From: AAAI-91 Proceedings. Copyright ©1991, AAAI (www.aaai.org). All rights reserved.
`
`Toward an ntelligent Agent Framework for Enterprise Integration
`Jeff Y-6 Pan and Jay M. Tenenbaum
`Enterprise
`Integration Technologies Corporation
`459 Hamilton Ave., Palo Alto, CA 94301
`and
`Center for Integrated Systems, Stanford University
`pan@cis.stanford.edu;
`marty@cis.stanford.edu
`
`Abstract
`
`’
`
`We propose a software framework for integrating people
`and computer systems in large, geographically dispersed
`manufacturing enterprises. Underlying the framework is
`an enterprise model that is built by dividing complex busi-
`ness processes into elementary
`tasks or activities. Each
`such task is then modeled in cognitive terms (e.g., what
`to look for, what to do, who to tell), and entrusted
`to
`an Intelligent Agent (IA) for execution. The IAs interact
`with each other directly via a message bus, or through a
`shared, distributed knowledge base. They can also interact
`with humans through personal assistants (PAS), a special
`type of IA that knows how to communicate with people
`through multi-media interfaces. Preliminary experimental
`results suggest that
`this model-based, man-machine ap-
`proach provides a viable path for applying DA1 to real-
`world enterprises.
`
`1
`
`Introduction
`
`We are creating a software framework for integrating
`people and computer systems in large, geographically
`dispersed manufacturing
`enterprises.
`It is based on
`a vision of augmenting human workers with a large
`number of- computerized assistants, known as intelli-
`gent agents, or IAs. Each IA supports a clearly dis-
`cernible task or job function, automating what it can
`and calling on the services of other IAs, as well as
`human beings, when necessary.
`IAs can interact di-
`rectly via a dedicated message bus, the IA Network,
`or through a shared knowledge-base,
`the MKS knowl-
`edge service[l,2],
`as illustrated
`in Figure 1.
`
`At the core of MKS is a comprehensive object-oriented
`model of the enterprise and how it functions.
`The
`MKS model includes descriptions of personnel,
`facil-
`ities, equipment,
`inventory, manufacturing processes,
`
`‘This work was supported by the Defense Advanced Re-
`search Projects Agency under contracts N00014-87-K-0729 and
`N00014-90-J-4016.
`
`206
`
`MULTIAGENT
`
`ARCHITECTURES
`
`the flow
`It also captures
`and other corporate assets.
`of information, decisions and materials
`through
`the
`- how things get done.
`The model
`is
`enterprise
`wired into the enterprise’s information
`infrastructure
`(databases, CIM systems, accounting systems etc.) so
`that
`it continuously
`rcflccts
`the actual state of the
`enterprise. Agents interact with these information re-
`sources through the model via high-level service pro-
`tocols that
`insulate
`them from details such as where
`information
`resides. They can also register with the
`model their
`interest
`in particular events, and be no-
`tified when they occur. MKS
`thus serves agents as
`a repository
`for shared knowledge, and a center for
`information exchange.
`reasoning, action and com-
`IAs model the perceptual,
`in performing human
`job
`munication
`skills involved
`functions,
`such as those of an equipment operator,
`production scheduler, engineer, purchasing agent or
`manager.
`IA’s must therefore know what to look for,
`what to do when they see it, and who to tell, just like
`the person they model.
`Interactions
`among IAs fol-
`low the established corporate
`lines of communication
`and procedures. Collectively,
`IAs and their communi-
`cation
`links form an active, operational model of an
`enterprise.
