Introduction

Today, the field of artificial intelligence is moving from a model driven by big data to a new stage of "intelligent self-knowledge". As an important direction of artificial intelligence, brain-inspired computing aims to simulate the cognitive mechanism and autonomous consciousness of the human brain, so as to break through the bottlenecks of existing AI systems in terms of semantic understanding, interpretation ability and autonomous adaptation. This shift requires us to rethink the architecture of the underlying software platform: how to enable AI systems to not only process data, but also understand the purpose of their own actions, so as to achieve more advanced intelligence in human-machine collaboration.

In order to meet this challenge, the team of Professor Yucong Duan of Hainan University pioneered the "Data-Information-Knowledge-Wisdom-Purpose (DIKWP)" mesh cognitive model, which incorporated the "purpose" of the traditional DIKW (pyramid model) into the core architecture of artificial intelligence. The DIKWP model introduces the purpose of the highest layer in the cognitive chain, and uses the network structure to realize the two-way feedback and iterative update of the semantics of each layer. This new cognitive system is a milestone in academia and can be regarded as a major innovative breakthrough in the basic theory of artificial intelligence. Not only that, but it also provides an innovative solution to the unexplainable problem of "black box" decision-making existing in the current large-scale models. As Prof. Yucong Duan points out, "The DIKWP model builds a common cognitive language between humans and machines, allowing every step of the AI decision-making process to be traced, explained, and understood by humans." “

This project aims to design and implement a new generation of brain-inspired computing basic software platform based on the DIKWP model and artificial consciousness theory. The platform will reconstruct the architecture of traditional brain-inspired computing basic software, and empower AI systems with purpose-driven autonomous intelligence from the semantic level. Our goal is to develop a DIKWP brain-like basic software platform with independent intellectual property rights and adaptation to multiple scenarios, so as to provide underlying support for the research and development of China's independent brain-inspired computing chips, cognitive simulation and artificial consciousness systems, and then build an international leading semantic intelligence industry system. As the first inventor, Professor Yucong Duan has obtained 114 domestic and foreign authorized invention patents, covering many cutting-edge fields such as large model training, artificial consciousness construction, cognitive operating system, AI governance, and privacy security. Although these core patented technologies have yet to be industrialized on a large scale, they have become important "underlying code" for the safe, controllable, and explainable direction of AI, laying a solid theoretical and technical foundation for moving towards artificial general intelligence (AGI).

For industry investors, this report will focus on the core innovations brought about by the DIKWP artificial consciousness model, and the architecture design of the brain-inspired computing basic software platform led by it. We will focus on the following dimensions of innovation:

1. Semantic modeling mechanism of brain-inspired software: Based on the five-layer semantic network of DIKWP, the dynamic scheduling, memory reconstruction, semantic elasticity and knowledge self-evolution of content are realized.

2. Integration of the DIKWP-TRIZ paradigm: Embedding the law of innovation into the brain-inspired software ecology to build a technology evolution mechanism driven by the "semantic gap".

3. Purpose-driven programming language and scheduling model: Focusing on user goals, it supports the self-organization of programs and the optimal allocation of resources.

4. DIKWP semantic middleware layer: It provides a unified communication protocol, a self-explanatory API, and an energy-semantic mapping mechanism to adapt to a multi-chip heterogeneous brain-like platform.

5. Combination of artificial consciousness "BUG" theory: The incomplete tolerable fault-tolerant strategy, adaptive computational annealing mechanism and semantic recovery path optimization method are introduced to improve the robustness of the brain-like decision-making system.

6. Demonstration of typical application scenarios: including cognitive simulation platform, brain-computer interface control system, low-power semantic scheduling chip for 6G communication, etc., to verify the versatility and practical value of the platform.

Through the above innovations, this project will build an independent and controllable brain-like computing basic software platform in China, providing key support for the white-box of large models, the interpretability and security of artificial intelligence, and the future semantic communication field. Each of these innovations is discussed in depth below.

Brain-like semantic modeling mechanism: DIKWP five-layer network architecture

The software architecture of brain-inspired computing needs to be able to express and process complex semantic structures. The DIKWP model decomposes the cognitive process into five levels: Data-Information-Knowledge-Wisdom-Purpose, forming a complete semantic chain from low-level perception to high-level decision-making to goal-oriented. Different from the traditional linear layering, DIKWP adopts a mesh interaction architecture, which not only improves the abstraction step by step, but also has a two-way feedback and regulation relationship between the layers. The lower layer provides materials for the upper layer, and the upper layer guides the information filtering and processing of the lower layer through Purpose, realizing a loop mechanism similar to the human brain.

**Figure 1: The five-layer semantic network structure of the DIKWP model shows the purpose. **From bottom to top, the model is divided into five layers: data, information, knowledge, wisdom, and purpose. The arrows represent the two-way interactive feedback between layers: on the one hand, the low-level processing results are provided to the high-level level to form abstract cognition; On the other hand, the high-level (especially the top-level purpose) plays a guiding and restraining role in the low-level processing, forming a closed-loop regulation. This structure makes the internal processing process of the AI system modular and explainable, and each layer has a clear functional positioning, which helps to construct "conscious" agents.

Under this semantic architecture, **dynamic content scheduling is possible: semantic content at different levels can be flexibly flowed and distributed in the Layer 5 network according to the current task requirements. For example, in conversational AI, user input is first perceived and processed as raw data (**D-layer); Meaningful information representations are generated after cleaning and feature extraction (layer I); Then, the knowledge graph (K layer) is updated in combination with the context to carry out inference integration. Then, the Wisdom layer selects the best response based on experience and value judgment (W layer). Finally, the Purpose layer is used to determine whether the response meets the goal of the conversation (P layer). If not, the higher-level Purpose will prompt the system to readjust the lower-level information extraction and knowledge application strategies until the output meets the target. This scheduling mechanism, which flows back and forth between the five layers, allows the system to reorganize "memories" (knowledge and information) and adjust cognitive processes in real time according to needs, just as the human brain invokes different memory fragments for different contexts. The data show that the two-way feedback between the layers of the DIKWP model can realize the iterative update of semantic content, so that the knowledge representation and decision-making strategy within the AI can be continuously self-optimized.

