I-X: Technology for Intelligent Agents & Tools Systems Architecture

Work in Intelligent Planning and Activity Management at the University of Edinburgh (See http://www.aiai.ed.ac.uk/project/plan/) has led to a number of significant assets that are re-used on a number of projects. These assets include: Nonlin, O-Plan, <I-N-OVA>, Enterprise Ontology, Enterprise Architecture, Task Manager, O-P³ Process Panels, Common Process Editor, etc. A new generation of the work will draw on these earlier efforts, generalise them, and significantly extend the application of the core concepts and assets, leading to new re-usable components, and create opportunities for applications and further research.

This new programme is called I-X and the core components are a shared model representation called <I-N-CA> and a related systems integration architecture called I-Core. A variety of re-usable components and systems will be build on the new architecture and these will be collectively referred to as I-Technology and I-Tools.

The "I" in I-X and I-Technology reflects the following:

• Intelligent - I-X supports the construction of intelligent systems and intelligent agents
• Intelligible - I-X supports the construction of systems which are intelligible to their users and to other systems and agents.
• Integration - I-X is a systems integration architecture
• Issue-based - I-X is an issue-based and issue handling architecture

• Human-friendly representations of plans for human communication and cooperation, as well as man-machine interaction in areas such as mixed initiative planning.
• Plan and activity representation ontologies and their potential to assist in the knowledge acquisition and knowledge management process.
• Plan representations well suited to storage and for use in markup languages such as XML.
• Plan representation languages which are also suitable for use to represent the planning (task setting, planning, scheduling and execution) process.
• Bringing AI plan and agent capability description experience into practical use.
• Relationship to emerging standards (NIST PSL, ISO, WfMC, etc).

I-X provides a new systems integration architecture. It can be used to create agents or non-agent systems. Its design is based on the O-Plan planning agent architecture. I-Core incorporates components and interface specifications which account for simplifications, abstractions and clarifications in the O-Plan work. I-Core provides an issue-handling workflow style of architecture, with reasoning and functional capabilities provided as plug-ins. Also via plug-ins it allows for sophisticated management and use of the internal model representations to reflect the application domain of the system being built in I-Core. I-Core systems or agents may be recursively or fractally composed, and may interwork with other processing cells or architectures. This is a systems integration approach now being advocated by a number of groups concerned with large scale, long-lived, evolving and diverse systems integration issues.

The I-Technology programme has 5 aspects:

1. Systems Integration - A broad vision of an open architecture for the creation of intelligent systems for the synthesis of a result or "product" which is based on a "two cycle" approach which uses plug-in components to "handle issues" and to "manage and respect the domain model".
2. Representation - a core notion of the representation of a synthesised product as a set of nodes making up the components of the product, along with constraints on the relationship between those nodes and a set of outstanding issues - <I-N-CA> - Issues, Nodes, Critical Constraints and Auxiliary Constraints.
3. Reasoning - the provision of reusable reasoning and model management capabilities.
4. Viewers & User Interfaces - to understand user roles in performing activities and to provide generic modules which present the state of the process they are engaged in, their relationships to others and the status of the artifacts/products they are working with.
5. Applications - work in various application sectors which will seek to create generic approaches (I-Tools) for the various types of Task in which users may engage.

• Task and Option Management - the Controller -- The capability to support user tasks via appropriate use of the processing and information assets and to assist the user in managing options being used within the model.
• Issue Handlers -- Functional components (distinguished into those which can add nodes to the model (synthesis) and those which analyse the model (to add information only).
• Model Management -- coordination of the capabilities/assets to represent, store, retrieve, merge, translate, compare, correct, analyse, synthesise and modify models.
• Constraint Managers -- Components which assist in the maintenance of the consistency of the model.
• Viewers -- User interface, visualisation and presentation viewers for the model - sometimes differentiated into technical model views (charts, structure diagrams, etc.) and world model views (simulations, animations, etc.)

A number of different types of internal interfaces are available within the framework to reflect the protocols or interfaces into which the various components can fit. The separation into viewers, Issue handlers or processing assets, constraint managers and information assets has been found to be useful in a number of AIAI projects. This also puts the I-X work on a convergent path with other Model/Viewer/Controller styles of systems framework.

