TL;DR
This paper argues that the development of "smart" application systems, particularly knowledge-based systems, has significantly advanced software engineering by introducing and refining techniques for managing "exploratory programming." The core contribution lies in adapting traditional software engineering methodologies to accommodate the inherent uncertainty and iterative nature of building systems where requirements are not fully understood upfront.
What Problem Does This Paper Address?
Building "smart" application systems, often referred to as expert systems or knowledge-based systems, presents a unique challenge: the initial requirements are frequently vague and aspirational. Unlike traditional software projects with well-defined specifications, these systems require an "exploratory process" to discover the necessary functionality, knowledge, and processing methods. This exploratory nature, characterized by building successive prototypes and incorporating feedback, clashes with the predictability expected in software engineering regarding cost, timing, and outcomes. The paper addresses the problem of how to manage this exploratory programming effectively within a software engineering framework, ensuring progress and maintaining a degree of predictability despite the evolving nature of the project.
Methods
The paper doesn't introduce new algorithms or techniques in the traditional sense. Instead, it focuses on adapting and extending existing software engineering management methodologies. The key approach is to integrate "exploratory programming" into the conventional software development lifecycle. This involves:
- Maintaining and Evolving Specifications and Project Plans: While initial specifications and plans are often vague, the paper emphasizes that they are still crucial. The novelty lies in understanding that these documents are expected to evolve significantly throughout the project. They serve as a "common agreement" among stakeholders, documenting what has been learned and what remains uncertain.
- Stepwise Stable Project Plans: The core idea is that while the overall project plan evolves, the plan for the current prototyping step should be stable. This allows developers to focus on the immediate task without constant disruption.
- Frequent Project Reviews: Regular reviews, including prototype demonstrations, are central to the process. These reviews provide a structured opportunity for all interested parties (sponsor, experts, users, developers) to provide feedback, identify issues, and revise their agreements.
- Clear Procedures for Updates and Communication: Establishing clear procedures for updating specifications and plans, and for informing all stakeholders about these changes, is vital for managing expectations and ensuring buy-in.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Exploratory Programming is a Valuable Methodology: The paper highlights that exploratory programming, driven by the needs of AI systems, is not just an AI-specific technique but a valuable methodology for any system development where functional requirements are uncertain.
- The Importance of Evolving Agreements: The traditional view of fixed specifications is challenged. The paper posits that for exploratory projects, the "agreement" itself is dynamic and evolves as understanding deepens.
- Predictability Through Iterative Stability: While the project as a whole is unpredictable, achieving predictability for each step through stable, short-term plans is key to managing the process.
- The Project Manager as an "Agreement Evolver": The role of the project manager shifts from simply executing a fixed plan to actively managing the evolution of the common agreement among stakeholders.
Key Results
The paper doesn't present quantitative results in terms of accuracy or performance metrics. Its key results are qualitative and methodological:
- Integration of Exploratory Programming: The primary outcome is the successful integration of exploratory programming as a manageable technique within software engineering.
- Development of Management Guidelines: A set of guidelines has emerged from building knowledge-based systems that contribute to making exploratory programming a viable technique for systems with uncertain functional requirements.
- Improved Predictability: By adopting these guidelines, projects with inherent uncertainty can achieve a greater degree of predictability in cost, timing, and results.
Limitations and Future Work
The paper acknowledges that the "exploratory phase" is characterized by uncertainty and that the initial specifications and plans will change substantially. The primary limitation, inherent to the nature of exploratory programming, is the challenge of maintaining an "acceptable degree of predictability." The paper suggests that the techniques discussed are significant contributions to software engineering, implying that further refinement and adoption of these management practices are future directions. The broad applicability of these techniques to both AI and non-AI projects is also a point of emphasis for future consideration.
Summary updated