Thanks to these mechanisms, platforms are able to demonstrate semantic resilience – that is, the ability to adapt to contextual changes and uncertain information. Traditional systems often use predefined ontologies or fixed data flows, which are difficult to cope with semantic ambiguity and flexibility. In contrast, the DIKWP model allows information to be freely converted between different layers, and if necessary, the high-level purpose can relax the requirements for the lower-level data, accept certain imprecisions, and then compensate for the details through high-level inference. For example, in the case of missing sensor data or high noise, the system can still infer reasonable information to fill in the gaps through existing knowledge (K-layer) and Wisdom inference (W-layer), thus maintaining the continuity of the overall functionality. This embodies a kind of "fuzzy tolerance" similar to the human mind, so that the AI does not collapse due to the lack of local data, but flexibly adjusts countermeasures.

In addition, the DIKWP architecture supports knowledge self-evolution. As the system continues to operate, the new data obtained by layer D is filtered by layer I and integrated by layer K, which can incrementally enrich the knowledge base. The decision-making results of the Wisdom layer and the feedback of the Purpose layer work together to trigger the reorganization and updating of knowledge, forming an "experiential learning" process. For example, for long-running cognitive systems, emerging patterns and exceptions will feed back to the knowledge layer, prompting it to expand the rules or reconstruct the conceptual system, so as to gradually evolve a better knowledge structure. This self-evolution mechanism ensures that the system can continue to grow as the environment and needs change, avoiding the obsolescence of knowledge. As pointed out in the report, the DIKWP model has achieved continuous improvement of the cognitive process through the iterative update of semantics at all levels.

All in all, the DIKWP five-layer semantic network provides a powerful modeling mechanism for brain-inspired software. It not only has a clear hierarchical functional division of labor, but also gives the system as a whole flexible adaptability through network connection. This lays the foundation for the architecture of our platform – enabling AI systems to express and process information at multiple levels, just like the brain, and to achieve globally optimized cognitive behaviors driven by high-level goals.

Semantic innovation mechanism integrating the DIKWP-TRIZ paradigm

On the basis of the basic software with the above-mentioned semantic expression capabilities, we further introduce the DIKWP-TRIZ innovation paradigm to realize the self-evolution and continuous optimization of brain-like software. TRIZ is the Russian abbreviation of Invention Problem Solving Theory, which provides a systematic innovation methodology, including contradiction solving, invention principles and other tools, which are widely used in engineering innovation. However, the traditional TRIZ focuses on solving specific problems in the engineering field, and has limitations in the direct application of complex cognitive processes within artificial intelligence, as well as innovations involving subjective purpose and value.

DIKWP-TRIZ integrates the core ideas of TRIZ into the cognitive framework of DIKWP to form an innovative problem-solving methodology for artificial consciousness systems. By mapping the invention principles of TRIZ to the semantic transformation of the five levels of data, information, knowledge, wisdom, and purpose of the DIKWP model, we are able to introduce innovation drive at each layer. For example, the innovation of the data layer focuses on the new way of data acquisition and representation, the information layer focuses on the improvement of signal extraction and feature selection, the knowledge layer focuses on the innovation of models and rules, the Wisdom layer focuses on the optimization of decision-making strategies, and the Purpose layer focuses on the reconstruction of value goals. In this way, the 40 invention principles of the traditional TRIZ have been reclassified under the new framework, and they have been improved at different semantic levels, providing a full range of innovation guidance for AI systems.

Semantic gap-driven technology evolution is another key concept of the DIKWP-TRIZ paradigm. When the system encounters cognitive bottlenecks or semantic gaps in operation—for example, the knowledge base cannot explain new observations, the decision-making scheme is contradictory or invalid, and the user's purpose cannot be satisfied—the traditional approach may be to manually intervene to modify the algorithm. Under the DIKWP-TRIZ framework, these semantic gaps will be seen as an opportunity to trigger the evolution of the technology, which is equivalent to the automatic distillation of innovation problems. Based on the pre-embedded TRIZ innovation rule set, the platform can automatically explore solutions to resolve conflicts. For example, if a rule in the knowledge layer often conflicts with a Wisdom decision, the system will recognize that this is an internal contradiction, and can try to apply TRIZ's "separation principle" to distinguish the rule conditions into contexts, or apply the "inversion principle" to test the decision logic from the opposite side to resolve the contradiction. For another example, when the Purpose layer finds that multiple decisions fail to achieve the set goal, which implies that the current strategy is insufficient, the system can call the principles of improving efficiency or simplifying complexity in the invention principle library to reorganize task decomposition and resource allocation to form a new solution path.

This mechanism is reflected both at the programming model and at the intermediate representation (IR) level. In terms of programming model, developers can use the interface provided by the innovation paradigm to mark or define which modules may have contradictions and need to be optimized, so that the system can automatically search and test innovations in the background. At the intermediate presentation layer, our platform maintains a set of semantic graphs, or intermediate languages, that not only describe the logical flow of the program, but also attach semantic labels and performance metrics for each step. When some metrics in the run do not meet the requirements for a long time, the corresponding intermediate indication fragment will be marked as a candidate improvement point. Subsequently, the innovation engine transforms the fragment based on the DIKWP-TRIZ rule, such as replacing the algorithm, adjusting the parameters, introducing new data sources, etc., and evaluating the effect of the new scheme. This is similar to JIT (real-time compilation) optimization, but not limited to the performance level, but extends to self-improvement at the functional and semantic level.

