditional pure perceptual action model, and injects "self" elements into AI. Whether it's structural transparency or functional autonomy, our designs are at the cutting edge. For example, through the DIKWP semantic framework, we have built an internal semantic white-box mechanism for large models, which can effectively alleviate the unexplainable pain points of the AI decision-making process. By being intent-driven, we transform AI from a passive responder to an active problem solver, reducing the reliance on human instructions and improving adaptability to complex and changing tasks. These technological innovations not only enrich the connotation of artificial intelligence architecture, but also take an important step towards engineering practice in the field of artificial consciousness, which will lay the foundation for achieving a higher level of machine intelligence.

5、Project team and implementation basis

This project is led by the team of Professor Duan Yucong, who has a deep research accumulation and achievement foundation in the fields of artificial consciousness modeling, brain-computer fusion, and multimodal interaction, which provides a strong guarantee for the smooth implementation of the project.

First of all, the **team is an international leader in artificial consciousness theory and model. Professor Duan Yucong is an academician of the International Academy of Advanced Technology and Engineering, a corresponding member of the National Academy of Artificial Intelligence, and the chairman of the World Association of Artificial Consciousness. The DIKWP artificial consciousness model proposed by him has aroused strong repercussions in the academic community, and the relevant theoretical results have been published through a number of papers and reports, and the first World Artificial Consciousness Conference (**AC2023) and other academic activities have been successfully held in 2023, which has greatly enhanced the team's influence in this field. In addition, the team led the development of a preliminary plan for the DIKWP international standard for AI assessment, which provides a standard framework for measuring the "level of awareness" of AI systems in the future. These reflect the team's academic discourse and leadership in the field of artificial consciousness.

Secondly, the team has a wealth of intellectual property rights and technical reserves, which proves its practical ability. As mentioned above, Professor Duan's team has obtained 114 domestic and foreign authorized invention patents. These patents cover a wide range of areas and constitute a complete DIKWP technology system, including core technologies such as mesh cognitive model, dual-cycle consciousness architecture, cognitive operating system, and semantic network coupling. Such a systematic patent layout is the first in the world, providing a solid independent technical foundation for the implementation of the project. It is worth mentioning that the team pays great attention to the translation of research results into actual systems. As early as 2023, the team developed the world's first DIKWP-AC artificial consciousness prototype system and publicly demonstrated it at an international conference. The prototype realizes the verification of the DIKWP theory with a small model, and adopts the mathematical/physiological twin system architecture, which has strong interpretability and high computational efficiency, which fully proves the engineering feasibility of the DIKWP model. The achievement has also been recognized by the industry: at the CCF China Digital Service Conference in September 2023, the team's graduate students presented a paper poster on "Intent-Driven Data-Information-Knowledge-Wisdom Integration DIKWP Physiological and Artificial Consciousness Prototype" won the Best Poster Award. This award shows that the team's work in applying intention-driven artificial consciousness models to specific scenarios, such as physiological simulations, has been highly praised by peer experts.

In terms of brain-computer fusion and multimodal interaction, the team also has forward-looking explorations. The so-called brain-computer fusion refers to the improvement of AI by combining neuroscience principles on the one hand, and the interaction between AI and the human nervous system on the other. In recent years, the team has carried out research on the correspondence between semantic cognition and brain signals, and tried to correlate the purpose output of the DIKWP model with biological neural signals to expand the physiological authenticity of the artificial consciousness system. This interdisciplinary attempt is expected to enable AI to obtain human brain feedback through brain-computer interfaces in the future, further improving the realism of artificial consciousness models. In terms of multimodal interaction, the team has built a unified semantic space covering text, speech, vision and other modalities, and developed a prototype of a multimodal human-computer dialogue system. In the medical scenario, the team developed an intelligent consultation dialogue simulation system, which converts the patient's language description (text/speech) into the semantics inside the DIKWP model, and then generates the diagnosis and treatment recommendations by the artificial consciousness system. As a physician assistant, the system can understand the patient's multimodal information and give personalized feedback, which fully demonstrates the team's strength in multimodal semantic fusion and interaction.

