Report on Reform Suggestions for China's Civil Procuratorial
通用人工智能AGI测评DIKWP实验室
Report on Reform Suggestions for China's Civil Procuratorial Supervision System
International Standardization Committee of Networked DIKWPfor Artificial Intelligence Evaluation(DIKWP-SC)
World Academy for Artificial Consciousness(WAAC)
World Artificial Consciousness CIC(WAC)
World Conference on Artificial Consciousness(WCAC)
(Email: duanyucong@hotmail.com)
I. Current Status and Problem Analysis of the Procuratorial Supervision System
1. Historical Evolution and Legal Framework
China's civil procuratorial supervision system was gradually established starting from the late 20th century and has now formed a legal framework based on the Civil Procedure Law and relevant judicial interpretations. According to Article 14 of the Civil Procedure Law, the People's Procuratorate has the right to exercise legal supervision over civil litigation, which is the constitutional positioning basis for civil procuratorial supervision. In terms of specific institutional design, procuratorial organs mainly have two supervision methods for effective civil judgments: first, protest (kangsu), i.e., lodging a protest against effective judgments or rulings that are definitely in error to initiate a retrial procedure; second, procuratorial suggestions, including retrial procuratorial suggestions (suggesting the court to retry and correct) and other procuratorial suggestions (targeting illegal acts by judicial personnel or execution activities). Current regulations such as the Rules of the People's Procuratorate on Civil Litigation Supervision provide detailed provisions on the applicable conditions and procedures for the above methods. For example, regarding effective judgment documents, procuratorial organs can propose measures such as retrial procuratorial suggestions or protests; for illegalities in trial procedures and execution activities, supervision is carried out by submitting procuratorial suggestions. Overall, the civil procuratorial supervision system has covered the entire process of civil litigation, but it still faces many difficulties in practical operation.
2. Difficulties in System Operation
First, it is difficult to initiate supervision and intervention is scarce. Compared with first-instance and second-instance procedures, the number of civil procuratorial supervision cases is relatively small, and procuratorial organs are extremely cautious and rare in exercising their right to protest. According to statistics, from 2021 to 2023, the number of applications for supervision of effective civil judgments accepted by procuratorial organs nationwide was 70,000-80,000 per year, but the actual number of cases where protests were lodged was less than 7%, and retrial procuratorial suggestions were less than 15%. Overall, the proportion of procuratorial organs taking supervision measures against effective judgments has remained below 20%. Procuratorates hold a "restrained" attitude towards initiating supervision, intervening only when there are judicial errors that definitely need correction, to maintain the res judicata of judgments and judicial authority. This reflects the principle of respecting judicial stability on the one hand, but on the other hand, it also leads to a significant number of supervision cases applied for by parties failing to enter the substantive review procedure. The conditions for initiating supervision are too harsh, and the threshold for procuratorial intervention is high. Parties often reflect that "applying for procuratorial supervision is harder than reaching the sky." Even if a supervision application is submitted, prompting the procuratorial organ to file a case for review and essentially initiate the supervision procedure is the most critical yet most difficult link.
Second, the impact of supervision effects is limited. Since procuratorial organs are cautious in intervening in effective judgments and the overall intervention rate is low, the macro influence of civil procuratorial supervision on correcting wrongful cases is insufficient. Admittedly, once procuratorial organs take protest measures, their effects are often significant—the retrial reversal rate of protest cases and the adoption rate of procuratorial suggestions in recent years have been as high as about 90%. This shows that once procuratorial supervision intervenes, it can often correct erroneous judgments and maintain judicial justice. However, restricted by the limited number of initiations, the role of the procuratorial supervision mechanism in preventing and correcting unjust and wrongful cases on a large scale has not been fully exerted, and there is a gap between institutional supply and the demand for judicial justice. Some parties reflect that even with obviously unfair judgments, due to the lack of intervention by procuratorial organs, there is ultimately a lack of avenues for correction. It can be seen that the uplifting effect of procuratorial supervision results on the entire judicial justice is still insufficient, and it is necessary to further enhance the coverage and actual influence of supervision.
Third, the post-judgment supervision mechanism is fragmented. The connection and closed-loop management of the current civil procuratorial supervision process are not sound enough, and there are phenomena of ineffective execution of supervision opinions and "considering the matter settled once a protest is lodged." After procuratorial organs lodge a protest or issue a retrial procuratorial suggestion, if the court does not adopt it or the retrial judgment still has obvious errors, there is a lack of unified norms on how to follow up. Some grassroots procuratorates have different understandings and inconsistent standards regarding the applicable circumstances and methods of "follow-up supervision," leading to the possibility that initial supervision opinions may be ignored, affecting the seriousness of supervision. For example, regarding situations where the court fails to provide feedback on procuratorial suggestions within the time limit, or where the retrial judgment is still problematic, there are confusions and divergences in practice on whether to initiate further supervision and how to grasp the "degree" of supervision. This weakness in the latter half of the supervision chain reflects the fragmentation of the post-judgment supervision mechanism: there is a lack of effective linkage and closed-loop management between links such as protests, procuratorial suggestions, and follow-up supervision by procuratorial organs. The result may lead to erroneous judgments not being corrected in time, or procuratorial suggestions being shelved without supervision for implementation, deviating from the original intention of the supervision system design.