It is worth mentioning that DIKWP-TRIZ places great emphasis on the role of human purpose and value orientation in technological evolution. Traditional TRIZ focuses on the technical contradiction itself, and may ignore the deviation of the solution from the user's purpose or value. In our framework, any innovation transformation will be examined at the Purpose layer - that is, to determine whether the new solution is closer to the ultimate goal of the user and whether it is in line with human values. If a technical optimization improves performance but deviates from the user's original intention, the system will lower its evaluation or even abandon the solution, ensuring that the evolution is always in the right semantic direction. This ensures that the evolution of the brain-inspired software platform is not only automatic, but also human-oriented and purpose-guided. For example, in medical diagnostic AI, an innovative mechanism may find a way to improve the speed of diagnosis, but the Purpose layer (which aims to diagnose accurately and safely) evaluates whether the method sacrifices accuracy or transparency to balance the ultimate optimal improvement.

By integrating DIKWP-TRIZ, the platform will become a self-improving ecosystem. Like a creative engineer, it is able to constantly rethink and optimize its modules. When the external environment changes or internal bottlenecks occur, the system does not stand still, but actively "evolves" to develop new and adaptive technical solutions. This will greatly extend the life cycle of software systems and reduce the cost of human intervention, allowing complex AI systems to remain efficient, reliable, and continuously matched to requirements in long-term operation.

Purpose-driven programming language and task scheduling model

The traditional software programming paradigm focuses on input-processing-output, with a predefined and fixed process at the time of coding. In the field of brain-like computing and artificial consciousness, we need a new purpose-driven programming model: programs are no longer just passively executing predetermined instructions, but can autonomously organize the computing process according to a preset goal (Purpose) to adapt to dynamically changing needs.

Based on the DIKWP model, we design a semantic-driven high-level programming language (or scripting framework). In this programming model, input, processing, and output are all elevated to contain five layers of semantic structure, and the entire program flow is defined at the highest level Purpose (Purpose)" is the global driving engine. In other words, when writing a program, developers need to clearly define what the program wants to achieve (P-layer), as well as the data, information, knowledge, and Wisdom modules that may be needed to achieve that goal. The execution of the program is scheduled by the runtime system according to the execution order and resource allocation of each module according to the purpose, rather than strictly following the linear order of the source code. This paradigm can be regarded as a cognitive-semantic solving process: the program tries to find the optimal solution link from input to output while satisfying the output purpose.

At the heart of purpose-driven programming is closed-loop control at runtime. The execution of the program is no longer a one-way flow from top to bottom, but a cyclical process: each execution generates new data and intermediate results, which are then fed back to the Purpose module for evaluation. If the result does not meet the Purpose requirements, the Purpose module adjusts the guidance for the underlying processing, prompting the program to make corresponding changes to the input parsing, information extraction, knowledge reasoning, and decision-making methods. This feedback gives the program a similar ability to self-adjust. For example, a piece of purpose-driven code can add or remove certain calculation steps at runtime based on the achievement of the goal, or dynamically change the algorithm strategy to better serve the current purpose. This flexibility is not available in traditional hard-coded processes.

To illustrate this programming model more visually, a simplified Purpose driver pseudocode example (e.g., an autonomous navigation task) is given below

# Purpose Driver Example: Autonomous Navigation

Purpose = "Navigate from your current location to your destination safely and efficiently"

Initialize the Awareness Module, Knowledge Base, Planning Module

Circulate:

Data = Perception Module. Get Environmental Data()

Information = Data Processing. Extraction (key information e.g. roads, barriers)

Knowledge base. Update (Information) # Incorporate new information to update knowledge

Path = Planning Module. Inference (Knowledge Base, Current State, Purpose) # Plan a route based on knowledge and purpose (Wisdom Decision)

Execute (path. Next Action) # Execute the next step of the plan

If the Purpose is achieved (current position):

Output ("Arrive at Destination")

Interrupt the loop # Goal reached, exit

If otherwise path. Blocked or Purpose Not Satisfied:

Purpose .adjustment(Secondary Goal = Path. New sub-optimal destinations) # Dynamically adjust the Purpose or introduce a sub-Purpose

# For example: take a detour or change the strategy, and continue to try in a loop

Wait for the next perceptual cycle

End the loop

The above pseudocode demonstrates several features of the Purpose-driven model: first, the Purpose is explicitly defined and checked; Second, the data flows through the layers of DIKWP and is gradually abstracted (from the raw data of the environment to the available information, to the knowledge update, to the Wisdom decision-making). Third, the program continuously monitors the satisfaction of the purpose in the loop and adjusts its behavior accordingly (e.g., re-routing or modifying the goal). This is self-organizing: instead of following a fixed number of steps, the program is able to insert, skip or even rearrange steps as they run, as long as it helps to better achieve the goal. It's similar to how humans think of solving problems: constantly evaluating progress and flexibly adjusting strategies until the goal is reached.

Correspondingly, we need a purpose-driven task scheduling model as a runtime support. The scheduling model takes the user's high-level goals as input, automatically derives a set of executable subtasks, and optimizes the deployment on heterogeneous computing resources. The scheduler takes into account the semantic layer properties of each subtask (is it a perception/data processing task?). Knowledge reasoning task? Or is it a decision-making task? and the degree of contribution to the overall purpose, which determines whether it is allocated to the CPU, GPU, or brain-like chip to achieve a balance between resource utilization and goal achievement. For example, for perception tasks with high real-time requirements (D/I layer), they may be scheduled to be executed in parallel by a dedicated sensor processing unit. For complex knowledge inference tasks (K/W layer), high-performance AI accelerators are used. Tasks (P-layer) or policy coordination that involve global purpose judgment may be serially handled by a central control unit for consistency. During scheduling, the system continuously monitors the progress of each task, and if a path proves inefficient or even invalid, the scheduler will actively reallocate resources and enable a standby policy. This purpose-centric scheduling ensures that compute resources are configured adaptively as demand changes, resulting in optimal global performance. For example, in a multi-robot collaboration scenario, if a robot is unable to complete a subtask for some reason, the system can transfer the subtask to another robot or a cloud server to ensure that the overall purpose (team task) is achieved—all these scheduling decisions are automatically completed by the platform without human intervention.