`
`in this society of agents, humans re-
`To participate
`quire the services of personal assistants
`(or PAS). PAS
`belong to a special class of IAs that know how to
`
`Figure 1: An IA-based Enterprise
`
`Integration Frame work
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 1
`
`

`

`communicate with humans through E-mail, graphical
`editors, and other standard modalities
`(e.g., beepers,
`faxes, telephone). They also know how to translate be-
`tween these human modalities and IA network proto-
`cols. To facilitate
`the interaction, one’s PA maintains
`a personal model that
`includes where they currently
`are, how they can be reached, what information
`they
`would like monitored
`through the knowledge service,
`and what actions
`to take, either in response to mes-
`sages from IAs or notifications
`from the knowledge ser-
`vice. For example, reports of minor trouble might be
`forwarded automatically
`to a human subordinate or an
`IA, whereas a report of major trouble might warrant
`being paged. Moreover, one’s PA model can include
`personal preferences for supporting
`information
`to be
`presented in conjunction with decision requests. Such
`an architecture
`supports the notion of man/machine
`cooperative work, as articulated by Winograd
`[3]. In-
`deed, it goes further
`in allowing tasks gradually
`to
`be handed off to IAs so that ultimately
`they can be
`performed interchangeably by a person or their agent.
`
`In addition to supporting traditional human activities,
`IAs can automate a variety of tasks for which peo-
`ple may be too expensive or otherwise
`limited (e.g.,
`too slow, too inattentive),
`Think of IAs as an in-
`exhaustible source of “cheap labor” that can provide
`dedicated human-like services such as watching over
`each piece of equipment
`in a factory and shepherd-
`ing each workpiece and piece of paperwork along their
`prescribed routes.
`IAs can also integrate software
`in
`interesting ways, such as by automating human job
`functions that primarily
`involve transferring
`informa-
`tion between several disjoint systems, or by serving
`as intelligent
`front ends that make existing software
`more useful and easier to use.
`
`set-
`in an operational
`For the framework to succeed
`ting, we believe
`it
`is essential
`that
`the enterprise
`models underlying
`IAs, PAS and MKS be created
`and maintained by the people they serve - the end-
`users that understand best the tasks to be modeled.
`Two features of our approach help make this possible.
`First, partitioning complex activities
`into simple tasks
`and modeling them in cognitive terms produces mod-
`els that are familiar to workers, and therefore easy for
`them to understand. Second, we are developing sim-
`ple modeling tools for end-users, that enable them to
`copy and customize generic activity models (e.g., for
`brokering)
`from a library.
`monitoring,
`transactions,
`Customization might require editing a script or flesh-
`ing out a decision tree, but such skills can be quickly
`mastered when the editing tools support represent*
`tions that are already familiar[4,5].
`
`In the remainder of this paper, we flesh out this vision
`and describe a prototype
`implementation,
`currently
`under development,
`that will run the semiconductor
`fabrication
`facility at Stanford’s Center for Integrated
`Systems (CIS). Sections 2 and 3 provide technical de-
`tails on the framework and its implementation.
`Sec-
`tion 4 reports on preliminary experiments with the
`prototype at CIS. Finally, Section 5 summarizes our
`results to date and draws implications
`for both dis-
`tributed AI and enterprise
`integration.
`For a fuller
`treatment of these issues, see [6].
`
`amework
`
`2.1
`
`Technical Overview
`
`in Figure 1 consists of
`shown
`IA framework
`The
`three synergistic
`technologies:
`intelligent agents, the
`MKS knowledge service, and a distributed
`system
`infrastructure
`(not shown).
`IAs
`interact with hu-
`mans through their Personal Assistants
`(PAS) over an
`enterprise-wide network using standard service proto-
`cols. IAs can also access information resources, control
`on-line equipment,
`and trigger other IAs indirectly
`through the MKS knowledge scrvice[2].
`
`is an active, autonomous
`agent
`Each computerized
`process that models a single discernible task. The task
`may be either one traditionally performed by a human
`or one intended specifically
`for a computerized agent
`(e.g., tracking a wafer through a process). We model
`activities
`in cognitive terms that make the models easy
`for our users to understand and maintain. Activities
`are described by corresponding “activity” objects
`in
`the MKS model, in terms of the information
`they con-
`sume, process and produce, and the other MKS model
`objects
`(people, equipment, wafer lots and so forth)
`that participate. From such descriptions,
`IAs can de-
`termine what information and events to monitor and
`how to respond.