II. Semantic Judicial Adaptability Theory of the DIKWP Model
1. Overview of DIKWP Model
The DIKWP model was proposed by Professor Yucong Duan and is an extension of the classic DIKW (pyramid) cognitive model. The traditional DIKW model includes four levels: Data, Information, Knowledge, and Wisdom, used to describe the evolution from raw data to wise decision-making in the cognitive process. DIKWP adds the highest level "P layer"—Purpose/Intent based on this. Through this extension, the model forms a five-layer semantic architecture of "Data → Information → Knowledge → Wisdom → Purpose." Introducing the Purpose layer endows the cognitive process with goals and motivations, making the entire model Purpose-driven, realizing a closed loop from data perception to decision execution. In other words, DIKWP emphasizes that the processing of any data and knowledge should be connected to predetermined purposes, explicitly considering purpose motivations during decision-making, and achieving connection at the semantic level. This innovation provides a structured, globalized cognitive framework for artificial intelligence and complex decision-making, enabling AI to possess goal-oriented cognitive abilities similar to humans. Some studies also refer to the top layer as the "Practice" layer to highlight the operational link of putting wise decisions into practice, but its essence is still to govern the entire cognitive process with purpose and intent.
2. Meaning of Five-Layer Semantic Structure in Judicial Field
The five-layer semantic structure of the DIKWP model is highly compatible with the judicial reasoning process. In the judicial cognitive scenario, legal provisions, case facts, and judgment purposes can be mapped to different levels of the DIKWP model for semantic expression respectively, and through feedback iteration between layers, it ensures that the reasoning process connects "legal purpose" and "case objectives" (i.e., the purpose of fairness and justice to be achieved by judicial judgment). Specifically:
Data Layer (D): Corresponds to the raw fact materials and evidence data of the case. For example, parties' statements, witness testimonies, documentary evidence, physical evidence, expert opinions, etc., all belong to the basic data in judicial cognition. In traditional trials, these scattered factual materials need to be extracted and organized. Under the DIKWP framework, the Data layer requires comprehensive collection and formalized representation of case fact elements, laying the foundation for subsequent reasoning.
Information Layer (I): Corresponds to the specific case information obtained after processing case data, as well as intermediate reasoning associating facts with legal norms. For instance, processing numerous evidences as recognized legal facts, drawing case timelines, or corresponding facts to specific legal provisions (such as matching behavior with constitutive elements). This layer associates evidence with legal points, forming structured case information. Semantic transformation at the Information layer enables AI to have a panoramic "self-awareness" of each case, clarifying core issues and known facts.
Knowledge Layer (K): Corresponds to legal knowledge and judicial experience rules. This includes statutory provisions, judicial interpretations, guiding cases, precedent rules, and legal theoretical concepts. The Knowledge layer provides the normative framework and professional knowledge background for explaining the case. At this layer, the model integrates legal rules with past cases, building a knowledge graph between case facts and legal norms. For example, linking the fact pattern of a "loan case" with corresponding legal regulations, burden of proof rules, and judgment rules of past similar cases. This is equivalent to embedding a legal expert system into AI reasoning, making the reasoning conform to legal provisions and precedent experience.
Wisdom Layer (W): Corresponds to value judgments and strategic choices in judicial adjudication, i.e., legal wisdom. In complex cases, there are often multiple legitimate judgment paths. The Wisdom layer reflects the judge's comprehensive consideration of fairness, justice, and social effects. For example, the balanced application of legal principles and the use of discretion. In DIKWP, the Wisdom layer semantics represent the value trade-offs implied in the judgment process. Through this layer, the model can evaluate the legitimacy of different handling schemes: for example, whether a scheme violates the principle of honesty and credibility, or whether it conforms to legislative purposes and social common sense. The Wisdom layer enables AI to possess preliminary value judgment capabilities, avoiding mechanical application of legal provisions that violates the essence of justice.
Purpose Layer (P): Corresponds to the final purpose and intent of judicial activities, i.e., realizing individual case justice and maintaining the authority of the rule of law. In legal reasoning, this layer manifests as the overall goal of the judgment—including both individual justice (realization of parties' legitimate rights and interests) and higher-level purposes such as maintaining institutional fairness and regulating social behavior. The Purpose layer gives direction to the entire model: all lower-level data and knowledge processing must ultimately serve the goal of realizing judicial justice. Therefore, in AI judicial applications, the Purpose layer can be set as an objective function optimizing judicial justice. When different goals (such as efficiency and fairness) conflict, the model can solve for the Pareto optimal solution through semantic mathematics, or weight and compromise various goals to seek value balance. The Purpose layer ensures that AI reasoning does not deviate from the track of rule of law values, thereby achieving so-called "semantic justice"—that is, the judgment result is semantically consistent with the spirit of the law and social public will.
Through the above five-layer structure, the DIKWP model semantically models the legal cognitive process panoramically, enabling the AI system to have clear semantic depictions and logical associations for the case from facts to jurisprudence, and then to value goals. The value of this hierarchical semantic framework in the judicial field lies in: it provides adaptability of cognitive semantics—it enables AI to accurately understand the meaning and applicable boundaries of legal provisions, and also allows AI to perceive the tension and fit between case facts and legal rules, and consider value purposes during decision-making, thereby improving the transparency and explainability of the judicial process. In short, the semantic mathematics model of DIKWP provides a rigorous methodology for judicial AI: transforming implicit conceptual judgments in traditional legal judgment processes into explicit, multi-level semantic graph expressions, promising to enhance the fairness of the judgment process and the verifiability of reasoning.
