KR-20260065752-A - System and Method for AI Revenue Distribution Based on Real-Time Quantification of Professional Knowledge Contribution

Abstract

The present invention relates to a system and method for an AI model that utilizes expert knowledge data provided by an expert group (100) as a training dataset, wherein when generating a result according to a user request, the contribution of each expert knowledge data to the generation of the result is quantified in real time by synchronizing with the inference stage, and the profit is automatically distributed to the expert group in proportion to the contribution index. The contribution quantification engine (400) calculates individual data weights by executing a gradient-based influence measurement algorithm (Influence Functions or Shapley Value) in the inference stage without retraining the model, and derives a contribution index by aggregating them in units of expert group identifiers. The profit distribution processing unit (500) automatically settles available profit in a first and second distribution structure according to the contribution index, and the contribution history blockchain record unit (600) stores all contribution index calculation history and distribution details as immutable records in a distributed ledger.

Inventors

• 안범주

Assignees

• 안범주

Dates

Publication Date

20260511

Application Date

20260324

Claims (1)

1. A data collection unit that collects expert knowledge data generated from an expert group through a communication network; A model training unit that generates or updates an AI model specialized in a specific professional field by utilizing the above-mentioned collected expertise data as a training dataset; A contribution quantification engine that calculates the contribution of specific expert knowledge data that contributed to the generation of the results in the process where the above AI model generates results and generates revenue in response to user requests; and It includes a profit distribution processing unit that distributes a portion of the revenue generated from the AI model to the relevant expert group based on the contribution calculated above, The above contribution quantification engine is, An AI revenue sharing system based on expert knowledge contribution, characterized by calculating weights for individual expert knowledge data using an algorithm that measures the marginal contribution of each training data to the output value of the above AI model, and aggregating the weights based on an expert group identifier to derive a Contribution Index for each expert group.

Description

System and Method for AI Revenue Distribution Based on Real-Time Quantification of Professional Knowledge Contribution System and Method for AI Revenue Distribution Based on Real-Time Quantification of Professional Knowledge Contribution The present invention relates to a system and method for quantifying the contribution of expert knowledge data used in the training of an artificial intelligence (AI) model in real time during the output generation stage and automatically distributing the resulting profit to the expert group. More specifically, the invention relates to a technology for calculating a contribution index in real time for each AI model output result generated by applying an influence measurement algorithm based on Influence Functions or Shapley Value to the expert knowledge data provided by the expert group (100) in synchronization with the inference stage, and for transparently settling and distributing profit by linking this with a blockchain-based smart contract. The rapid advancement of generative AI technology has made it possible for AI models to generate expert-level results in fields requiring high levels of expertise, such as medical diagnosis, legal consulting, patent specification drafting, technical translation, and financial analysis. These AI models are built by learning from vast amounts of knowledge data accumulated over decades by experts in their respective fields. For example, medical AI learns from medical records and clinical findings written by tens of thousands of doctors; legal AI learns from judgments, contracts, and legal opinions produced by lawyers and judges; and technical translation AI learns from high-quality translation corpora generated by professional translators and technical experts. However, according to conventional technology, despite AI service providers generating massive profits by utilizing this expert knowledge data for training, there is a structural problem in which the experts who provide the source data do not receive fair economic compensation. This is the so-called "Data Sovereignty" issue, where there is virtually no technical means to individually track and compensate for the extent of an expert's contribution once their knowledge assets are internalized into the AI model. Conventional data contribution evaluation techniques include the Leave-One-Out (LOO) method and the Data Shapley method (Ghorbani & Zou, 2019). The LOO method measures performance changes after retraining a model excluding specific data, but it has limitations in that the cost of retraining increases exponentially when applied to large-scale AI models. The Data Shapley method applies the Shapley Value from cooperative game theory to evaluate the contribution of training data; while it offers excellent mathematical fairness, it has the problem of being impossible to apply in real time because it similarly requires iterative retraining on multiple data subsets. Furthermore, conventional data contribution evaluation technologies focus on calculating contributions at the data point level, failing to provide a specific method for deriving an aggregated contribution index at the level of the "expert group"—the ownership entity that serves as the actual unit for revenue distribution. Moreover, conventional technologies are limited to a post-hoc attribution method following the completion of the AI model's training phase, and thus fail to propose an automated structure for real-time contribution calculation and revenue distribution linked to the inference phase where actual service revenue is generated. Meanwhile, conventional technology has not sufficiently addressed technical measures to prevent falsification or alteration of contribution calculation results and profit distribution details, as well as to ensure transparency. When contributions and distribution details are managed on a centralized server, the possibility of data manipulation by platform operators cannot be ruled out, which becomes a fundamental cause of undermining the trust relationship between the expert group and the platform. Figure 1 is an overall network configuration diagram of an AI revenue sharing system based on expert knowledge contribution according to one embodiment of the present invention. FIG. 2 is a block diagram showing the detailed components and data processing flow of the contribution quantification engine (400) of the present invention. Figure 3 is a conceptual diagram illustrating the process of calculating contribution weights for individual expert data and aggregating them into contribution indices at the expert group level using a Shapley Value-based influence measurement algorithm in the present invention. FIG. 4 is a flowchart illustrating the operation flow of the profit distribution processing unit (500) and the contribution history blockchain record unit (600) of the present invention, and the process of automatic profit settlement and distributed ledger recording through a smart co