A multi-level model of collective decision-making and a negotiation method for IT project task and resource allocation

Abstract

The article addresses the transition from local, incomplete and potentially conflicting proposals of specialized artificial intelligence agents to an agreed management decision and an executable allocation of IT project tasks and resources. Separate use of voting, centralized arbitration or greedy assignment does not simultaneously provide decision stability, controlled autonomy and consistent allocation of interdependent actions. A multi-level model of coordination and collective decision-making by specialized AI agents has been improved. The model combines distributed monitoring, versioned formation of the feasible decision set, role-based and weighted quorum, robust aggregation, iterative stabilization, collective confidence assessment and regulated human oversight. Unlike one-shot voting, the model produces a traceable decision prepared for decomposition and two-phase dispatch. The negotiation method for IT project task and resource allocation has been further developed. It creates versioned lots and package bids, considers competence matching, temporal feasibility, execution cost, current workload, fatigue, assignment risk and performance guarantees, and jointly optimizes interdependent assignments. Context-dependent protocol selection, two-phase contracting and local renegotiation improve agreement stability and limit reconfiguration after a failure. The experimental evaluation comprised two parts. The collective decision model achieved an overall conflict resolution rate of zero point nine one, a decision stability index of zero point eight seven and an autonomous dispatch rate of zero point eight one. In resource allocation, the proposed method reduced mean project duration from one hundred forty-two to one hundred twenty-four days, lowered the Gini coefficient to zero point one eight, reduced overload frequency to four percent and decreased critical-path failure probability to zero point one seven.