III. Case Reconstruction Analysis: Application of DIKWP Model in Civil Wrongful Case Supervision
To visually demonstrate the application value of the DIKWP model in civil procuratorial supervision, we reconstruct the DIKWP semantic graph of a desensitized "Private Lending Wrongful Judgment Retrial Supervision Case" to analyze how procuratorial organs identify semantic tension in wrongful cases and accurately initiate supervision to realize "Semantic Justice." This case prototype is taken from a civil lending dispute protest case handled by procuratorial organs (Procuratorial Case No. 154), reflecting a typical process where erroneous evidence determination led to a wrongful judgment, which was corrected through procuratorial supervision.
1. Case Background (Data Layer)
Between the Plaintiff (Lender) and the Defendant (Borrower), there was a private loan of 1.4 million yuan. The Plaintiff provided a loan receipt as evidence, claiming the loan was not returned; the Defendant defended that the loan had been fully repaid and presented a so-called "Repayment Note" as proof. The content of the Repayment Note was "The loan has been fully repaid, the loan receipt is torn up by myself," signed by the Lender with a date, but the Lender had passed away before the loan maturity date. In the first instance, both parties fiercely disputed the authenticity of the Repayment Note: the Defendant commissioned an appraisal agency which concluded that the signature and fingerprint on the note were made by the Lender; later, the court commissioned two other authoritative appraisal centers for re-appraisal, resulting in a conflict—one appraisal deemed the signature not the Lender's handwriting and the fingerprint unconfirmable, while the other deemed the signature written by the Lender. Under the circumstance of inconsistent appraisal conclusions, the first-instance court adopted the appraisal opinion favorable to the Defendant, determining the Repayment Note to be authentic and valid, and thus ruled to reject the Plaintiff's claim for repayment. The second-instance court and the provincial high court's retrial both upheld the first-instance opinion, ruling the loan had been repaid and the Plaintiff's claim was baseless. The Plaintiff's appeal against the retrial judgment was finally submitted to the procuratorial organ, seeking supervision and correction.
The elements of the Data Layer in this case include: 4 loan receipts (Plaintiff proving existence of loan), 1 Repayment Note (Defendant proving repayment), 3 judicial appraisal reports (conclusions contradicting each other), and bank fund transaction records (possible traces of repayment), etc. These raw evidence data constitute the nodes of the DIKWP model's Data Layer. The prosecutor reviewed the entire case file to comprehensively obtain the above evidence, laying the data foundation for subsequent analysis.
2. Case Information Extraction (Information Layer)
At the Information Layer, procuratorial organs process the evidence semantically, extracting key facts and revealing relationships between evidences. First, the prosecutor used AI tools to automatically parse file materials, sorting out the factual context and focus of controversy. The dispute in this case centered on: Was the loan actually paid off? The authenticity of the Repayment Note and the delivery of funds were decisive facts. Information extracted by the prosecutor included: loan receipts proved the existence of the loan and no return record upon maturity; the Repayment Note was the sole evidence proving payment, but its form and content were suspicious; the Defendant claimed to have repaid 1 million yuan via bank acceptance bill discounting but failed to submit any objective vouchers like bank statements. Through semantic graph association, the Information Layer linked the core evidence "Repayment Note" with the fact to be proved "Fact of Repayment," marking its credibility as doubtful; it associated the claim of "Repayment via Acceptance Bill" with bank transaction record elements, identifying evidence missing on this chain. Meanwhile, information on contradictory appraisal reports was also extracted: there were two mutually contradictory handwriting appraisal conclusions, and anomalies in the appraisal procedure (such as unclear source of appraisal samples, appraiser contacting the Defendant privately). These Information Layer contents were recorded in the DIKWP graph in the form of semantic nodes and relationships, allowing the prosecutor to clearly see: the information chain supporting the conclusion "Loan Repaid" had breaks and anomalies, with numerous doubts.
Especially important is that the Information Layer revealed signs of semantic tension: on one hand, the Defendant provided the Repayment Note and appraisal opinion as supporting information for "Repaid"; on the other hand, multiple signs indicated this support was contradictory and insufficient (isolated evidence without corroboration, flaws in the note, repayment method leaving no objective traces). This contradiction was depicted as a conflict relationship node in the semantic graph at the Information Layer, suggesting the conclusion was inconsistent with evidence semantics.
3. Application of Legal Knowledge (Knowledge Layer)
Entering the Knowledge Layer, the procuratorial organ mapped the case to corresponding legal norms and rules for review. Relevant legal knowledge in this case included: conditions for retrial protest in Civil Procedure Law (effective judgment is definitely in error), requirements for lending evidence in the Supreme Court's Provisions on Several Issues Concerning the Application of Law in the Trial of Private Lending Cases, and general principles of evidence law such as Standard of High Probability and Allocation of Burden of Proof rules. In the Knowledge Layer semantic graph, the prosecutor focused on the following rule nodes: First, according to evidence rules, in lending cases where the debtor claims repayment, they should bear the burden of proof for the fact of repayment; and when the evidence presented is obviously doubtful, the claim cannot be deemed established. Second, according to judicial interpretations, vouchers proving loan delivery or repayment often need to form a complete evidence chain; isolated evidence with flaws should require reinforcement, otherwise it is insufficient to determine facts. Third, the Civil Procedure Law stipulates that effective judgments lacking evidence for fact-finding or applying law erroneously belong to "definitely in error" situations, and protest supervision can be initiated. The prosecutor performed semantic matching between this legal knowledge and the information extracted from the case: finding that the judgment in this case likely violated requirements for burden of proof and sufficiency of evidence—the court determining repayment established based only on doubtful isolated evidence was an obvious impropriety in application of law. The Knowledge Layer analysis further aggravated the semantic tension previously hinted at by the Information Layer: the semantics of legal norms conflicted with the case judgment result. According to Knowledge Layer evaluation, the original judgment in this case was definitely in error in fact-finding and application of law, satisfying conditions for protest.