`
`in two
`is built
`for an enterprise
`An IA framework
`stages. First, enterprise activities are modeled in cog-
`nitive terms and added to the MKS model as A&iv-
`ity Objects.
`Second,
`these activity models are selec-
`tively activated as IAs, to actually perform the mod-
`eled tasks. We shall now cover each of these steps in
`some detail.
`
`PAN & TENENBAUM
`
`207
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 2
`
`

`

`2.2 Activity Modeling
`
`,
`
`Following the object-oriented methodology of MKS,
`individual
`activities
`are modeled by customizing
`generic activity models from the MKS
`library. The
`library of activity objects
`is organized as a classifi-
`cation hierarchy according
`to the nature of the work
`performed.
`
`Each activity object contains the knowledge necessary
`for performing a specific job (i.e., the modeled activ-
`ity). The knowledge is provided in the form of a tem-
`plate that defines the essential elements necessary
`to
`construct a cognitive model for performing
`the task:
`the players, capabilities,
`states, and sensory
`inputs,
`and the core model consisting of message and sensory
`input patterns,
`sets of actions, and patternaaction
`rules (i.e., reasoning)
`that associate
`them. Players
`describe
`the parties
`involved in an activity, and are
`specified generically where possible (eg., a job role vs.
`a specific person); Capabilities
`specify the tasks an
`agent is capable of performing, with applicable “cost”
`to use when bidding for jobs; States are the natural
`contexts that people use in deciding what sensory pat-
`terns and actions are appropriate.
`(Perceptions
`and
`actions can also be conditioned on goals, resources and
`other cognitive concepts, where appropriate.);
`Sen-
`inputs designate
`the sources of data to moni-
`sory
`tor and the specific patterns
`to look for (as functions
`of state); Core Model defines the IA’s behavior
`in
`terms of sets of pattern-action
`rules, indexed by cur-
`rent state.
`(Sensory patterns define what to look for,
`Message patterns define what IA messages are mean-
`ingful in the context of a particular activity, and Ac-
`tion Sets detail the sequences of actions to accomplish
`some desirable results.)
`
`rules are the main form of
`While PatternJAction
`reasoning model used in our experiments,
`other in-
`tuitive behavioral
`representations
`can be used where
`they contribute
`to ease of expression and compre-
`hension. Examples
`include state transition diagrams,
`petri nets, flow charts, decision trees, and scripts. The
`choice depends on what is most natural
`for the task
`at hand.
`
`design practice, models of
`Following object-oriented
`specific activities can be generalized to create reusable
`that model generic job roles. Ac-
`libraries of objects
`cordingly, we have begun building a comprehensive
`library of generic activity models for semiconductor
`manufacturing,
`together with supporting
`libraries of
`basic methods for sensing, reasoning, communicating,
`and taking action. We are also developing activity
`
`208
`
`MULTIAGENT
`
`ARCHITECTURES
`
`modeling tools analogous to the MKS modeling tools
`for processes, equipment and the like. They will make
`it possible to select an activity object
`from a library
`that has been predefined for common high-level tasks
`(e.g., equipment control) and customize it by selecting
`and composing desired sensing, data processing, deci-
`sion making, and action behaviors
`from the method
`Libraries and CASE
`tools thus effectively
`libraries.
`raise the level of vocabulary used for modeling.
`
`2.3
`
`Intelligent Agents
`
`In the following subsections, we first discuss the agen-
`tification process by which a passive activity object
`is transformed
`into an active IA, and then elaborate
`on the process by which IAs communicate with each
`other.
`
`2.3.1
`
`Agentification
`
`of IAs
`
`refers to the three steps involved in cre-
`Agentification
`ating a computational process(es)
`(i.e. an IA) whose
`behavior mimics that described in a corresponding ac-
`tivity object.
`First, a unique instance of an activity
`object
`is created from the MKS library - a surrogate
`for a particular
`agent performing a particular
`task.
`Second, one or more computational processes are cre-
`ated to implement
`the “autonomous” agent’s sensors
`and behavior.
`Finally,
`these processes are activated
`so that
`they can begin receiving and responding
`to
`sensory data and incoming IA messages. We will now
`elaborate on each of these steps.