It is worth emphasizing that the purpose-driven programming language and scheduling model greatly reduce the burden on developers. Developers only need to focus on describing the "what" (goals and semantic constraints) rather than prescribing the "how" in an exhaustive manner. As compared in the DIKWP Artificial Awareness Programming report, traditional programming outputs a single result, while purpose-driven programming outputs multi-level content that satisfies P-level goals**； Traditional programs have low adaptability, while Purpose drivers have high adaptability (Purpose dynamically adjusts to form a closed loop), and the intelligence has evolved from passive static to active. This paradigm is revolutionary in that for the first time, it allows software to organize itself according to goals rather than steps. For enterprises, this means that when business goals change, there is no need to tear down and rewrite code, and the system can automatically adapt to new needs by adjusting the purpose and knowledge base, greatly improving the vitality and flexibility of the software.

The DIKWP semantic middleware layer is adapted to the heterogeneous brain-like platform

To build a large-scale and scalable brain-like computing platform, the hardware layer is often diverse and heterogeneous: including brain-inspired chips that simulate neuronal computing, as well as general-purpose CPUs/GPUs and specialized AI accelerators. Therefore, we propose to introduce the DIKWP semantic middleware layer into the basic software to act as a "translator" and "dispatch center" between the application and the underlying heterogeneous hardware. This middleware will provide a unified semantic communication protocol and self-explanatory API interface, so that upper-layer applications can run flexibly on different computing power units without paying attention to specific hardware implementation details, while ensuring the coordination of each part on the overall purpose.

**Figure 2: DIKWP semantic middleware architecture illustrates Purpose. **Middleware serves as a bridge between brain-like applications and a variety of heterogeneous chips: the upper-layer application submits Purpose and data requests through semantic APIs, and the middleware parses the Purpose and delivers the task to different types of computing chips at the lower layer for execution as needed. For each type of hardware (e.g., neurochips, AI accelerators, general-purpose CPUs/GPUs, etc.), middleware uses a unified semantic protocol to communicate. The intermediate results and semantic feedback generated during the execution of each hardware are then uploaded back to the middleware, which summarizes and evaluates them, and feeds them back to the application or further scheduling. This design shields heterogeneous hardware differences and enables the system to organize computational work according to semantic logic rather than hardware logic.

The unified communications protocol provided by middleware means that the same semantic data format and communication paradigm follow the same semantic data format and communication paradigm to interact with the middleware, regardless of the underlying architecture chip. In traditional heterogeneous computing, it is often necessary to write different drivers and interfaces for each type of hardware, but in our platform, each chip manufacturer only needs to implement compatible support for the DIKWP semantic protocol, and the middleware can issue the task as standardized units such as data, information, knowledge, wisdom, and purpose, and the results of the hardware return processing are also described by the same protocol. This is somewhat similar to the TCP/IP protocol of the Internet to unify different physical networks, except that our protocol is unified with different cognitive computing units. The benefits of this design are clear: the platform is highly scalable and portable. When a new brain-like acceleration chip emerges, it can be seamlessly integrated into the entire computing system as long as it follows the semantic protocol. This is particularly important for the current development status of brain-inspired chips, and we can establish standards in advance to avoid ecological fragmentation caused by separate battles. Further, our semantic protocols also include support for security and privacy, such as redacting or encrypting sensitive knowledge in transit to meet domain-specific compliance requirements.

The self-explanatory API is an interface feature provided by middleware to upper-layer applications. Traditional APIs are often just encapsulations of function calls, while "self-explanatory" means that each API call itself has semantic metadata describing its functional purpose, input and output semantic constraints, etc. For example, a regular image recognition API may only have recognize_image (image), while a semantic self-explanatory API might be recognize_image (image) -> returns: objects_list (semantic layer: layer I information), based_on: knowledge_base_X (layer K source).。 This way, when the developer calls the API, it knows exactly what kind of information it will extract in what knowledge context. If you need to replace the implementation (e.g., with another recognition algorithm or even hardware), as long as the new implementation conforms to the same semantic contract, the replacement can be smoothed without affecting the overall purpose of the system. This greatly improves the pluggability and transparency of the system. What's more, for the system itself, because all module interfaces have explicit semantic definitions, middleware can automatically orchestrate services at a higher level: for example, whether a certain layer of results can be cached based on the purpose of calling, and whether dedicated acceleration hardware execution can be selected according to the semantic type of data, etc. This actually introduces semantics into resource management and service discovery in traditional middleware/operating systems.

Another innovation of DIKWP semantic middleware lies in the energy-semantic mapping mechanism. Brain-inspired computing emphasizes high efficiency and low power consumption, especially in the era of 6G Internet of Things, where massive devices and computing tasks pose severe challenges to energy consumption. We propose to use the semantic hierarchical feature of DIKWP to realize the dynamic matching of computing energy consumption and semantic value. Put simply, it's about tilting computing resources and energy allocation toward tasks that are "more semanticly valuable". In our middleware, each task is submitted with its Purpose level of importance and the level of each subtask in the DIKWP hierarchy. The system can carry out intelligent energy management accordingly: priority is given to ensuring the resource supply of the relevant calculations of the Purpose layer and the Wisdom layer, because these have the most direct impact on the achievement of the overall goal; For redundant calculations at the data or information layers, you can choose to skip or reduce the frequency when energy is tight. If there is a conflict between global performance and energy consumption, the system can try to achieve an approximate effect with fewer calculations through a knowledge-layer approach, such as a predictive model, to save energy consumption. For example, in a distributed monitoring network, middleware can adjust the working mode of each camera node based on a purpose (such as "detecting abnormal events"): usually reduce the frame rate to save power, and only increase the frame rate when the knowledge layer model predicts a potential anomaly. Another example is on a mobile device, when the battery is low, middleware allows the device to keep only the core tasks that meet the primary purpose running, and postpone or offload non-critical semantic updates to the cloud. Through this semantic-driven energy mapping, we are expected to break through simple power optimization algorithms and achieve energy savings at the semantic level, that is, let every piece of energy be used on the cutting edge and on the calculation that makes the most sense for the current goal.