4. Consideration of Judicial Purpose (Wisdom/Purpose Layer)
The Wisdom Layer and Purpose Layer reflect the prosecutor's scrutiny of judicial fairness and justice goals. From the Wisdom Layer perspective, the original judgment in this case was obviously unfair: the Lender had passed away and was unable to protect rights, while the Borrower provided suspicious evidence and received support, leading to the creditor's interests not being realized. This contradicts the value orientation that justice should protect honesty and trustworthiness and punish fraud. Especially given the significant doubt regarding the authenticity of the Repayment Note, the court's direct adoption of the appraisal conclusion favorable to the debtor without requiring further fact-finding obviously violated common sense and principles of fairness. The value judgment of the Wisdom Layer held: if this judgment were allowed to take effect, it would send a wrong signal to society, damaging both the legitimate rights and interests of the parties and judicial credibility.
The Purpose Layer further focused on the judicial purpose of this case: on one hand, to correct individual injustice and safeguard the legitimate creditor rights of the Lender; on the other hand, aiming to maintain rule of law integrity, preventing fabricators of evidence from profiting through judicial procedures, thereby maintaining normal financial order and judicial authority. At this level, the supervision Purpose of the procuratorial organ was highly consistent with the purpose of the rule of law—through supervision and correction, realizing individual case justice and manifesting the rule of law value of honesty and credibility. The DIKWP semantic model showed its guiding function of the "Practice Layer" at this moment: the target node at the highest layer of the model pointed to the requirement of "realizing semantic justice," i.e., making the judgment conclusion consistent with factual truth and legal spirit. Based on this, the prosecutor determined the necessity and direction of supervision intervention.
5. Supervision Decision under Semantic Graph
Synthesizing the analysis of various layers of the DIKWP graph, the procuratorial organ identified obvious semantic tension and reasoning breaks in this case—the evidence system relied upon by the original judgment did not match legal requirements, and the conclusion was semantically inconsistent. Specifically: The Data Layer had isolated evidence with flaws failing to form an evidence closed loop; the Information Layer showed the "Repaid" conclusion lacked objective support and evidences were contradictory; the Knowledge Layer pointed out the court's error in burden of proof allocation, violating the high probability standard; the Wisdom Layer judged the ruling result violated fairness and common sense; the Purpose Layer required correcting the error to uphold justice. The inconsistency appearing from bottom to top across the entire DIKWP hierarchy meant a failure of the judicial judgment in semantics. Such a case is a typical situation where procuratorial supervision should intervene. Thus, based on the model analysis conclusion, the procuratorial organ made a supervision decision to lodge a protest. The Henan Provincial Procuratorate proposed the protest to the Supreme People's Procuratorate, which, after review, agreed to initiate the protest procedure and submitted the case to the Supreme People's Court for retrial.
In the retrial protest opinion, the procuratorial organ closely followed the problems revealed by the DIKWP analysis, focusing on two points for supervision grounds: (1) The original judgment's fact-finding seriously lacked evidence support: the Repayment Note was isolated evidence with major doubts, and the debtor could not produce other payment vouchers; the original judgment determining repayment based on this violated the principle of sufficiency of evidence; (2) The original court applied the law improperly: it failed to correctly allocate the burden of proof according to law, not requiring the debtor to reinforce proof when evidence was doubtful, constituting an obvious error in application of law. Through the above semantic-level argumentation, the protest opinion clearly reproduced the loopholes in the case reasoning chain and clarified the semantic causes of the judgment error. After trial, the Supreme People's Court adopted the procuratorial organ's opinion, revoking the original judgment in the 2019 retrial and ordering the defendant to repay the loan principal and interest. The wrongful case was finally corrected, and the lender's rights were legally confirmed. This supervision result confirmed the effectiveness of DIKWP model-assisted analysis—the procuratorial organ accurately pinpointed the judicial fallacy through the semantic graph, achieving the supervision effect of "Breaking the Target of Injustice with the Arrow of Semantics."
6. Realization of Semantic Justice Assisted by Model
In this case, the DIKWP model acted like a "Semantic Microscope," helping the prosecutor magnify key details and logical contradictions in the trial process. From evidence data collection to information association, then to knowledge comparison and value measurement, the model screened semantic consistency layer by layer. When a break or tension appeared at a certain level (e.g., evidence not matching conclusion, rules deviating from judgment), the model issued a timely warning. This based procuratorial supervision on comprehensive, rational semantic analysis rather than relying solely on experience or intuition, greatly improving the precision and persuasiveness of supervision intervention. Through the DIKWP reconstruction of this case, we see: The semantic mathematics model provided a brand-new perspective for wrongful case supervision. Prosecutors no longer just discover unfair results after the fact, but can explain "WHY the judgment is wrong"—because the judgment is untenable in semantic logic, and evidence semantics do not match legal semantics. This semantic-based injustice detection fits the essential function of "supervision and correction" of procuratorial organs and reflects the powerful support of theoretical innovation for practice.