`
`The Instantiation
`step involves customizing an ob-
`ject template copied from the MKS activities
`library
`for the specific task at hand. Normally, this involves
`filling in situation-specific
`information
`for Players and
`other slots in the template.
`(e.g., The operator of the
`Tylan Furnace
`is Mary).
`
`step initiates computational
`The Process Creation
`processes that efficiently implement the prescribed be-
`havior in a given runtime environment. Think of these
`processes as interpreters
`that translate
`the concise be-
`havioral descriptions provided by activity objects
`into
`the best possible runtime implementations.
`For exam-
`ple, suppose that an activity model calls for periodi-
`cally monitoring a data source. Under MKS, such a
`requirement can be translated
`into a simple registra-
`tion of interest with the notification mechanism
`[l].
`The process can then remain suspended until a change
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 3
`
`

`

`to the subject data is reported to the MKS model. If
`such a mechanism
`is not available, a process would be
`created to periodically
`sample the data source in re-
`sponse to clock interrupts. The intent,
`in either case,
`is to insulate
`the model builder from implementation
`details. To maximize flexibility, separate processes are
`established
`for an IA’s reasoning engine and each of
`its primary sensory and communication
`inputs. Each
`such receptor process is then “trigger-wired” to its cor-
`responding data source in the most computationally
`parsimonious
`fashion. The IA network is now ready
`for activation.
`
`The final Activation
`step arms the sensor processes
`so that they “watch” their assigned data sources. Si-
`multaneously, all the IA-message
`listeners and trans-
`mitters are switched on so that IAs can exchange re-
`quests and inquiries through the IA network. The pro-
`cess implementing
`the reasoning engine is placed in a
`continual stimulus-response mode. When a “‘stimu-
`lus” arrives, either as a symptom detected by one of
`the IA’s sensory sub-processes, or through a message
`received from another IA, the reasoning process is in-
`voked to generate
`the proper responses and actions,
`following
`the decision procedure
`(e.g., production-
`rule, state transition diagram, decision tree) contained
`in the activity model.
`
`2.3.2 The IA Network
`
`The IA Network is a Jog&Z communication bus, de-
`signed exclusively for exchanging messages among IAs
`in a special
`format, known as the IA protocol. An
`IA message, based on the
`IA protocol,
`allows an
`IA to report
`to or request services
`from other IAs.
`Even though IA messages may be broadcast over the
`same physical network as other
`logical communica-
`tion protocols (e.g., the MKS protocol, over an Ether-
`net), their high-level semantics provide concise, nat-
`ural, and comprehensive communication among IAs.
`Describing communications
`among IAs in this high-
`level vocabulary also insulates an enterprise’s activity
`model from the implementation details of its network
`infrastructure.
`
`An IA-message consists of four parts: 1). the message
`type, specifying the kind of communication pattern to
`be engaged in; 2). a “target” description, addressing
`the intended IA(s) either directly by name(s), or indi-
`rectly by role, interest, or capability/qualification;
`3).
`the “body” of the message which is to be sent to the
`targeted IA(s) and interpreted within its context; and
`an optional
`list of keyword arguments detailing
`4.
`interactions
`(e.g., what to do should a message fail to
`find its targeted IA within a specific time limit).
`
`There are four types of IA messages that are currently
`adopted for our IA system: REQUEST,
`INFORM,
`IN-
`QUIRY, and BID. A REQUEST message is used to
`issue commands to an IA, resulting in desired actions.
`An INFORM message
`is a special type of request,
`intended primarily
`for passing textual
`information
`to
`an IA(s), for forwarding to the most appropriate per-
`son. The receiving IA gets to decide who is the most
`appropriate person in its own context (e.g., Mary, the
`equipment operator on duty). It can then forward the
`message to that person’s PA, which may decide, for
`example,
`to send him/her an E-mail with the mes-
`An INQUIRY
`message is de-
`sage as its contents.