In addition, the team also has perfect scientific research conditions and a foundation for industry-university-research cooperation. The core members of the project include experts in artificial intelligence, computer science, cognitive psychology, neuroscience and other fields, as well as a number of postdoctoral and doctoral students, forming an interdisciplinary talent echelon. The School of Computer Science of Hainan University provided a special DIKWP artificial consciousness laboratory for the project, equipped with a high-performance computing platform and an EEG/physiological signal acquisition device, which can meet the needs of simulating the cognitive process of the human brain, large model training and multimodal data processing. The team has maintained cooperation with many well-known scientific research institutes and enterprises in China, including cooperating with medical institutions to develop intelligent diagnosis and treatment (such as the aforementioned test scenario with Hainan Provincial People's Hospital), cooperating with educational technology companies to explore intelligent teaching assistants, and cooperating with industrial enterprises to research intelligent control systems. These collaborations not only provide rich application scenarios and data support for project implementation, but also lay the foundation for subsequent achievement transformation.

To sum up, Professor Duan Yucong's team has outstanding advantages and accumulation in theoretical research, patented technology, prototype system, talent equipment, cooperation network, etc. This team has created a number of firsts in the field of artificial consciousness models: the first DIKWP theoretical system report, the first prototype of artificial consciousness evaluation standards, and the first demonstration of artificial consciousness prototype...... These pilot results are a testament to the team's ability and confidence in navigating the project. On this solid foundation, we have reason to believe that the team can complete the research tasks of this project with high quality and successfully implement the theoretical ideas into engineering practice.

6、Stage results, assessment indicators and transformation paths

In order to ensure the realization of the project objectives, we will gradually advance according to the five-tier structure of DIKWP, and set up phased results and assessment indicators to gradually verify the perfection of the system functions and prepare for the subsequent industrial transformation. The specific phases are as follows:

Phase 1 (Data-Information Layer, Cognitive Infrastructure): The time span is approximately 1 year after the start of the project. In this stage, the development and integration of data acquisition system and information fusion system are focused on providing high-quality input for upper-level cognition. Expected outcomes include: a perception data platform that supports multimodal input, which can realize the unified collection and preprocessing of image, voice, text and other data; A rudimentary information fusion engine capable of transforming multi-source data into structured information representations and storing them. On this basis, we will demonstrate the information processing process in simple scenarios, such as the whole process from multimodal questions to semantic understanding in the intelligent Q&A prototype. The assessment indicators include the throughput rate of multimodal data processing, the accuracy of information extraction, and the proportion of performance improvement compared with single-modal processing. The sign of passing the acceptance is that the system has stable operation capabilities at the perception and information levels, and initially realizes the semantic perception of the environment.

Stage 2 (Knowledge Layer, Cognitive Memory Formation): The time span is approximately the 2nd year. In this stage, a knowledge modeling system is constructed, and the information output of stage 1 is precipitated into usable knowledge, and the system is endowed with basic reasoning and memory functions. Expected deliverables include: prototype knowledge graph related to the project area, covering key concepts, relationships and rules; A set of inference algorithm modules, which can answer queries and make logical deductions based on the knowledge base; and a knowledge update mechanism that can automatically refine or revise the knowledge base content based on new data. In the test, we will construct several benchmark reasoning tasks to evaluate system performance, such as Q&A in the field of medical diagnosis (given a patient's symptoms, the system reasoning about possible diseases and the basis). The assessment indicators cover the size of the knowledge base (number of nodes/edges), the degree of automation of knowledge acquisition, the accuracy of inference, and the average length of the inference chain. The acceptance criteria are that the system demonstrates reliable memory and reasoning ability at the knowledge level, such as answering professional questions in a specific field with the expected accuracy, and has a mechanism to update knowledge on a rolling basis as the data grows.