IV. Model Calculation and Semantic Propagation Mechanism: Auxiliary Deduction of Supervision Judgment
Using the DIKWP semantic model to assist procuratorial supervision is not limited to the conceptual description level but can also quantifiably deduce the supervision judgment process through Weighting Calculation and Propagation Algorithms. This mechanism involves assigning weight values to semantic nodes and paths, calculating the magnitude of semantic tension by propagating weights along the graph, thereby providing a quantitative reference for prosecutors to judge whether a case is "definitely in error." Its core idea is: let every piece of evidence, every legal rule, and every value goal participate in reasoning in the form of "weighted semantic nodes." When the support intensity for the conclusion node is insufficient or there is high conflict tension, it indicates a potential wrongful case requiring supervision initiation. Specifically:
In the DIKWP graph, elements at different levels can be viewed as nodes, and each node can be assigned a certain weight value based on credibility or importance. For example, at the Data Layer, each piece of evidence can be assigned a Reliability Weight (e.g., witness testimony might be assigned a lower weight due to subjectivity, while objective documentary evidence gets a higher weight); at the Knowledge Layer, each legal rule can be assigned a Normative Strength Weight (e.g., mandatory rules have higher weight than arbitrary rules). The basis for weighting can be a combination of statistical analysis and expert experience: AI systems can preliminarily estimate the general credibility of certain types of evidence in proving specific facts by learning from massive cases, while combining prosecutors' professional judgment to calibrate weights. This node weighting endows the semantic graph with "numerical semantics," laying the foundation for subsequent calculations.
2. Path Weight Propagation
Reasoning processes such as evidence supporting conclusions and legal norms applying to facts can be abstracted as paths between nodes in the semantic graph. When an evidence node points to a conclusion node through a series of intermediary nodes (Facts → Legal Elements, etc.), its initial weight will propagate and convert along the path level by level, influencing the conclusion. Propagation can use methods like product or weighted accumulation: for example, if the reliability of evidence is 0.8 and the strength of the applicable rule is 0.9, the combined impact can be evaluated as 0.72. For situations where multiple evidences jointly support the same conclusion, Aggregation Functions (such as weighted sum or Bayesian update) can be used to synthesize weights from different paths to obtain the Support Degree of the conclusion node. Similarly, if there is evidence refuting the conclusion, negative weights can be assigned and propagated to weaken the conclusion support degree. In this way, each potential judgment conclusion obtains a Probative Strength Value accumulated from bottom to top on the graph. This propagation calculation process ensures Explainability: how much each piece of evidence and each reasoning chain contributes to the conclusion is traceable. At the same time, by quantifying the complex proof process, it assists prosecutors in intuitively grasping the sufficiency of the probative force of the whole case.
3. Semantic Tension Function
Semantic tension can be defined as a measurement function of the degree of conflict or contradiction in the semantic network. When a set of semantic paths supporting proposition A coexists with a set of paths supporting counter-proposition ¬A, the difference and conflict in accumulated weights in different directions generate tension. We can design a Tension Function T, for example, making it a function of the difference or ratio between the "positive support degree" and "negative support degree" of the conclusion node. If the T value is very low (support degrees differ greatly), it indicates high semantic consistency; conversely, if the T value is close to 0 or in an indeterminate state, it indicates tense conflict within the semantic system, requiring key attention. When semantic tension exceeds a certain critical threshold, it can be viewed as a Warning Signal for the correctness of the judgment. Taking this case as an example, the support for the conclusion "Loan Repaid" mainly came from the highly flawed Repayment Note (low weight), while the support for the opposite direction "Loan Unpaid" implied facts like loan receipts and lack of fund vouchers (relatively higher weight). Model calculation might result in little difference in support degrees or even the counter-evidence dominating, thus producing a high semantic tension value. This aligns with intuitive judgment: the case has obvious doubts, and the original judgment is likely wrong. This conflict is presented quantitatively through the tension function, enabling prosecutors to "measure" injustice. Notably, the tension function considers not only evidence support strength but can also incorporate Inter-layer Tension: for instance, a mismatch between requirements at the legal level and satisfaction at the factual level will also be reflected in some tension indicator of the model. In short, the semantic tension function provides a formalized method to detect the consistency of judgment reasoning; significant tension often implies the judgment conclusion cannot be self-consistent in semantic logic, indicating a necessity for supervision.
4. Assistance in Determining "Definitely in Error"
Based on the above node weighting and tension calculation results, procuratorial organs can obtain a Quantified Judgment Reference. The specific practice is: set several indicator thresholds; whenever the model shows the conclusion support degree is far below the legal requirement value, or semantic tension is higher than the preset threshold, it can be preliminarily judged that the original judgment may be "definitely in error," thereby initiating further manual review. Model calculation plays a Warning and Filtering role here. For example, assuming the model stipulates evidence support degree needs to reach above 0.8 to be considered clear facts, while a case's support is only 0.4, it prompts the prosecutor to focus on it. Or if the tension value is close to 1 indicating equal force of evidence on both sides or even counter-evidence dominance, yet the judgment favors one side, it implies a possible wrongful judgment. These indicators can be calibrated corresponding to traditional legal standards (such as the "preponderance of evidence" principle). Through model assistance, prosecutors can have a more objective and intuitive grasp of supervision application cases, reducing deviations from subjective judgment based solely on experience. It must be emphasized that model calculation cannot and should not replace humans: it only provides reference basis, and whether to protest ultimately requires the prosecutor's comprehensive discretion combining law and facts. But this auxiliary decision-making mechanism can greatly improve work efficiency and scientificity. As some views point out, Prosecutors should lead, AI assists, and AI analysis conclusions need independent review and gatekeeping by procuratorial personnel. Therefore, in practical application, mechanisms like AI-assisted decision logs and objection annotation should be established to record the interaction between model output and prosecutor decisions throughout, with careful review and correction by the undertaker for model conclusions that do not conform to common sense. This human-machine collaborative process is also constantly optimizing the model itself: prosecutor feedback will be used to adjust weight calculations, making the model increasingly conform to judicial reality.