On the whole, DIKWP semantic middleware provides standardized and intelligent underlying support for our platform. It unifies the invocation of heterogeneous computing resources and ensures that various computing power modules work together around a common semantic goal. It improves interface transparency and adaptability, allowing the system to adjust the module composition in real time according to the purpose. It optimizes the balance of power consumption and performance, freeing up valuable computing resources for the most critical semantic tasks. For the industry, this middleware layer means that our brain-inspired basic software platform can quickly connect with chips and devices from different manufacturers to build a cross-domain semantic computing network. Whether you're deploying a large-scale cognitive middle office in the cloud or an intelligent agent on edge devices, this architecture will play a huge advantage.

The "BUG" Theory of Artificial Consciousness and the Adaptive Mechanism of Brain-like Fault Tolerance

The human brain is not perfect, on the contrary, our cognition is full of all kinds of "incompleteness", "uncertainties" and even "fallacies" – a phenomenon that Professor Yucong Duan vividly refers to as "bugs" in consciousness. Different from the traditional engineering concept that regards bugs as flaws and needs to be completely eliminated, the "BUG Theory" believes that consciousness is essentially an efficient but incomplete computing mode, and it is the tolerance and utilization of incompleteness that makes human beings uniquely creative and adaptable. This theory provides important implications for the fault-tolerant and adaptive design of artificial consciousness systems. This project will combine the BUG theory and introduce a series of mechanisms into the brain-like decision-making module, so that the system can still operate robustly in an imperfect information environment, and continuously approach the optimal through self-correction.

The first is a "incomplete tolerable" fault-tolerant strategy. Real-world data and knowledge are often incomplete or even contradictory, and a truly intelligent system should learn to make informed decisions with incomplete information, rather than shut down when it's missing. To this end, our software platform is designed with fault tolerance at all DIKWP layers: at the data and information layers, we introduce a semantic padding strategy for missing data or features, i.e., using context or prior knowledge to infer missing parts. For example, when a small number of readings from the sensor are lost, they can be filled by data interpolation at the approaching moment; When a text sentence is missing, it is completed according to the language model. At the knowledge and wisdom levels, for incomplete rules or models, we employ probabilistic reasoning and fuzzy reasoning to obtain the most likely conclusion, rather than requiring absolutely accurate derivations. If the relevant knowledge is missing, the system marks "uncertain" but continues the decision-making process and seeks external information if necessary (e.g. querying a database or asking a user). This is equivalent to creating a "fault-tolerant safety net" for the system: allowing a certain level of knowledge gap to exist, as long as the overall purpose can be advanced, it will not collapse completely because of a gap.

The fault-tolerant strategy is complemented by an adaptive computational annealing mechanism. The idea of simulated annealing algorithm is borrowed here, which means that when the system explores the solution space, it is allowed to introduce random perturbations or reduce the target requirements in stages to jump out of the local optimal or dead-end state. Specifically, when the AI fails to find a solution that satisfies the purpose of the Wisdom decision-making layer for many consecutive times, the middleware will trigger the "annealing" process: temporarily relax or change certain constraints, such as lowering the priority of secondary goals, giving the decision-making module more room for random attempts, etc., and then try to solve it again. This process is similar to a person who "changes his mind" and tries a different path after racking his brains to no avail. For example, in a complex planning problem, the system can first randomly try some seemingly suboptimal solutions to gain new insights, and then gradually converge back to a more optimal solution. This kind of active stochastic exploration can effectively avoid the goal never being achieved due to the narrow thinking at the beginning, and improve the probability of the system finding a feasible solution. Of course, the annealing process sets the "temperature" parameter to ensure that the amplitude of the disturbance is controllable and gradually converges over a period of time, avoiding the system from being unlimitedly random. The practice shows that when solving the uncertainty problem, the introduction of annealing-like mechanism can improve the robustness and global search ability.

Another key is semantic recovery path optimization. When the system deviates from its original goal due to wrong intermediate decisions or external interference, it needs an efficient way to self-correct and recover. We draw on the process of "introspection" in the human mind: when people realize that they have made a mistake, they usually look back to the previous steps, find the possible problem areas, and re-plan the way forward. In brain-like software, we achieve similar traceability by saving decision trajectories and semantic context information within the DIKWP framework. Specifically, the platform records the multi-layered status of DIKWP at the time of each important decision (e.g., what information was entered, what knowledge was used, and what was the basis for the judgment). When the end result is not ideal, the system can reverse-traverse these trajectories to find out which part of the output does not match expectations. For example, in a dialogue system, if the answer is not satisfactory to the user, the system can trace the historical context of the conversation (layer I) and the knowledge used (layer K), and find that it may be a common sense conflict that leads to the misunderstanding, so it adjusts the knowledge base or reinterprets the user's purpose, and then follows the new path of secondary reasoning to give an improved answer. This recovery process is not a simple retry, but a targeted adjustment based on semantic cues. For example, if it is identified that the problem is a misunderstanding of a technical term, the system partially updates the definition or context of the term, rather than overturning the entire reasoning. This makes the recovery process much more efficient. In addition, we also introduce the idea of multi-path parallel exploration: for key decision points, the middleware can back up the current state, and then try a variety of possible branch decision parallel simulations to move forward, compare which path has the best final effect, and then select the best path as the recovery path actually executed by the system. It's similar to a person considering different options to "take it one step at a time" and rehearse the results so as not to waste a lot of time on a dead end.

Through these strategies, our platform will be resilient and resilient to humans in the face of an imperfect world. As the "bug theory" reveals, it is those imperfections that create creativity. Rather than stagnating in pursuit of absolute perfection, our system allows for a certain level of "compromise" and "trial and error", but constantly self-correcting and learning in the process. Each fault-tolerant process, annealing attempt, and recovery optimization actually enriches the system's understanding of the problem, improves its knowledge base, and expands its policy base. Over time, the system will become more perfect in the midst of imperfections. This shift in design philosophy will transform AI from a fragile mechanical actuator to a robust autonomous agent – like an experienced human expert who can learn from mistakes and robust its mission in an uncertain environment.

