On the Hybrid Nature of ABPMS Process Frames and it... | AI Research

This paper introduces a conceptual framework for AI-Augmented Business Process Management Systems (ABPMS). The authors propose a "process frame"—a hybrid structure that defines the boundaries within which an AI system can operate autonomously. By combining procedural models (which define specific sequences of steps) with declarative models (which define rules and constraints), the framework allows for a more flexible and realistic representation of business processes that can adapt to different types of knowledge, such as best practices or legal requirements.

A Hybrid Approach to Process Modeling

The core idea is to move away from rigid, single-model representations. Instead, the authors suggest that an ABPMS should use a collection of smaller, overlapping specifications. Some of these may be procedural, like a traditional flowchart, while others may be declarative, like a set of rules. By adopting an "open-world assumption," the system treats each specification as a constraint that only applies to the activities it explicitly mentions. This allows different parts of a business process to be governed by different rules without requiring a single, monolithic model that covers every possible scenario.

Enabling Framed Autonomy

The goal of this hybrid representation is to enable "framed autonomy." In this model, the AI system is not just a tool for executing tasks; it acts as an agent that can make decisions within defined boundaries. Because the process frame is permissive—meaning it only restricts what is explicitly forbidden—the AI has the freedom to optimize performance or handle unexpected situations, provided it stays within the "fences" set by the procedural and declarative rules.

Implications for Automated Discovery

The authors highlight that discovering these process frames automatically from historical data is a significant challenge. Traditional discovery tools are usually designed to find a single, complete process model. Because the proposed process frame is a hybrid of different types of models, the authors propose mapping discovered declarative constraints into procedural fragments. This approach helps simplify complex models by identifying "pockets of rigidity"—specific areas where strict procedural rules are necessary—while keeping the rest of the system flexible.

Key Considerations

While this framework offers a powerful way to manage complex, autonomous systems, the authors note that it is not a "plug-and-play" solution. Automated discovery is unlikely to produce a perfect process frame on its own due to noise in real-world data. Instead, the authors suggest that discovery should be an initial step, followed by human adjustment to ensure the system’s boundaries align with organizational goals. Furthermore, because these specifications can interact in complex ways, developers must be careful to account for hidden dependencies that might arise when combining multiple rules.