In general, the Semantic Weighting Propagation Algorithm provided by the DIKWP model gives procuratorial supervision wings of "quantitative analysis." In complex and difficult cases, the model can help quantitatively evaluate the evidence basis and logical consistency of judgments, providing technical support for the judgment of "definitely in error." This exploration fits the current development direction of digital procuratorate, integrating subjective judgment into objective data analysis, helping to overcome limitations of vague standards and experience-based decision-making in past supervision practices. Of course, model calculation is only an auxiliary tool and should not be superstitiously followed mechanically. But as long as it is well utilized under the premise of Prosecutor Leadership, it will become a sharp tool for enhancing supervision precision and reasoning strength, making procuratorial supervision more scientific, transparent, and just.
V. Reform Suggestion System Based on the DIKWP Model
Centering on the institutional dilemmas and theoretical tools analyzed above, this paper proposes a Systematic Reform Suggestion set, hoping to better integrate the DIKWP semantic model into civil procuratorial supervision practice, enhancing supervision quality and efficiency and theoretical innovation value. Policy suggestions cover technical support, mechanism improvement, and concept renewal:
(1) Construct AI-assisted supervision semantic analysis tools: Develop an artificial intelligence analysis platform based on the DIKWP model to assist prosecutors in semantic review of effective judgments. In specific case handling, introduce large model applications similar to "DeepSeek." This tool should possess Rapid Scanning and Element Extraction functions for file materials, automatically parsing judgment documents and evidence materials, extracting key information such as focus of controversy and judgment points. Meanwhile, combining with the DIKWP semantic network, support Automatic Association of Evidence and Legal Articles, interfacing with laws, regulations, and case databases to hint at relevant legal application risks and similar case judgment results. For example, in false litigation supervision, let the AI model analyze massive civil case data, automatically screening out clues like suspected "Routine Loan" or "False Evidence" for prosecutor review; in civil execution supervision, the model can monitor and identify illegal situations like "Sealing up properties exceeding the value of the subject matter" or "Non-compliant termination of execution," helping prosecutors discover supervision clues on a large scale. In practice, local large models deployed by Hubei procuratorial organs can already batch analyze recent civil case data, accurately discovering supervision problem clues like "violation in applying summary procedure for service by publication" and pushing them to business systems for filing. This AI-assisted mode liberates procuratorial supervision from "Human Wave Tactics," realizing a transformation from passive supervision to active supervision. It is suggested that the Supreme People's Procuratorate lead the R&D of a unified civil procuratorial supervision intelligent platform, promoting its application in procuratorial organs nationwide, and clarifying the principle of "AI assistance, Prosecutor leadership": AI provides analysis opinions and similar case references, and its conclusions are decided for adoption after prosecutor review. Through technology empowerment, the breadth of supervision clue discovery and case handling efficiency can be greatly improved, reducing the rate of missed cases, ensuring this supervision procedure truly becomes a reliable avenue for parties' "last remedy."
(2) Establish a semantic comparison database for similar cases to enhance homogeneous supervision capabilities: Establish a "Civil Supervision Case Semantic Graph Database," collecting typical supervision cases handled by procuratorial organs everywhere, semantically annotating each case with the DIKWP model to form a structured knowledge base. This semantic comparison database should cover information such as cause of action, basic facts, focus of controversy, supervision points, judgment results, as well as semantic evaluation indicators like evidence chain completeness and legal application doubts. Through big data analysis, procuratorial organs can conduct Semantic Comparison between newly received supervision application cases and processed cases in the database, retrieving whether there are "Similar Cases with Same Errors." Once the semantic graph of a new case is found highly similar to a corrected wrongful case (e.g., both due to missing certain evidence leading to wrongful determination), it prompts the prosecutor to refer to precedent experience, maintaining consistent supervision standards. This helps Standardize Supervision Scale, preventing inconsistent supervision selectivity due to regional or personnel differences. In addition, this semantic database can serve as training corpus for AI models, constantly enriching the model's understanding of legal semantics and recognition ability of error patterns. For example, by learning from past supervision cases, the model can summarize common types of judicial trial errors (insufficient evidence, procedural violation, improper application of law, etc.), improving sensitivity in discovering similar semantic contradictions. In technical implementation, experience from judicial big data platform construction can be borrowed, combining knowledge graphs with machine learning to realize Intelligent Retrieval and Comparison of Similar Cases. Simultaneously, it is suggested that the Supreme People's Procuratorate establish a Guiding Supervision Case release system, regularly publishing various typical supervision and correction cases, and entering their semantic analysis points into the database for reference by procuratorial organs nationwide. This will enhance the supervision fairness of "Same Case, Same Standard," making parties' expectations of supervision results more predictable.