Analysis of typical application scenarios

In order to verify the practicability of the above-mentioned DIKWP brain-like basic software platform, we selected a number of representative and forward-looking application scenarios for demonstration. These scenarios cover different fields from cognitive simulation, brain-computer interfaces, to next-generation communication chips, fully reflecting the platform's universal adaptability and core value. In each scenario, we'll show how the platform leverages the strengths of the DIKWP model, solves existing technical pain points, and creates new features or performance breakthroughs.

Scenario 1: A cognitive simulation platform based on DIKWP semantic middleware

Background: The cognitive simulation platform is designed to simulate the cognitive processes of humans or animals, and is used for cognitive science research, brain-inspired algorithm validation, and decision support for complex systems. For example, simulating the cognitive response of a human driver in a virtual environment to test the safety of autonomous driving systems; Or simulate the cognitive process in doctor-patient dialogue to train intelligent consultation assistants. Traditional simulation platforms often use preset rules or machine learning models to simulate cognition, but they lack a unified semantic framework, which makes it difficult to achieve high-level interpretability and flexible adjustment.

DIKWP Platform Solution: With our basic software platform, you can build a multi-agent cognitive simulation environment. Each agent, such as a simulated driver, doctor, or patient, runs an instance of the DIKWP model internally. In this way, the agent's behavior in the virtual environment will be determined by the five layers of cognitive states within it. For example, the visual input of a simulated driver enters as "data", the "information" layer extracts key points such as traffic signals and road conditions, the "knowledge" layer contains his driving experience and traffic rules, the "wisdom" layer makes driving decisions based on it, and the "purpose" layer represents his driving purpose (e.g., arriving on time, safety first, etc.). These cognitive processes are transparent to the researcher because the platform allows us to examine the content of each agent at each DIKWP layer at any given moment – the equivalent of reading "what it is seeing, thinking, and wanting" in real time. This provides valuable white-box analysis capabilities for cognitive science experiments. Instead of only observing behavioral outputs, we can now drill down into internal processes. For example, we can verify whether an improvement allows the agent to learn new rules at the knowledge layer, whether the decision at the wisdom layer is more in line with its purpose, and so on, thus greatly improving the explanatory and persuasive power of the simulation.

Dynamic interaction and adaptation: Multiple agents interact with each other through the platform's semantic middleware. The "Wisdom" decisions output by one agent are performed by the environment and may become the "data" input of another agent. Since all agents share a unified DIKWP semantic protocol, they can naturally understand the meaning of each other's actions. For example, in the doctor-patient dialogue simulation, both the doctor AI and the patient AI are driven by the DIKWP model, so the doctor's words are not only used as string input by the patient model, but also rise to the patient's knowledge and Purpose layer to influence their next response. Previously, Tang et al. have tried to use the DIKWP theory to simulate the generation and flow of data, information, knowledge, wisdom, and purpose in the brains of doctors and patients in the consultation scenario, and construct a deep cognitive interaction model to better explain the dialogue process. Our platform makes it easier to implement such complex interactions. In addition, each agent has the ability to adapt to faults supported by the BUG theory. This means that during the simulation process, even if there are anomalies (such as sensor errors, blind spots in the character's knowledge), the agent can continue to talk or operate with uncertainty like a human, and will not be stalled in the event of an accident. This brings the simulation closer to reality and allows us to test the behavior of the system under various anomalies.

Significance and value: Based on the DIKWP platform, cognitive simulation is no longer a simple replication of behavior, but a real mechanistic simulation - we are simulating the cognitive process itself. This has significant implications for cognitive science research and can be used to test new theories about consciousness and decision-making. For example, what will be the macro performance of agents with different "Purpose" settings in a cooperative game? For example, if the intelligence layer capabilities of individuals are gradually enhanced in the swarm agent, how will group decision-making evolve? These scientific questions can be easily experimented on this platform. For engineering applications, this simulation can help us uncover potential problems with complex AI systems. Since we can examine the internal state of each step in the simulation, when the simulation finds that an AI makes an inappropriate decision in a particular situation, we can trace the cause, is it a perceptual error? Knowledge deficits? Or is it a Purpose conflict? In turn, the algorithm can be improved or training data can be provided in a targeted manner. In conclusion, the DIKWP cognitive simulation platform will become an important tool for AI system research and development, shorten the commissioning cycle and improve safety and reliability.

Scenario 2: Brain-Computer Interface (BCI) control system

Background: Brain-computer interface technology is dedicated to the direct interaction between the brain and computers/mechanical devices, such as allowing paralyzed patients to control robotic arms with their minds and type with brain waves. Current BCI systems rely primarily on pattern recognition of brain signals (e.g., EEG, ECOG) and mapping them to a limited number of instructions. However, due to the noisy brain signals that vary from person to person, the system often misjudges or fails to recognize. In addition, the user's purpose is often a complex high-level target, and the existing BCI decoding can only capture simple instructions, lacking an understanding of the user's real needs.

DIKWP Platform Solution: Our platform can provide an end-to-end semantic decoding and control framework for BCI systems. Specifically, the DIKWP model is introduced into the BCI signal processing pipeline: the brain signal is input as the lowest level of data D, and the meaningful brain signal features (which may correspond to the user's initial thoughts, such as the original impulse of "left" and "right" shift) are extracted through the information layer, and then the higher-level semantic purpose is deduced by the knowledge layer combined with the user's background (such as known preferences and current situation), and finally the Wisdom layer formulates the action plan, Purpose Layers ensure alignment with user expectations. For example, a patient wearing a BCI wants to grab a cup on the table and drink water. Traditional BCI may need to train brain signals such as "raise your hand" and "clench your fist", and then send them out in turn. The DIKWP-based BCI system can try to directly decode the user's goal (P-layer: "drinking water"), and use the knowledge layer (cup position, grasping common sense) and the Wisdom layer to plan a series of actions to adaptively complete - such as adjusting the position of the wheelchair first and then reaching out to grab the cup. If there is a deviation in a certain step (the patient's distraction causes the signal to be abnormal), the system can automatically tolerate errors, such as pausing the action or continuing the execution based on the camera's visual compensation information, and will not stop due to a momentary signal instability.

