TL;DR
This paper introduces SMILE and MARVEL, two software development environments that aim to provide intelligent assistance. SMILE achieves this without AI by hardcoding its knowledge, while MARVEL builds upon SMILE's architecture with a knowledge-based approach and a focus on "opportunistic processing" to automate tasks. The work highlights the critical role of granularity in defining software entities and processing tools for effective intelligent assistance.
What Problem Does This Paper Address?
The paper addresses the long-standing dream of an "intelligent assistant" for programmers, a concept first articulated by Winograd in 1973. Traditional programming environments often treat tools as passive agents, requiring explicit user commands. While AI research has explored knowledge-based systems for programming, these have often been complex and not easily integrated with existing tools. The authors aim to bridge this gap by developing environments that actively participate in the software development process, understand the context of user actions, and automate routine tasks, thereby increasing programmer productivity and reducing errors. They explore two distinct approaches: one that avoids AI entirely (SMILE) and another that leverages AI for more sophisticated assistance (MARVEL).
Methods
The paper presents two systems:
- SMILE: This system provides "intelligent assistance without artificial intelligence." It achieves this by hardcoding its knowledge of software objects and the programming process directly into the environment. SMILE offers a "fileless environment," automatically invokes tools, and manages software development activities without relying on AI techniques.
- MARVEL: This system is built on a general architecture for intelligent assistance, combining an "objectbase" (storing software objects and their relationships) and a "model of the software development process" (defined by rules, metarules, hints, and strategies). MARVEL's key contribution is "opportunistic processing," where the environment automatically performs activities when their preconditions are met and their results are likely to be needed soon, rather than waiting for explicit user commands. This involves forward and backward chaining among rules, guided by metarules. The paper also emphasizes the importance of "granularity" in defining software entities and tools for effective processing.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- "Intelligent Assistance without AI" is Achievable: SMILE demonstrates that a sophisticated, seemingly intelligent assistant can be built by carefully designing the environment and hardcoding its knowledge, without resorting to complex AI techniques. This offers a pragmatic approach for certain environments.
- The Power of "Opportunistic Processing": MARVEL's core innovation is opportunistic processing. By automatically performing tasks when the opportunity arises (e.g., compiling a module after a change, even if the user hasn't explicitly requested it), the environment can significantly reduce user burden and ensure consistency.
- Granularity is Crucial: The paper strongly emphasizes that the "granularity" of software entities (how finely they are broken down into objects) and the granularity of tools (how they process these entities) profoundly impacts the intelligence, friendliness, and performance of a programming environment. Choosing the right granularity is a key design decision.
- Objectbase and Process Model are Prerequisites: Both SMILE and MARVEL rely on an objectbase to represent software entities and their relationships, and a model of the development process. This foundational structure is essential for any assistant to "understand" its actions.
- Separating Checking from Reporting: For feedback mechanisms (like error reporting), separating the act of checking from the act of reporting provides flexibility. Users can choose when and how they want to be informed, reducing intrusiveness.
Key Results
- SMILE: Successfully provided a fileless environment, automated tool invocation, and supported teams of programmers on projects up to 61,000 lines of code. It proved robust enough for production use.
- MARVEL: The paper outlines the architecture and design principles of MARVEL, focusing on its knowledge-based approach and opportunistic processing. While specific quantitative results are not detailed, the design aims to offer more sophisticated assistance than SMILE by understanding context and anticipating user needs. The discussion of granularity issues provides a framework for designing such systems.
Limitations and Future Work
- SMILE's Limitations: SMILE's knowledge is hardcoded and not extensible, meaning users cannot easily augment its understanding. Its user interface is also described as less advanced than what modern systems offer.
- MARVEL's Implementation Status: MARVEL is presented as an ongoing development. The paper focuses on its architecture and design principles, with implementation details and empirical results to follow.
- User Modeling for Strategies: MARVEL's strategies are currently selected by users, not autonomously chosen by the assistant. The authors suggest that additional research on user modeling is needed to enable the assistant to select strategies automatically.
- Extensibility: While MARVEL aims for extensibility through its rule-based model, the paper notes that SMILE was not extensible. The design of MARVEL aims to address this.
Summary updated