IBM Intelligent Agents activities, overall

NEWS: STATUS of Intelligent Agents Project at IBM T.J. Watson Research:

1. Business Rules for Electronic Commerce: fundamental core technology and pilot applications, about intelligent rule-based agents in e-commerce. This is the closest follow-on to the former Intelligent Agents project. In particular, it is developing the CommonRules alpha prototype, a Java library of business rules capabilities, that supersumes almost all of the functionality of IBM Agent Building Environment (ABE). (ABE is in C++; its pilot prototype was RAISE.) CommonRules builds upon the techniques developed in ABE and RAISE for conflict handling and for procedural attachments (i.e., "situating").

   The first alpha version of CommonRules will be released (free, with trial license) on IBM's AlphaWorks in late June or July of 1999.

2. Information Economies: investigating market economies composed of multiple intelligent agents, including micro-economic interactions (and decision-making) and macro-economic emergent phenomena, especially for information goods and services.

3. MailCat: a learning assistant for categorizing e-mail -- essentially, an intelligent agent.

Agent Building Environment:

Overview of FORMER Intelligent Agents Project (1994-1997):

Our mission is to develop intelligent agent technology that is highly reusable and easy to integrate with a broad spectrum of networked applications. Towards this end, we prototype applications in tandem with developing reusable componentry. We also contribute to company-wide efforts in strategy and in common architecture, e.g., for inter-agent knowledge-level communication and inter-operability. For more information about company-wide efforts, you can see an overview of some of IBM Intelligent Agents activities, as well as our papers etc..

Our reusable intelligent agents technology is embodied as an extensible structured class library, called RAISE. RAISE is mnemonic for: Reusable Agent Intelligence Software Environment. The first phase of RAISE's capabilities include: rule-based inferencing, user authoring of rule bases, integration with external software components, and basic support of inter-agent knowledge-level communication. RAISE is deeply object-oriented in design and is implemented in C++. RAISE also features dynamic pluggability of user-authored rule sets (including easy merging and updating), and development-time pluggability of reasoning engines.

RAISE is at the core of the Agent Building Environment (ABE) developer's toolkit product alpha [which was] in release by IBM Development [during 1996-1998]. ABE [was] available free for Internet download.

There are several aspects to our work on RAISE/ABE, which can be described via an anthropomorphic analogy.

• "Brain": reasoning and learning. To begin with, this includes inferencing with yes-no rules, in a primarily forward direction, e.g., driven by a set of facts that represent the event of an information item's arrival. (By "yes-no", we mean beliefs that are true or are false, as opposed to having a degree of truth in the manner of probabilities or fuzzy logic.)
• "Body": situating the "brain" in a larger environment.

  Historically, reasoning systems have been difficult to marry with other software technologies. RAISE/ABE has an especially innovative technique for situating reasoning engines ("brains") by connecting them to sensors ("eyes"/"ears") and effectors ("hands"). Sensors and effectors are procedural attachments, packaged and dynamically registered as pluggable adapter components. Linkages to sensors and effectors are treated as a syntactic and semantic extension of the pure-belief knowledge representation.

• "Society": inter-agent communication, especially at the knowledge level. To begin with, this includes support for exchanging rules and facts in a standard fashion, using the approach of the ARPA Knowledge Sharing Effort, e.g., Knowledge Interchange Format. Sensing, effecting, and delivering events also provide modes of inter-agent knowledge-level communcation.
• Human-agent interaction: both the deep and the surface aspects of user interface (UI). To begin with, this includes the user authoring of rule sets. Employing rules (with clean semantics) as a method for a user to instruct an agent helps ease the user's job in modifying the instructions, and in understanding what the agent does. (Rules are also simpler to use than scripting-type programming languages, and are more powerful than menus.) This is an example of deep UI. Employing graphical template forms helps structure the authoring of rules that are particular to a specific application; this is an example of surface UI.

RAISE is especially appropriate for enhancing Internet applications with embedded intelligent agents that perform information flow functions: finding, searching, filtering, categorizing, storing, routing, and/or selectively disseminating information items. Pilot prototype applications for RAISE, currently running, include: electronic commerce shopping and customer service support workflow, on the Web and in Lotus Notes; and news and e-mail on Internet. For more details, see our papers on RAISE and its applications.

A number of enhancements to RAISE itself are currently under development. We have designed (and currently have running as a prototype) an especially innovative technique for handling conflicts between rules. Rules may override each other, based on specified override orderings. The approach, called courteous logic programs, is computationally low-overhead, guarantees a consistent set of conclusions, and is semantically clean. It facilitates merging and updating rule sets, as well as agents learning from data mining (statistical induction) or from inter-agent knowledge-level communication (taking "advice" from "friends"). (For cognoscenti: it constitutes a practical form of "prioritized default" reasoning, a kind of "non-monotonic" reasoning.) For more details, see our papers.

Our further technical directions include: extension of reasoning capabilities to include machine learning and probabilistic / fuzzy reasoning; closer integration with text analysis and search; closer integration with data mining; and inter-agent knowledge-level communication, especially to exchange rules and facts, e.g., about electronic commerce. We have designed an architecture for itinerant (mobile-execution) intelligent agents that interact at Agent Meeting Points.

For more about our work, you can see our:

Papers etc. that are most closely related to the project.

Massively Distributed Systems page, which has more about our intelligent agents, mobile agents, information economies, and other related work.

For further information, please contact:

Benjamin Grosof at grosof@us.ibm.com
Jeff Kephart at kephart@watson.ibm.com

[ IBM home page | Order | Search | Contact IBM | Help | (C) | (TM) ]