`signed to acquire information
`through another IA. On
`sending this type of message, an IA will be suspended
`until the expected
`information becomes available. A
`special
`feature
`for this type of message
`is that de-
`fault behaviors for an aborted situation (e.g.,
`:if timed
`out or :if-rejected) will terminate
`the suspension - a
`precaution
`to prevent permanent “hanging” of the IA
`due to an inquiry
`that cannot be satisfied.
`Finally,
`a BID message encompasses
`three stages of behav-
`ior: 1). broadcasting messages
`to a set of targeted
`IAs inviting
`them to submit bids to supply service,
`with associated costs; 2). evaluating all bids received
`within the :max-response-time
`and selecting the “win-
`ning” bid according to the :cost-function;
`3). sending
`out a request message with the winning bidder as the
`targeted IA.
`
`The IA message format allows targeted IA(s) to be de-
`scribed by their names, by their roles (including
`their
`interests), or by their capabilities. Cull by role requires
`that the targeted IA be determined dynamically
`in the
`context of the sending IA. CudZ by capability, on the
`other hand, requires that a generalized pattern spec-
`ifying a qualification be broadcast and interpreted by
`all IAs within the specified broadcast scope.
`
`2.3.3
`
`Personal Assistants
`
`As discussed earlier, Personal Assistants are a special
`class of IA distinguished by their ability to communi-
`cate with people as well as with other IAs. They en-
`individuals, enabling
`them to interact with
`cupsdute
`other IAs using the network protocols and to act as
`their own (i.e., living) activity model.
`
`Each PA maintains a personal activity model of the
`individual
`it serves. One’s PA model would include
`the following
`information about
`them: where they
`are; how to contact
`them
`(e.g., E-mud,
`fax, phone,
`pager) their capabilities
`(a list of tasks
`X-terminal,
`
`PAN & TENENBAUM
`
`209
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 4
`
`

`

`their responsibilities
`to perform);
`they are qualified
`(a list of tasks currently assigned to them); IAs sup-
`porting
`their current
`tasks;
`their general and task-
`specific
`information needs and interests;
`their pref-
`erences for how information should be presented; an
`activity model for personal tasks.
`
`Using this information, one’s PA can perform a variety
`of services on their behalf. For example, it can monitor
`incoming
`information
`for them 24 hours a day (e.g.,
`read their E-mail buffer, messages from their IAs, or
`notifications
`from the MKS knowledge service). Rou-
`tine events can be handled autonomously, such as by
`dispatching
`them to a human subordinate or an IA,
`while important
`issues and information are brought
`immediately
`to their attention.
`
`3
`
`Implementation
`
`Our experiments with IAs are being conducted at
`Stanford’s CIS, whose fabrication
`line provides an ac-
`cessible real-world manufacturing
`environment. The
`IA architecture
`is being implemented as an extension
`to the MKS framework [l] , by adding a sub-hierarchy
`to the MKS model taxonomy
`that includes the activ-
`ity library and instantiated
`activity objects.
`IAs also
`use the MKS knowledge service [2] to access real-time,
`manufacturing data from the CIS fabrication
`line.
`
`in
`is being prototyped
`the IA architecture
`Like MKS,
`HyperClass, an object-oriented programming environ-
`ment [7] implemented on Lucid CommonLisp version
`3.0. A distinctive
`feature of HyperClass
`is MetaClass,
`a toolkit for rapidly constructing
`customized
`interac-
`tive graphical editors. MetaClass
`is being used exten-
`sively to create specialized editors for building IA ac-
`tivity models. Additionally,
`it helps us prototype
`the
`graphical user interfaces by which PAS interact with
`their human masters. The multi-tasking
`capability of
`Lucid CommonLisp 3.0, provides an easy way to im-
`plement the multiple autonomous processes required
`for IAs. While all experiments,
`to date, have been
`done on Sun4 workstations,
`the system can be readily
`ported to other hardware platforms
`(e.g., DEC Sta-
`tions 3100 and 5000) running Lucid 3.0.