<I-N-CA> Ontology

The <I-N-CA> constraints model is a means to represent synthesised or designed artifacts as a set of constraints on the space of possible designs. By having a clear description of the different components within a synthesised artifact, the model allows for them to be manipulated and used separately from the environments in which they are generated.

The constraints which add a node (these are in the form ``include node'') in the <I-N-CA> model set the space within which the description of the artifact may be further constrained.

The I (issues) state the outstanding items to be handled and can represent unsatisfied objectives, problems which analysis has shown need to be addressed, etc. The I constraints can be thought of as implying further constraints which may have to be added into the design in future in order to address the outstanding issues.

The CA constraints restrict the relationships between the nodes to describe only those artifacts within the design space which meet the requirements. The constraints are split into "critical constraints" and "auxiliary constraints" depending on whether some constraint managers (solvers) can return them as "maybe" answers to indicate that the constraint being added to the model is okay so long as other critical constraints are imposed. The maybe answer is returned as a disjunction of conjunctions of critical constraints.

The use of "maybe", delayed and/or conditional constraint results from problem solvers is common in AI planning where one of the main world state condition/effect solvers many of us use is the modal truth criterion (MTC) based on Tate's work on the Nonlin "QA Algorithm" in the mid 1970s. This established whether some condition was satisfied at some point in a partial ordering of activities in a plan. In later Edinburgh planning work, the same interface is used a general model for constraint solvers being plugged into a planner - or indeed any synthesis system.

For example, if the product or artifact is an activity plan, some types of ordering (temporal) and variable (object) constraints are distinguished from all other auxiliary constraints since these act as critical constraints or cross constraints, usually being involved in describing the others -- such as in a resource constraint which will often refer to objects/variables and to time points or intervals.

We are also promoting a knowledge modelling approach in AKT which is thought of as a synthesis of the model from multiple human and system sources using any modelling methods and approaches or tools. This would build "nodes" in the model as appropriate to the domain, and build constraints on those nodes. It would maintain a set of outstanding modelling issues. It would use terminology in a domain ontology that was being concurrently defined alongside the modelling, so it itself would be a result of the modelling. It would allow a wide range of reasoning approaches and sub-solvers to be employed as appropriate to the tasks being conducted being found through the use of a capability description that allowed access to the library of capabilities (i.e. agent capabilities or problem-solving method libraries).

Issues can be categorised as to whether they will, may or will never add a "node". The I constraints which can lead to the inclusion of new nodes are of a different nature in the synthesis process to those which cannot.

The choice of which constraints are considered critical is itself a decision for an application of I-X and I-Core. It is not pre-determined for all applications. A temporal activity-based planner would normally have objects/variable constraints (equality and inequality of objects) and some temporal constraints (maybe just the simple before(time-point1, time-point-2) constraint) as the critical constraints. But, in a 3D design or a configuration application object/variable and some other critical constraints (possibly spatial constraints) might be chosen. It depends on the nature of what is communicated between constraint managers in the application of the architecture.

The types of constraints in an <I-N-CA> model is shown below:

I - Issues

- which may add an "include node" constraint (i.e. add nodes to the model)
- which will not add an "include node" constraint

N - Node Constraints

- include node constraints
- other node constraints

C - Critical Constraints

E.g., if the artifact is an activity plan:
O - Critical Ordering Constraints
V - Critical Object or Variable Constraints

A - Auxiliary Constraints

E.g., if the artifact is an activity plan:
O' - Non-critical Ordering Constraints
V' - Non-critical Object or Variable Constraints
X - eXtra constraints (such as):
- Authority Constraints
- World State Constraints
- Resource Constraints
- Spatial Constraints
- Miscellaneous Constraints

A number of concepts are being used as the basis for exploring task orientated multi-agent and mixed-initiative work involving users and systems. Together these provide for a shared model of what each agent can and is authorised to do and what those agents can act upon. The concepts are:

1. Shared Object/Product Model -- a structured representation of the object being modelled or produced using a common constraint model of the object or product (<I-N-CA>).
2. Shared Task Model -- Mixed initiative model of "mutually constraining the space of objects/products"
3. Shared Space of Options -- option management.
4. Shared Model of Agent Processing -- handlers for issues (functional capabilities) and constraint managers
5. Shared Understanding of Authority -- management of the authority to do work (to handle issues) and which may take into account options and levels of abstraction of the model of the object or product.