Purpose Adaptive and human-machine integration: The introduction of the DIKWP framework can also allow the BCI system to form a closed-loop interaction with users. Traditional BCI is one-way: from the brain to the machine. We can use the Purpose layer to enable two-way communication: when the system is unsure about the user's Purpose, it can generate feedback (through visual cues or stimuli) to guide the user to clarify the Purpose. This is somewhat similar to a brain-controlled robot asking the user, "Are you going to take a cup?" "It's just that the way to do it is more subtle (e.g., using the P300 brain wave response test to test user confirmation). In addition, because we structure the user's brain state information into the DIKWP layers, the system can accumulate the user's personalized cognitive model. For example, if a user has low attention (W-layer efficiency) in the morning, he may reduce complex operations or strengthen the Purpose layer validation in the morning. If it is found that the user's brain signal interference is large when listening to music, the knowledge layer will mark the low decoding reliability at this time, so that the Wisdom layer can adopt a more conservative action strategy to ensure security.

Security and ethical considerations: BCI is also supported by the platform's built-in semantic middleware and security mechanisms. For example, we can set the value criterion of the Purpose layer: even if the user is impulsive (the brain signal shows that he wants to do a dangerous action), the system can politely refuse or ask for additional confirmation after evaluating the risk at the Wisdom layer. This is akin to a "conscious" assistant who weighs the consequences of actions rather than blindly executing them, thus avoiding potential harm. For example, in the medical scenario, the BCI system needs to be highly reliable, and any decoding error can have serious consequences. Our bug-tolerant mechanism can rather not do it when the system is uncertain: when the confidence level of brain signal interpretation is insufficient, the system would rather pause the action and guide the user to try again, rather than rush to execute the questionable instruction. At the same time, multi-channel information fusion (such as combining eye movements, electromyography, etc.) can also improve the accuracy through the knowledge layer. These designs have significantly improved the security and user trust of BCI controls.

Overall, with the DIKWP platform, BCI systems will be upgraded from "signal control" to "purpose understanding". It's no longer just listening to the brain's superficial electrical signals, it's really trying to read the user's thoughts and get things done. This not only improves the user experience (the user does not need to learn too many rigid instructions, the system will "guess" and help complete the purpose), but also expands the application boundaries of BCI - from simple device control to more complex daily assistants, and even brain-controlled creation. For users with disabilities, this is an important step towards the goal of "everything comes true".

Scenario 3: A low-power semantic scheduling chip framework for 6G communication

Application background:
In the future 6G era, the communication paradigm will shift from the pursuit of bit rate to the focus on semantic efficiency and task orientation. The so-called semantic communication, that is, the transmitting end only sends information that is meaningful to the receiving end to reduce the transmission of useless data. This is seen as a key way to break through the limits of traditional Shannon and improve communication efficiency. At present, the industry has experimentally verified the great potential of semantic communication: by introducing semantic understanding on 4G links, it can achieve an effective transmission capability comparable to that of 6G. However, implementing semantic communication requires an end-to-end system with an understanding of the content being transmitted and awareness of the user's purpose, which is far beyond what current communication protocols can support. Therefore, there is a view that 6G will integrate communication and AI to achieve a "semantic network", in which the network can intelligently schedule resources according to business semantics.

DIKWP platform solution:
Our brain-like basic software platform is very suitable for the development needs of 6G semantic communication. Specifically, a semantic scheduling chip can be deployed in a communication terminal or base station, and the underlying layer is supported by DIKWP middleware, which is responsible for introducing semantic intelligence in all aspects of the communication process. For example, on the sender side, after the data is processed by the DIKWP model, only the information that reaches a certain knowledge/wisdom value will be actually sent through the channel, reducing redundancy. On the network side, middleware schedules spectrum and power resources according to a global purpose (such as ensuring low latency for a certain type of emergency service), and gives more resources to data streams with high semantic value. To give an example: in the 6G application for augmented reality (AR), the purpose of the user terminal may be to "pay attention to the road navigation information ahead", then the terminal can use the DIKWP framework to conduct local analysis of the camera video (extract information such as roads and road signs and other information layer I, combined with the map knowledge K layer), and only send the extracted navigation semantics (such as "turn right at the next intersection, go 200 meters" such as W-layer decision) to the server, without uploading the entire HD video stream. This not only saves bandwidth, but also reduces the energy consumption of the terminal. Based on this semantic information, the receiver synthesizes it locally and gives a simple prompt to the user to complete the task.

Low-power advantage:
Since DIKWP middleware has an energy-semantic mapping function, it can dynamically weigh communication consumption and semantic value. When the network load is high or the terminal battery is low, the system automatically cuts the transmission of data with low semantic benefits. For example, for an environmental monitoring sensor network, in the stationary state, each node only needs to periodically upload summary information (the abnormal indicator extracted by the Wisdom layer), and a large amount of raw data (D layer) is not uploaded. Only when the middleware determines that an anomaly may occur (for example, a sudden change in the data of a station requires more detailed information K layer confirmation), the policy will be temporarily relaxed to allow the node to upload more raw data for analysis. Once confirmed, throttling mode is resumed. This on-demand transmission strategy greatly reduces the usual power consumption and ensures accuracy at critical moments.