Stage 3 (Intelligent Layer, High-Level Decision-making and Adaptation): The time span is about the second to the first half of the third year. In this stage, an intelligent evolutionary system is developed to realize complex decision-making and adaptive learning functions on the basis of existing knowledge. Expected outcomes include: implementation of a decision planning module capable of handling multi-objective, multi-constraint situational decision-making; Establish a continuous learning framework to enable the system to continuously optimize strategies through repeated training in the simulation environment (e.g., the combination of reinforcement learning and imitation learning); The safety ethics constraint module is introduced to ensure that the system considers the preset value standards and safety rules when making decisions. In this phase, we will test the intelligence of the system in near-real complex scenarios, such as simulating autonomous driving or medical diagnosis and treatment scenarios, so that the system can make continuous decisions based on the dynamic environment, and can improve the strategy through repeated training in the event of a mistake. The assessment indicators include task completion rate (such as the safe arrival rate in driving simulation, the correct diagnosis rate in medical diagnosis simulation), decision response time, and performance degradation under environmental changes. The sign of acceptance is that the system has the ability to make independent decisions and adapt at the intelligent level, can deal with more complex problem situations and shows a gradual improvement learning curve.

The fourth stage (purpose layer, the prototype of artificial consciousness): the time span is about the second half of the third year. In this stage, the purpose control system is realized, and the dual-cycle architecture is completed, so that the system has preliminary self-awareness characteristics. Expected outcomes include: implementation of an internal purpose representation and management module that allows the system to generate and switch targets based on external task requirements and internal states; The metacognitive monitoring module can be implemented to monitor the operation of each layer and detect anomalies and deviations. Implement an purpose calibration mechanism that automatically adjusts intents and corresponding policies when target deviations or conflicts are detected. At this stage, we plan to verify the prototype function of artificial consciousness in a comprehensive scenario. For example, deploy an intelligent assistant robot to perform tasks in a simulated home or medical environment, and have it multitask (e.g., answering questions, monitoring the environment, planning actions) at the same time, observing whether it can set the order autonomously, complete it in an organized manner, and self-reflect and correct strategies when it fails. The assessment indicators include the success rate of multitasking, the reasonableness of purpose switching (whether it meets the priority expectations), and the self-adjustment time of the system for abnormal conditions. The acceptance criteria are that the system shows obvious autonomy and reflection at the level of intent, such as the ability to reformulate the plan according to the changing situation without human intervention, and avoid repeating the same mistakes. This will mark the basic closed-loop formation of the architecture of artificial consciousness.

Phase 5 (System Integration and Application Demonstration): The time span is about half a year before the end of the project. In this stage, the results of the previous layers are integrated into a complete DIKWP brain-like intelligence model, and the demonstration application test is carried out in the real scene. Expected outcomes include: prototype systems for specific application areas (e.g., smart medical diagnosis prototypes, intelligent teaching assistant prototypes, or smart manufacturing decision-making prototypes) that enable models to interact with users in a real-world environment; Compile technical specifications and evaluation reports, and summarize the indicators of the DIKWP model in terms of function, performance, safety, etc.; Form the next step of industrialization implementation plan. In the demonstration application, we will work with partners to build pilots, such as deploying intelligent diagnosis and treatment assistant prototypes in hospitals, intelligent tutor assistance systems in pilot classrooms, or intelligent control decision-making modules on factory floors to verify the practicability and reliability of the model. Metrics are domain-specific and include user test feedback (e.g., physician satisfaction with diagnosis and treatment recommendations, students' acceptance of teaching aids), system operational stability (uptime, failure rate), and performance comparison (efficiency or effectiveness improvement compared to existing solutions). The final acceptance of the project is marked by the delivery of a white paper on the roadmap of the "New Generation of Brain-like Intelligence Model", as well as at least a prototype of the artificial consciousness system that has been tested in real life, and a clear path and cooperation purpose for subsequent industrial development.

On the basis of completing the above stage goals, the project will also actively plan the transformation and application of the results to promote the implementation of artificial consciousness technology in various industries and create actual social value. Specifically, the artificial consciousness brain-like model of this project has broad application prospects in the following fields:

Healthcare: Using the intelligent diagnosis and treatment model built in this project, more advanced medical assistance decision-making can be realized. Traditional medical AI is mostly a diagnostic suggestion tool, which lacks a comprehensive understanding of patients' complex symptoms and active tracking of diagnosis and treatment goals (such as cure and remission). Our model has the characteristics of multimodal information fusion and intention-driven, which can synthesize the patient's symptom description and test results (data/information layer) to construct a medical knowledge graph (knowledge layer), and then combine medical experience and ethical constraints (wisdom layer) to formulate personalized treatment plans, and dynamically adjust the treatment goals (purpose layer) according to patient feedback. For example, in the diagnosis and treatment of complex diseases such as gout and systemic lupus erythematosus, the artificial consciousness system can assist doctors to weigh the pros and cons of multiple treatment methods, propose a plan that takes into account the efficacy and quality of life of patients, monitor the patient's status during the treatment process, and recommend adjustments in a timely manner. This will improve the efficiency of diagnosis and treatment of difficult cases, reduce the burden on doctors, and provide patients with a more humane medical experience.