(3) Promote semantic assessment for civil procuratorial early intervention mechanism: Under the existing supervision mode, procuratorial organs often intervene in supervision only after the court's final judgment. However, for certain major, difficult, or suspected false litigation cases, exploring "Procuratorial Early Intervention" is a mechanism for error correction and trust enhancement. Specific suggestion: Select some pilot areas to introduce procuratorial organ semantic assessment based on the DIKWP model during the first or second instance of cases. During court trials, if the AI model identifies Potential Error Risk Points (such as doubts on key evidence, legal application deviation), the analysis report can be sent to the procuratorate at the same level for reference. When deemed necessary after assessment, the procuratorate can remind the court to pay attention to correction in the form of Procuratorial Suggestions in advance, or immediately initiate a supervision plan after the judgment. The value of this mechanism lies in: moving the tentacles of procuratorial supervision forward, killing signs of wrongful cases in the bud as much as possible, reducing the generation of unjust and wrongful judgments. The development of large model technology makes this idea operable—the model can sort out evidence chains, distinguish priorities, and hint at potential supervision points in real-time during case trials. For example, for new types of difficult cases, the model can instantly retrieve thousands of cases and legal articles for judges' and prosecutors' reference, avoiding wrongful judgments due to cognitive blind spots. The early intervention mechanism can also be piloted in High-incidence Areas of False Litigation, such as private lending and divorce disputes, screening abnormal litigation patterns through AI, with prosecution and court collaborating to identify false lawsuits, protecting judicial resources and parties' rights from the source. It should be noted that early intervention must be based on law and proceeded with caution, preventing interference with normal judicial independence. Therefore, procuratorial organs can be granted limited procedural participation rights, such as observing, reviewing files, and offering opinions and suggestions, rather than fully intervening in the trial. With the rational analysis of the DIKWP model as a basis, early intervention will be more persuasive and fair, and easier for judicial organs to accept, exploring new paths for procuratorial supervision for Smart Justice.
(4) Establish a feedback system for identifying semantic tension in supervision cases: To overcome the current problem of fragmented post-judgment supervision, it is suggested to build a national unified Procuratorial Supervision Feedback Platform, implementing full-process monitoring for every supervision case from application acceptance to case closure, paying special attention to the adoption of supervision opinions and result execution. The platform should utilize the DIKWP model to conduct semantic comparison analysis between supervision opinions and court responses, automatically identifying "Residual Semantic Tension": i.e., when procuratorial suggestions are not adopted or retrial judgments still have doubts, the system can perceive that the semantic conflict between judgment reasons and procuratorial opinions still exists, and issue timely warnings. For example, in a case where the procuratorate issued a retrial procuratorial suggestion believing the original judgment's fact-finding was unclear, but the court rejected the application with insufficient reasons, the system would mark the case as high-risk and retain it in the supervision sequence. The "Feedback System" can realize Dynamic Tracking of every supervision case by interfacing with the court judgment document network and the procuratorial organ case management system. Once conditions for follow-up supervision are found (such as court failing to reply to procuratorial suggestions overdue, retrial reversal still obviously improper), the system automatically reminds the superior procuratorate to intervene for review, achieving "Follow up whenever necessary, no omission or abandonment." At the same time, at the case management level, a special ledger should be established for unadopted initial supervision opinions, ensuring procuratorial personnel do not miss subsequent supervision opportunities due to personnel changes or negligence. The feedback system should also have built-in evaluation and learning mechanisms: statistically analyzing the adoption rate and effect of supervision opinions in various places, feeding experience back to the DIKWP model, continuously Optimizing Model Parameters, and improving the model's ability to predict court reactions. In the long run, this system will be deeply integrated with the construction of digital procuratorates as part of smart procuratorial infrastructure. By combining technology with institutions, it truly endows procuratorial supervision with "Rigidity and Resilience"—having both rigid closed-loop management to prevent supervision from becoming a formality, and resilient continuous improvement to constantly improve supervision levels.
VI. Risk Assessment and Future Prospects
1. Risk Assessment of Model Application
While introducing the DIKWP semantic model to assist procuratorial supervision, potential risks must be carefully assessed to prevent "intelligent assistance" from alienating into "intelligent interference." First is the Risk of Misjudgment. If model training data is biased or algorithms are flawed, false warnings (misjudging correct judgments as wrong) or missing true wrongful cases may occur. To this end, the wide representativeness of model training data should be ensured, especially strengthening balanced learning of unjust, false, and wrongful cases and normal judgment samples, and calibrating model output through continuous White Box Testing, checking whether semantics at each layer are coherent and reasonable. Second is the Risk of Value Deviation. AI models ultimately lack human conscience; no matter how precise the "semantic tension" they measure is, they may not fully reflect humanistic care and value trade-offs in justice. Therefore, it is necessary to prevent models from overly pursuing logical consistency while ignoring fairness and ethics. For example, sometimes mechanically reducing tension (balancing conflict) might sacrifice the interests of vulnerable groups, which is impermissible. In response, on one hand, core rule of law values such as the principle of priority of fairness and protection of the weak should be implanted in the model's Purpose layer, converting Judicial Conscience into objective functions for constraint; on the other hand, persist in Human-Machine Collaboration, letting prosecutors conduct value scrutiny and adjustment of AI suggestions during decision-making. Finally, there are Technical Security and Liability Risks. Model application needs to guarantee data security and privacy protection, deploying AI in the procuratorial organ's intranet environment to prevent leakage of case information. At the same time, clarify the subject of responsibility—any supervision decision is legally borne by the prosecutor, and AI should not become a "second subject." Establish mechanisms for AI-assisted decision-making full-process tracing and objection annotation; for conclusions provided by the model, prosecutors record reasons for acceptance or rejection for ex-post review. Through these multiple measures, the risk of AI misleading can be minimized, making it truly a reliable assistant rather than a source of hidden dangers.