Purpose-driven network resource scheduling:
In 6G scenarios, not only terminal devices but also the network itself can have a purpose. For example, the purpose of an emergency communication network is to "maximize the delivery rate and timeliness of important messages". Our platform can introduce this network-level purpose into the scheduling algorithm: the middleware of the base station or routing device continuously evaluates whether the currently transmitted stream serves the network purpose, and prioritizes the weighted scheduling for important semantic traffic. Traditional QoS may be based on service type or user level, while semantic QoS truly schedules resources according to the importance of the information itself. Academician Zhang Ping and other experts pointed out that modern semantic communication will promote the paradigm shift of communication, and its excellent performance has been verified in scenarios such as unmanned driving and satellite Internet of Things. Our platform further enables semantic cognitive networks through Purpose scheduling: the network is able to "understand" what information is most critical in the current environment and make decisions similar to those of a human dispatcher. This is extremely important for the many scenarios that 6G needs to support (such as massive IoT, industrial control, and real-time interaction in the metaverse).

Expected results:
The DIKWP semantic intelligence communication chip can realize semantic processing at the hardware level, greatly reduce the end-to-end communication volume and latency, and improve reliability. According to reports, semantic communication is expected to improve the network by an order of magnitude in terms of capacity, coverage, and efficiency. More importantly, it changes the status quo of "transmitting useless data", allowing the network to transmit more meaningful data and truly achieve on-demand communication, which also makes the network resource allocation more green and energy-saving. Imagine the future of 6G terminals: equipped with our semantic scheduling chip, it can intelligently adjust the uplink and downlink data strategy according to the user's current task, so that the terminal traffic utilization and battery life are significantly increased, and the network congestion is also reduced. This will be a model of the deep integration of communication technology and artificial intelligence.

Through the analysis of the above three scenarios, we can see the broad application prospects of the DIKWP brain-like computing basic software platform. Whether it is in the field of AI cognitive research and simulation, human-computer interaction control, or new intelligent communication, this platform can play a unique role. This is due to its core semantic modeling, purpose-driven, adaptive fault tolerance and other innovations, which can be embedded in various complex systems to become the underlying "Wisdom Engine". For industry investors, these scenarios also represent huge market opportunities in the future: cognitive simulation and intelligent middle platform will serve the digital transformation of enterprises, BCI and auxiliary medical equipment market prospects are considerable, and 6G semantic communication is a major strategic highland for future communication infrastructure. Once the platform is implemented in these areas, it will generate significant business value and social benefits.

Conclusions and prospects

To sum up, the basic software platform for brain-inspired computing based on DIKWP and artificial consciousness theory builds a basic software solution for the future intelligent era through original architecture design and core technology integration. It uses the DIKWP five-layer semantic network as the skeleton to make the AI system have a clear cognitive level and internal interpretability. Integrate the TRIZ innovation paradigm to ensure that the platform itself can continue to evolve and meet the needs of users; The use of purpose-driven programming and scheduling greatly improves the software's adaptability to complex and dynamic tasks. The semantic middleware layer is introduced to break down the barriers between heterogeneous hardware and high-level semantics, and realize the optimal utilization of resources and energy semantic mapping. Combined with the "BUG" fault tolerance theory, the system is endowed with robust performance in a human-like imperfect environment; And it shows its versatility and transformative potential in a number of typical scenarios. It can be said that this platform reflects a deep insight and leadership into the next generation of artificial intelligence technology trends from architecture, methods to applications.

Our ultimate goal is not only to develop a software system, but to create a set of brain-like computing infrastructure platforms with independent intellectual property rights and suitable for a variety of scenarios, and become a strategic base support for China in the field of brain-like artificial intelligence. Globally, AI technology is entering a critical period of competition for the underlying architecture and ecosystem. Through the implementation of this project, we are expected to take the lead in establishing an industrial system of semantic intelligence, including cognitive middle-end solutions for different industries, software stacks for collaborative optimization with domestic brain-like chips, standard tools for artificial awareness evaluation and security governance, and so on. This will fill the gap in the field of artificial consciousness and brain-like operating system in China, and greatly enhance the core competitiveness of China's AI industry.

At present, Professor Yucong Duan's team has built a patent pool covering 99 domestic and foreign authorized invention patents and 15 PCT international patents around the DIKWP model, forming a solid intellectual property moat in the field of AI cognition and artificial awareness. The combination of these patented technologies is highly systematic and innovative, giving us a unique voice and cooperation value in the competition and cooperation with international technology giants. With the boom of large models such as ChatGPT and Claude, there is an urgent need for all walks of life to solve the AI black box and improve credibility. The "DIKWP semantic operating system" and related technologies are expected to be quickly integrated into existing platforms through patent licensing or joint development, and become the core components of next-generation AI systems, with a huge potential market size. We believe that with the help of industrial capital, these cutting-edge achievements will accelerate from the laboratory to industrial application and be transformed into real productivity.

Finally, we warmly appeal to the cooperation and support of the industry. As the report said, these innovative technologies are still in the stage of large-scale transformation, and there is an urgent need for funds, resources and platforms to accelerate their implementation. The project team will uphold an open and cooperative attitude, give full play to the advantages of multidisciplinary interdisciplinarity, and work with investors and industry partners to improve product solutions and expand the application ecosystem. In the next 3-5 years, we plan to gradually establish the standard status of DIKWP brain-like basic software in China through demonstration application verification and large-scale promotion, and promote it to the international market to participate in the competition and cooperation of the global artificial intelligence underlying architecture.

Looking forward to the future, an era of universal language and wisdom is coming. From smart factories to smart cities, from personal assistants to swarm intelligence, strong semantic understanding and autonomous decision-making are needed everywhere. We have reason to believe that the brain-inspired computing basic software platform based on the DIKWP artificial consciousness model will become one of the key foundation technologies leading this era. The paradigm of "semantic intelligence" it represents will enable China to seize the commanding heights in the new round of AI transformation and create a new world-leading industrial ecosystem. With the joint efforts of all parties, we look forward to this innovative achievement with independent IP taking root and bearing fruitful results, contributing to China's wisdom and creating historical opportunities for promoting artificial intelligence from perceptual intelligence to cognitive intelligence, and from tool-based AI to autonomous AI.