Education and training: The brain-like intelligence model of this project can be used to develop intelligent teaching and personalized learning systems. Traditional educational AI (such as intelligent tutors) mostly answer questions based on preset rules, which makes it difficult to truly understand students' learning intentions and emotional states. After the introduction of artificial consciousness, the intelligent tutor can obtain the students' expressions, intonation and behavior through multimodal perception (data/information layer), understand their knowledge mastery and psychological needs (knowledge/wisdom layer), and adjust teaching strategies and goals accordingly (purpose layer). For example, slow down the pace of explanations for students who are confused, or add challenges to students who have spare energy. The system can also reflect on the teaching effect after class and optimize the teaching plan. Through this awareness-driven interaction, students will receive the same attention and guidance as a real teacher, significantly improving their learning efficiency and experience. The path to industrialization in the field of education can be to cooperate with online education platforms to launch AI tutors with dialogue and emotion recognition capabilities, or introduce intelligent teaching assistant robots in the classroom to realize the automation of teaching according to aptitude.

Industry & Others: Artificial consciousness models have the potential to be used in a wide range of fields, including industrial manufacturing, urban management, and financial services. In industrial manufacturing, AI systems need to have the ability to independently adjust production plans and ensure safety in the face of complex production processes and emergencies. Our model can be embedded in the industrial intelligence middle platform as an intelligent decision-making center, real-time perception of data at each link of the production line (data/information layer), use equipment knowledge and experience rules to optimize production scheduling (knowledge/intelligence layer), and set production targets according to global capacity demand and safety regulations (purpose layer). In the event of an equipment failure or order change, the system autonomously assesses the impact and re-plans the production rhythm to ensure a balance between efficiency and safety. In terms of urban management, the artificial awareness system can be applied to smart transportation or energy allocation, through the integration of traffic flow/energy consumption data (data/information), combined with urban planning knowledge (knowledge) and real-time optimization algorithms (intelligence), it can automatically adjust signal timing or load allocation targets (intentions) to achieve adaptive city-level regulation. In financial services, AI can better understand market dynamics and user needs through awareness models, proactively adjust investment or service strategies, and provide solutions that are more in line with customer goals, while strictly following the principles of risk control. In short, the results of this project can be used as a general intelligence hub in all walks of life, and the introduction of brain-like models with artificial consciousness will bring significant improvements in system performance and safety reliability in scenarios that require complex environments, long-term planning, and multi-objective trade-offs.

Finally, on the path of achievement transformation, we will use the team's rich industry-university-research cooperation network to accelerate the implementation of technology. On the one hand, after the completion of the project, it is planned to build a joint laboratory or demonstration base with relevant industry leaders or government departments, customize and optimize the DIKWP artificial awareness model according to actual needs, and verify the business model in a small area. On the other hand, we make full use of the advantages of the team's patent pool to promote industrialization through patent licensing and technology shareholding. He mentioned that Professor Duan's team adheres to an open and cooperative attitude, is willing to work with the government, enterprises, and universities to develop commercial products based on the DIKWP model, and has expressed its willingness to donate some patents free of charge to promote the formulation of industry standards. These measures will lower the threshold for cooperation and attract all parties to invest resources to jointly improve the technology ecosystem. In key areas such as medical care and education, we will also strive for policy and financial support (such as joining the smart medical and smart education projects) to ensure that the closed loop from R&D to application is connected. Through the above steps, the innovative results of this project are expected to be quickly transformed into real productivity after the end of the project cycle, forming a sustainable new generation of brain-like intelligence industry. Looking forward to the future, the "brain-like intelligence model with artificial consciousness support" depicted in the roadmap will not only be a scientific research achievement, but also one of the core engines to promote the upgrading of the AI industry, and contribute to China's leading position in the new round of global artificial intelligence competition.