`
`to run
`In our initial prototype, all IAs are restricted
`within a single workstation,
`though they can remotely
`access the MKS model and enterprise-wide
`informa-
`tion through the distributed MKS knowledge service
`protocols
`[2]. H owever, future generations of the sys-
`tem must be fully distributed,
`so that IAs can live
`
`210
`
`MULTIAGENT
`
`ARCHITECTURES
`
`in a variety of geographically dispersed workstation
`environments
`(C++, Lisp, Unix, VMS and so forth).
`A distributed
`infrastructure,
`itself based on low-level
`agents known as Proxies,
`is being developed for this
`purpose [8].
`
`4 An IA-run Enterprise
`
`We shall now illustrate, with our ongoing experiments
`at CIS, how Intelligent Agents can be used to model
`and run an enterprise. Additional experiments using
`the framework to coordinate design and manufactur-
`ing decisions for concurrent engineering are reported
`in Brown[S].
`
`The following scenario is a slightly dramatized version
`of currently
`running code, focusing on a few generic
`tasks such as routing wafers through processing steps,
`assigning equipment, and monitoring
`for equipment
`malfunctions. The action begins when the operations
`manager starts a new wafer lot. At the same time
`that a wafer lot is started
`in the fab-line, a “shad-
`owing” wafer-lot genie
`is created
`in the IA world.
`This wafer-lot
`IA is entrusted with moving the lot
`expeditiously along the routes defined in the process
`recipe, making sure it is on schedule and receiving
`its
`fair share of resources. At each process step, it per-
`forms dynamic equipment assignment by selecting the
`most suitable equipment (e.g., capable and least busy)
`among all available machines
`in the fab-line
`[2]. The
`wafer-lot IA subsequently sends an IA-message
`to the
`chosen equipment’s operator
`IA, requesting
`that the
`lot be added to the incoming waiting queue of the
`equipment. The wafer-lot IA then switches itself to a
`“holding-in-queue”
`state where it awaits potential ab-
`normal reports from other IAs (e.g., the equipment’s
`IA reporting
`that the machine
`is being shut down).
`It also wakes up periodically
`to make sure that
`its
`job request
`is not unfairly stalled
`in the equipment’s
`queue.
`
`each piece of equip-
`From a production perspective,
`ment has an operator
`that maintains a waiting
`IA
`queue for incoming lots and is responsible for keeping
`its machine running at peak efficiency. Whenever
`the
`equipment
`is in an “idling” state,
`its operator IA will
`attempt
`to select a wafer lot from the waiting queue
`following a given prioritization
`rule, and instruct
`the
`equipment to load and process the lot.
`
`Each piece of equipment also has a monitor IA that
`emulates a technician watching for anomalous sensor
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 5
`
`

`

`readings (including gauges, instruments, and measure-
`ments).
`For example, an IA continually
`tracks
`the
`health of CIS’ Tylan furnace, watching for an abnor-
`mal temperature gradient
`indicative of a leak.
`If a
`malfunction
`is detected, an IA-message requesting an
`emergency shutdown of the equipment will be sent
`to the Tylan furnace’s operator IA. This IA, in turn,
`may attempt
`to contact the operator-on-duty
`through
`her PA, which may, for example, activate her personal
`beeper.
`If an acknowledgement
`from the targeted IA
`is not received within 120 seconds, a doomsday alarm
`will be forwarded directly
`to the Facility-Manager’s
`PA. The monitor IA will also at tempt to locate a main-
`tenance person by putting out an open bid-call on the
`IA network within CIS, requesting a qualified Tylan
`furnace serviceperson.
`If no one responds on time, a
`message will be sent to the facility manager instead.
`
`5 Discussion
`
`in which human and
`We have presented a framework
`intelligent agents can interact
`to facilitate
`the infor-
`mation flow and decision making in real-world enter-
`prises. Underlying
`the framework
`is the notion of an
`enterprise model that
`is built by dividing complex
`enterprise operations
`into a collection of elementary
`tasks or activities. Each such task is then modeled in
`cognitive terms and entrusted
`to an IA for execution.
