Intelligent Information Infrastructure Project

This project is developing intelligent systems for storing, retrieving, manipulating, and distributing information or knowledge, primarily over major internet protocols like email (STMP) and World-Wide Web (HTTP). There are three research tracks in the project:

o Server Track
The goal is to structure communication with intelligent agents so that the services they provide are readily accessible to any user over networks. Structured communication starts from automatic-form processing and builds towards stereotypical models of user interactions, or conversation analysis. Initial services include, but are not limited to, information search, retrieval, and routing.

o Organization Track
The goal is to develop systems that can allow large communities of users to collectively solve problems through hierarchical task decomposition and reintegration. The project seeks to enable organizations to cope more effectively with complexity by developing systems that allow users to build argument hypertexts. A key issue is managing fan-in bottlenecks, where communications from many people converge on a small number who make sense of it. Automatic survey research is one technology we have already developed and tested.

oNatural-Language Track
The goal is to parse unrestricted text into a semantic representation from which a generator can generate sentences, thus making the earlier tracks significantly more intelligent. The emphasis is on new declarative, parser/generator pairs that can build on an extensive base of implemented research systems. This track seeks to advance new paradigms for semantic perception, knowledge representation, and common-sense reasoning.

This project operates a primary hub for distributing electronic publications by the White House and is developing technology with relevance for citizen communication with government. There are many practical research possibilities in the server track, including issues of distributed hypermedia, automatic polling, group decision-making, and workflow management. Enabling technologies include statistical document categorization/clustering, machine learning (inductive rule learning and neural networks), modular data sharing, document authentication, security/encryption, and tools for developing complex software systems.

The natural language track offers opportunities to help build new, scalable paradigms for machine intelligence. These include reversible parser-generators for unrestricted text, common-sense learning (graphed-based analogy, induction), intersentential coreference, precise control of recursive graph walkers, composing multi-sentence utterances, and a host of deep but operational issues in philosophy of mind. In general, this project offers unique opportunities to pursue research that can have major impacts on the ``information highway,'' electronic access to government, or pushing artificial intelligence beyond microworlds to real-world scale.

o Seminars Overviewing the Project

John C. Mallery