LLM Mediation Workflow for Consistent Multi-Party Decisions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing general artificial-intelligence LLMs face challenges in efficiently and cost-effectively providing fair decisions or recommendations among multiple parties, as they require significant human expertise and time to learn and summarize information.
Innovation Solution
A method and apparatus utilizing a general artificial-intelligence LLM as a hub to process multiple information modalities, including images, video, and text, to provide decisions or recommendations by serializing and labeling party inputs, prompting for supplementary information, and ensuring consistency before outputting unified results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a LLM is used to mediate multi-party decisions, then decision speed and cost reduction are improved, but information completeness and consistency may deteriorate
Solution Approach 1:
The system performs preliminary actions by prompting parties to supplement information before the LLM makes its decision. Steps S6-S7 identify missing information, and steps S8-S11 detect and resolve inconsistencies, ensuring complete and consistent information is provided to the LLM before final decision-making occurs.
Solution Approach 2:
The system implements feedback mechanisms where the LLM evaluates submitted information and identifies what is missing or inconsistent. The system then feeds this evaluation back to the parties, prompting them to supplement or correct their information before the final decision is rendered.
2Reliability
If a LLM processes all party information, then neutrality and fairness are improved, but processing time and computational resources may worsen
Solution Approach 1:
The system segments the information processing into distinct phases: information collection (S2), supplementary information prompting (S6-S7), consistency checking (S8-S11), and final decision-making (S5). This segmentation allows the LLM to process information in manageable stages rather than all at once, reducing overall processing time while maintaining neutrality.
Solution Approach 2:
The system performs preliminary filtering and organization of party information before it reaches the LLM for final decision-making. Steps S3-S4 serialize and label party inputs, while steps S6-S11 pre-process the information by identifying gaps and inconsistencies, so the LLM receives cleaned, organized data that requires less processing time.
3Measurement precision
If the LLM requests supplementary information from parties, then decision accuracy is improved, but the number of interaction rounds increases
Solution Approach 1:
The system performs preliminary identification of missing information in steps S6-S7 before the final decision is made. By using the LLM to analyze submitted information and identify specific gaps, the system can target supplementary information requests precisely, reducing the number of rounds needed compared to requesting all possible information upfront.
Solution Approach 2:
The system requests only the partial information that is actually missing or insufficient, rather than requiring all possible information from all parties. The LLM evaluates each party's submission and identifies only the specific supplementary information needed, avoiding excessive interaction rounds while maintaining decision accuracy.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Social attribute is an important attribute of human beings. Among two or more individuals, various problems often arise. One important way to solve such problems is to have someone mediate decision making or advice making. For example, the profession of judge is a typical profession for mediation and adjudication among two or more subject parties. A person who can implement decision making or advice making mediation is usually the one having a high level of knowledge and prestige, and the cost of cultivating and retaining such a person is generally high. Moreover, understanding and summarizing information from all parties for research and judgment is another time-consuming task. General artificial intelligence (AI) large models emerging in today's human society possess extremely high levels of knowledge and neutrality, and it would be worthwhile to work on using the general AI large models to carry out the above task to obtain a fair result as soon as possible and greatly reduce the cost.