`Tasks that require human involvement are referred to
`the appropriate person through their Personal Assis-
`tant. Our experiments,
`though preliminary,
`suggest
`that this divide and conquer strategy
`involving a part-
`nership of man and machine
`is a viable path toward
`real-world distributed AI.
`
`in contrast, has focused
`Traditional DA1 research,
`largely on sophisticated distributed problem solving
`and negotiation
`techniques. However, we have found
`that many everyday enterprise activities
`can be ac-
`complished with only primitive reasoning abilities.
`In-
`deed, most human organizations are designed (or have
`evolved) to minimize precisely
`the type of agent in-
`teractions
`that dominate DA1 research. Modeling be-
`havior after the reasoning and inter-agent communica-
`tion processes found in human-run organizations helps
`make IAs comprehensible
`to people. Through our ex-
`periments, we discovered that this is a key factor
`in
`people accepting enterprise
`integration.
`
`in-
`to agent research evolved
`While our approach
`dependently
`from
`traditional
`distributed AI ap-
`proaches [10,11,12,13,14,15],
`there are many similar-
`
`ities in motivation and philosophy, particularly with
`Hewitt’s work on Actors, Orgs and Teams.
`Both
`for example, are predicated
`IAs and Actor/Orgs.
`on local, modular models of computation
`that deal
`with complexity
`the same way real-world organiza-
`tions do: by delegating
`responsibilities
`for tasks
`to
`individual agents and leaving the details of coordina-
`tion up to them. Extending
`the analogy, the shared
`archive
`in Kornfield and Hewitt’s scientific commu-
`nity metaphor
`[16] plays a role similar
`to the MKS
`model, both facilitating
`agent interactions.
`The dif-
`ferences between
`IAs and other DA1 paradigms are
`largely ones of emphasis. The emphasis
`in our work
`is squarely on finding pragmatic approaches
`to co-
`operative, distributed problem solving (CDPS)
`[12]
`that work in real-world organizations. Our pragma-
`tism should not be construed as being anti-theoretical.
`On the contrary,
`the IA framework provides an ideal
`real-world environment for experimenting with sophis-
`ticated agents and formal approaches
`like Hewitt’s.
`In the future, we hope to collaborate with DA1 re-
`searchers on establishing a set of common, tested pro-
`tocols
`that would enable agents with sophisticated
`planning and reasoning capabilities
`to be integrated
`into our framework.
`is that we are fundamentally com-
`Another distinction
`mitted to a cooperative man-machine approach to en-
`terprise
`integration. We believe that completely au-
`tomated
`factories are neither possible nor desirable
`in the foreseeable
`future. We are therefore concerned
`with how human and computerized agents can interact
`symbiotically, and how automation can be achieved in
`an orderly, incremental way. In this sense, our work
`owes an intellectual debt to concepts from collabor&
`tion theory and computer-supported
`cooperative work
`[3,17,18,19,20,21,22].
`However, most CSCW research
`to date has not explicitly concerned
`itself with shar-
`ing decision making responsibilities between humans
`and computer agents, a central focus of our work. We
`would now like to combine multi-media CSCW
`tools
`with our Personal Assistants so that the IA framework
`supports all information-related
`enterprise activities
`in an integrated way. It is our belief that such an in-
`terdisciplinary approach is precisely what is needed for
`distributed AI to have a major impact on mainstream
`information
`technology.
`
`Acknowledgement
`
`The authors gratefully acknowledge our colleagues,
`J. Glicksman
`and B. Hitson of Enterprise
`Integra-
`tion Technologies Corporation, Paul Losleben of Stan-
`ford CIS, and Victor Lesser, a Stanford visitor from
`the University of Massachusetts,
`for their stimulating
`comments and constructive critiques.
`
`PAN & TENENBAUM
`
`211
`
`Petitioner STMICROELECTRONICS, INC.,
`Ex. 1021, IPR2022-00681, Pg. 6
`
`

`

`eferences
`
`PI
`
`PI
`
`PI
`
`PI
`
`[l] J. Y-C, Pan, J. M. Tenenbaum, and J. Glicksman,
`“A Framework
`for Knowledge-Based Computer-
`Integrated Manufacturing,”
`IEEE
`Trans.