2. Prospects for Deep Integration
Looking at the development of the procuratorial supervision system, it is closely related to technological changes and the evolution of cognitive paradigms. In today's digital age, procuratorial organs face the dual challenges of massive information and complex cases, urgently needing AI to enhance supervision efficiency. The DIKWP model, as an emerging semantic cognitive framework, is expected to play a key role in the construction of Digital Procuratorate and Smart Justice. Its prospects can be viewed from the following aspects:
Enhancing Supervision Quality and Efficiency: Empowered by DIKWP, prosecutors can quickly extract key issues from massive files, obtain associated legal articles and similar cases within a limited time, thereby making supervision judgments faster and more accurately. This will effectively alleviate the contradiction of "many cases, few people," allowing prosecutors to devote more energy to substantive review and deep thinking rather than consuming it on tedious file reading. Taking domestic pilots as an example, provincial procuratorates have already deployed large model auxiliary systems, exploring over 500 business application scenarios, realizing full-process intelligent support from case acceptance screening, evidence review to document drafting. It is foreseeable that future civil procuratorial supervision, supported by semantic models, will usher in Multidimensional Improvement in Quality and Efficiency: supervision clue discovery will be more comprehensive, supervision reason explanation more sufficient, and supervision decisions made more timely.
Advancing Smart Procuratorial Affairs: The DIKWP model fits the concept of "Business-led, Technology-supported" in smart procuratorial affairs. It not only provides tools but also reshapes the thinking paradigm of procuratorial case handling—Human-Machine Collaboration, Semantic Leadership. As the model integrates into daily procuratorial case handling, a Virtuous Cycle will form: the model continuously optimizes the legal knowledge base and reasoning patterns by analyzing massive closed cases; procuratorial organs improve internal training and guidelines accordingly, enhancing personnel capabilities. For instance, when the model captures similar dispute points frequently appearing in numerous cases, the procuratorate can improve similar case supervision guidelines or issue judicial suggestions to prevent common problems. The model will also self-iterate in continuous human-machine interaction, constantly improving accuracy and credibility. This co-evolution will revitalize the procuratorial supervision system—maintaining the value judgment role of humans as judicial subjects while fully leveraging the machine's powerful advantages in data processing and knowledge integration, realizing an efficiency leap of "1+1>2."
Leading Institutional Innovation: In the long run, the application of the DIKWP semantic model may prompt innovation in the civil procuratorial supervision system itself. On one hand, the supervision procedure will be more transparent and visible: semantic graphs make supervision reasons clear at a glance, and the reasoning of supervision decisions will be more scientific and detailed, conducive to enhancing understanding and adoption of procuratorial suggestions and protests by all parties. On the other hand, the scope of supervision is expected to expand and deepen: with AI, procuratorial organs can perform duties more actively and proactively, such as the aforementioned early intervention and intelligent screening of false litigation, thereby enriching the realization forms of legal supervision. Furthermore, at the institutional level, new normative requirements may be spawned, such as evidence standard quantification and judgment document structuring, to coordinate with semantic analysis, which in turn promotes the integrated innovation of litigation systems and technology. It is expected that in the near future, Data Procuratorates and Smart Courts will collaborate closely, and the legal professional community using semantic AI models to solve judicial problems will become the norm: prosecutors and judges will call upon professional large models like using search engines to quickly obtain information support spanning regulations-cases-theories. This change not only improves efficiency but also reconstructs the judicial cognitive mode, injecting new momentum into the modernization of the rule of law.
In conclusion, the DIKWP semantic mathematics model provides solid theoretical support and technical paths for the reform of the civil procuratorial supervision system. Facing the opportunities and challenges of the information age, we have reason to believe: by introducing this model into civil procuratorial supervision, China's procuratorial organs will gain a head start on the track of Digital Justice, achieving a leapfrog upgrade of supervision functions. Its significance lies not only in correcting more wrongful cases and guaranteeing individual case justice but also in exploring a new path of Semantic-led, Intelligence-empowered judicial reform, which will also play a demonstrative leading role for global AI rule of law practice. Letting the light of semantics illuminate the path of the rule of law is a vision worth expecting for future civil procuratorial supervision. As argued in this report, only with active and proactive role positioning, scientifically complete institutional mechanisms, and prudent innovation in technology application—complementing and integrating with each other—can China's civil procuratorial supervision welcome innovation amidst integrity, achieve leaps in reform, and earnestly shoulder the heavy responsibility of Maintaining Judicial Justice endowed by the Constitution and laws.
Gao Xiya: "A Few Things About Applying for Civil Procuratorial Supervision - Case Handling Notes", Tiantong Law Firm, 2024.
Zhou Huikuan: "Improving Mechanisms to Solidify Civil Procuratorial Follow-up Supervision", in Procuratorial Daily, Theory Edition, September 20, 2025.
Supreme People's Procuratorate: "Typical Cases of Civil Litigation Supervision by Procuratorial Organs (Batch 38)", Released in 2022 (Procuratorial Case No. 154, etc.)
Yucong Duan et al.: "Application of DIKWP Model in Legal Semantic Reasoning", ScienceNet Blog, 2025.
Yucong Duan et al.: "On Building a 'Semantic Sovereignty' Driven Smart Judicial Pilot System", Zhihu Column, 2024.
"AI + Procuratorate: Legal Supervision Also Has 'Three Heads and Six Arms'", Website of Enshi City People's Procuratorate, March 25, 2025.
Shen Yang: "The Trial Road of Intersection between Law and AI", in Procuratorial Daily, Digital Procuratorate Special Topic, January 8, 2025.
Zhang Yu: "Application and Development of Artificial Intelligence in Criminal Prosecution", in Procuratorial Daily,
玩透DeepSeek：认知解构+技术解析+实践落地
人工意识概论：以DIKWP模型剖析智能差异，借“BUG”理论揭示意识局限
人工智能通识 2025新版 段玉聪 朱绵茂 编著 党建读物出版社
邮箱|duanyucong@hotmail.com