In particular, this work will carry forward the development of a strong systematic ontology to underpin the models of processes and activity - including continuing to engage in and promote its use as a basis for standards. The work will draw on the initial efforts to create an ontology suitable for the conceptual description of all aspects of an organisation - the Enterprise Ontology and on the development of a constraint model of activity <I-N-OVA>. The work will promote the aim that PIF, NIST PSL, OMWG CPR and SPAR all converge on a single core model of activity based on <I-N-CA>.

Input/Output Scenarios

A range of Input and Output scenarios are being studied to drive the development of I-X.

• Inter-agent messages
• Robot effectors and sensors
• Thermometer with preset min and max alarms
• Continuous readout thermometer
• Database access
• Monitor output for continuous status updates and warning lights for specific events
• User viewing agent process and model(s)
• User controlling agent process and model development

1. Characterisation of the separate components of a single I-X agent to establish what features each has and how they should differ. Establish whether the a priori view on the separate componets is desirable.
2. Initial Java encoding experiments to establish that the envisaged plug-in framework is feasible and to experiment with different approaches.
3. Consideration of how to describe plug-in components. This is envisaged as being in the <I-N-CA> ontology, reified in Sorted First Order Logic and storable in XML/RDF.
4. The provision of an ontology for issues and issue handler descriptions for some sample applications of I-X: (e.g. I-Plan, I-Config).
5. Consideration of how to describe the design, architecture and implementation. Possible use of UML models.
6. Initial applications of I-X (e.g. I-Plan, I-Config).
   • A very simple initial application of I-X fully worked through (PicoIX and I-Sim).
   • A simple, but realistic application of a mainly manual use of I-X (user level process management and collaboration support).
   • A DERA-related sample application, possibly based on Coalition Operations in the CoAX Project. and which could incorporate an initial generative I-Plan component.

• Java encoding limitations and opportunities
• Delivery of any user interfaces on multiple platforms via suitable web browsers.
• Use of XML/RDF to store intermediate results passed between suitable agents and components and to assist in storage of those components in a form which can be retrieved and used.
• Design to allow for the resulting components to be repackaged and re-used as embeddable components in a wide range of other systems.

• Presentation (about product/artifact and process)
  • Visible indications of users' workflow state
  • Multiple views tailored to user role, information type, etc.
  • Selection of what information to display.
  • Views (e.g. tables) that facilitate comparison.
  • Summary form with drill-down /expansion to details.
  • TechWorld Viewers
  • Colour-coding of status
  • Colour choice principles
  • Layout principles
• Multi-user cooperation
  • User roles (as distinct from physical users)
  • Mixed-initiative
  • Authority
  • Workflow
  • Issues (explicit representation and tracking of)
• Delivery (of applications, e.g. on Web)
  • Portable
  • Standards (XML/RDF markup)
  • Familiar interfaces (e.g. Browsers)
  • Low bandwidth use (e.g. cellular phones, WAP, PDA)
  • Reusable frameworks and modules
  • Embeddability
• Workflow
  • Workflow of using the system itself (hence meta-workflow in workflow tools)
  • Process Issues
• Openness
  • An architecture / framework
  • Plug-in concept and implementation support
  • Plug-in Issue handlers, Artifact analysers, Constraint managers, TechWorld viewers

• I-Plan Planning Agent
• DERA MBP Operator Support Agent
• Emergency and Unusual Procedures Assistant (Context sensitive Cell Phone or web access to procedural assistance)
• Multi-national Coalition Operations
• Multi-perspective Modelling framework for Advanced Knowledge Technologies - AKT
• Expository Scenarios
  • Meeting Room Booking Agent (Non-activity based, nodes are meetings)
  • Interior Decoration Design Agent (No time points)
  • PicoIX and I-Sim
  • For simple coding experiments and explanations
  • Example that can use probabilities within Issue Handlers and/or Constraint Managers