`on
`Semiconductor Manufacturing,
`SM-2, 2, pp. 33-
`46, May 1989.
`J. Glicksman, B. L. Hitson, J. Y-C. Pan, and J,
`M. Tenenbaum, “MKS: A Conceptually Central-
`ized Knowledge Service for Distributed CIM En-
`(to be published
`in Journal of In-
`vironments,*
`telligent Manufacturing,
`1991).
`“Understanding
`T. Winograd
`and F. Flores,
`Computers and Cognition: A New Foundation for
`Design,” pp. 220,lv Norwood, NJ: Ablex, 1986.
`Paperback
`issued by Addison-Wesley, 1987.
`J. Y-C. Pan and J. M. Tenenbaum, “PIES: An
`Engineer’s Do-It-Yourself Knowledge System for
`Interpretation of Parametric Test Data”, AI Mag-
`azine, Volume VII, No. 4, pp. 62-71, Fall 1986.
`P. Dishaw
`and
`J. Y-C. Pan,
`“AESOP: A
`Simulation-Based Knowledge System Approach
`to CMOS Process Diagnosis,”
`IEEE Transactions
`on Semiconductor Manufacturing, Vol. SM-2, No.
`3, 1989.
`J. Y-C. Pan and J. M. Tenenbaum, “An Intel-
`ligent Agent Framework
`for Enterprise
`Integra-
`tion,”
`to appear
`in IEEE Transactions
`on Sys-
`tems, Man, and Cybernetics
`- Special
`Issue on
`Distributed Artificial
`Intelligence, 1991.
`R. G. Smith, P. S. Barth, and R. L. Young, “A
`Substrate
`for Object-Oriented
`Interface Design,”
`in Research Directions
`in Object-Oriented
`Pro-
`gramming, B. Shriver and P. Wegner (eds.), MIT
`Press, Cambridge, MA.. pp. 253-315, 1987.
`181
`B. Hitson,
`for a
`“Distributed
`Infrastructure
`Prolific Manufacturing Enterprise,” Proceedings,
`Hawaii International Conference on System Sci-
`ences, January 1991.
`D. R. Brown, M. R. Cutkosky, and J. M. Tenen-
`baum, “Next-Cut: A Second Generation Frame-
`work for Concurrent Engineering,”
`to appear
`in
`Aided Cooperative
`Product Develop-
`Computer
`ment, D. Sriram and R. Logcher (Eds.), Springer-
`Verlag, 1991.
`PO1
`L. P. Kaelbling and S. J. Rosenschein, “Action
`and Planning
`in Embedded Agents,” Robotics
`and Autonomous Systems, vol. 6, nos. l&2, June
`1990.
`
`PI
`
`PI
`
`PI
`
`212
`
`MULTIAGENT ARCHITECTURES
`
`[ll] C. Hewitt, “Offices are open systems,” ACM
`Transactions
`on Office
`Information
`Systems,
`4(3):271-287,
`July 1986.
`and D.D. Corkill,
`5.H. Durfee, V.R. Lesser,
`Cooperative Distributed Problem Solving,” The
`Intelligence, Volume IV,
`Yandbook of Artificial
`IL.B. Barr, P. Cohen, and E. Feigenbaum
`(eds.),
`Addison Wesley, 1989, pp. 83-147.
`of
`g. J. Nilsson, “Action Nets,” Proceedings
`;he Rochester Planning Workshop: From Formal
`systems to Practical Systems, J. Tenenberg, et al.
`leds.), University of Rochester, Rochester, New
`York, 1989.
`
`J. S. Rosenschein and M. R. Genesereth, “Deals
`among rational agents,” Proceedings 9th IJCAI,
`pp. 91-99, 1985.
`Programming,”
`“Agent-Oriented
`Y. Shoham,
`Technical Report
`STAN-CS-90-1335,
`Stanford
`University, 1990.
`W.A. Kornfiel