TL;DR
This paper introduces TELOS, a novel AI programming language built on PASCAL, designed to address the complexity of AI development by providing powerful abstraction mechanisms. TELOS aims to facilitate the creation of well-structured, modular AI models by offering specialized features for data, control, and procedural abstraction, ultimately enhancing programmer productivity and the theoretical exploration of intelligence.
What Problem Does This Paper Address?
Developing sophisticated AI systems often involves managing immense complexity, particularly in areas like knowledge representation, planning, and reasoning. Traditional programming languages, even those with some abstraction capabilities, can struggle to adequately support the diverse and often unconventional structuring needs of AI research. This gap makes it challenging to build modular, maintainable, and theoretically grounded AI models, hindering both practical development and the advancement of AI theory. The authors highlight that while many AI languages focus on embedding specific high-level constructs, TELOS instead prioritizes providing powerful, general-purpose abstraction mechanisms that allow these constructs (and others) to be defined and implemented with greater ease and flexibility.
Methods
TELOS is built as a superset of PASCAL and is implemented in PASCAL, leveraging existing language features while introducing novel ones. The core of TELOS's innovation lies in its abstraction mechanisms:
- Data Abstraction ("Capsules"): TELOS extends PASCAL's data typing with "capsules." These allow users to define problem-specific data types by encapsulating representation details, implementation logic, and primitive operations within a single unit. This promotes data hiding and allows objects to be characterized by their defining operations, rather than their internal structure. Features are also included to manage how these capsule-defined types integrate with TELOS's associative database.
- Control Abstraction ("Overseers"): TELOS introduces "overseers" to manage interprocess control-transfer and communication, enabling the definition of custom control regimes. This allows for flexible implementation of various AI control strategies, from simple backtracking to complex planning, with features for descriptively invoking controlled procedures and processes.
- Compromise for AI: TELOS attempts to bridge the gap between general programming principles and the often conflicting demands of AI programming. It achieves this through several compromises:
- Heterarchical Control: Allows for heterarchical control structures while centralizing communication and control-transfer.
- Conditional Interrupts: Supports conditional interrupts (demons) but unifies them under a general event handling scheme that explicitly manages signal, escape, and suspend events within defined scopes.
- Context Switching: Enables alternative data base versions (contexts) but mandates a structured form for switching between them (
InContext <context> Do <statement>).
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Abstraction as the Key: The paper strongly advocates for abstraction mechanisms as the primary tool for managing AI complexity, rather than hardcoding specific AI constructs into the language. This approach offers greater flexibility and reduces theoretical bias.
- Bridging Theory and Practice: TELOS aims to facilitate the development of AI theories by providing program abstractions that are as distinct and well-defined as the theoretical constructs they represent.
- Productivity Gains: By managing complexity and offering powerful structuring tools, TELOS is expected to significantly improve AI programmer productivity, which in turn boosts AI research productivity.
- Evolutionary Programming: The language's focus on comprehensibility and abstraction makes it well-suited for the iterative nature of AI research, where hypotheses are frequently tested and programs evolve.
Key Results
The paper focuses on the design and rationale of the TELOS language rather than presenting empirical results from its application. Therefore, specific quantitative results are not available. However, the authors outline several anticipated benefits:
- Managing Complexity: TELOS provides aids for managing and containing program complexity, crucial for large AI projects.
- Improved Productivity: The language features are designed to enhance programmer productivity, leading to more efficient AI research.
- Easier Evolutionary Programming: The improved comprehensibility of TELOS programs is expected to make the iterative development and refinement of AI models more efficient.
Limitations and Future Work
- Theoretical Bias: While TELOS aims to minimize theoretical bias, the authors acknowledge that the language's design choices inherently reflect certain assumptions.
- Future Work: The paper references two companion documents ([1] and [2]) that likely contain more detailed design specifications and rationale, suggesting that the work presented is part of a larger ongoing effort. The authors imply that TELOS's utility will be demonstrated through its application in various AI research areas, particularly in knowledge representation and the investigation of novel control regimes.
Summary updated