Using a community of distributed electronic agents to support a highly mobile, ambient computing environment7036128Abstract A highly mobile, ambient computing environment is disclosed for serving a knowledge worker away from the their desk. The present invention allows a knowledge worker to get increased leverage from personal, networked, and interactive computing devices while in their car, airplane seat, or in a conference room with others. An Open Agent Architecture is used to incorporate elements such as GPS agents, speech recognition, and opportunistic connectivity among meeting participants. Communication and cooperation between agents are brokered by one or more facilitators, which are responsible for matching requests, from users and agents, with descriptions of the capabilities of other agents. It is not generally required that a user or agent know the identities, locations, or number of other agents involved in satisfying a request, and relatively minimal effort is involved in incorporating new agents and "wrapping" legacy applications. Extreme flexibility is achieved through an architecture organized around the declaration of capabilities by service-providing agents, the construction of arbitrarily complex goals by users and service-requesting agents, and the role of facilitators in delegating and coordinating the satisfaction of these goals, subject to advice and constraints that may accompany them. Claims We claim: Description FIELD OF THE INVENTION
The symbols '+' and '-', indicating input and output arguments, are at present used only for purposes of documentation. Most parameters and permissions have default values, and specifications of default values may be omitted from the parameters and permissions lists. Defining an agent's capabilities in terms of solvable declarations effectively creates a vocabulary with which other agents can communicate with the new agent. Ensuring that agents will speak the same language and share a common, unambiguous semantics of the vocabulary involves ontology. Agent development tools and services (automatic translations of solvables by the facilitator) help address this issue; additionally, a preferred embodiment of the present invention will typically rely on vocabulary from either formally engineered ontologies for specific domains or from ontologies constructed during the incremental development of a body of agents for several applications or from both specific domain ontologies and incrementally developed ontologies. Several example tools and services are described in Cheyer et al.'s paper entitled "Development Tools for the Open Agent Architecture," as presented at the Practical Application of Intelligent Agents and Multi-Agent Technology (PAAM 96), London, April 1996. Although the present invention imposes no hard restrictions on the form of solvable declarations, two common usage conventions illustrate some of the utility associated with solvables. Classes of services are often preferably tagged by a particular type. For instance, in the example above, the "last_message" and "get_message" solvables are specialized for email, not by modifying the names of the services, but rather by the use of the 'email' parameter, which serves during the execution of an ICL request to select (or not) a specific type of message. Actions are generally written using an imperative verb as the functor of the solvable in a preferred embodiment of the present invention, the direct object (or item class) as the first argument of the predicate, required arguments following, and then an extensible parameter list as the last argument. The parameter list can hold optional information usable by the function. The ICL expression generated by a natural language parser often makes use of this parameter list to store prepositional phrases and adjectives. As an illustration of the above two points, "Send mail to Bob about lunch" will be translated into an ICL request send_message(email, 'Bob Jones', [subject(lunch)]), whereas "Remind Bob about lunch" would leave the transport unspecified (send_message(KIND, 'Bob Jones', [subject(lunch)])), enabling all available message transfer agents (e.g., fax, phone, mail, pager) to compete for the opportunity to carry out the request. Requesting Services An agent preferably requests services of the community of agent by delegating tasks or goals to its facilitator. Each request preferably contains calls to one or more agent solvables, and optionally specifies parameters containing advice to help the facilitator determine how to execute the task. Calling a solvable preferably does not require that the agent specify (or even know of) a particular agent or agents to handle the call. While it is possible to specify one or more agents using an address parameter (and there are situations in which this is desirable), in general it is advantageous to leave this delegation to the facilitator. This greatly reduces the hard-coded component dependencies often found in other distributed frameworks. The agent libraries of a preferred embodiment of the present invention provide an agent with a single, unified point of entry for requesting services of other agents: the library procedure oaa_Solve. In the style of logic programming, oaa_Solve may preferably be used both to retrieve data and to initiate actions, so that calling a data solvable looks the same as calling a procedure solvable. Complex Goal Expressions A powerful feature provided by preferred embodiments of the present invention is the ability of a client agent (or a user) to submit compound goals of an arbitrarily complex nature to a facilitator. A compound goal is a single goal expression that specifies multiple sub-goals to be performed. In speaking of a "complex goal expression" we mean that a single goal expression that expresses multiple sub-goals can potentially include more than one type of logical connector (e.g., AND, OR, NOT), and/or more than one level of logical nesting (e.g., use of parentheses), or the substantive equivalent. By way of further clarification, we note that when speaking of an "arbitrarily complex goal expression" we mean that goals are expressed in a language or syntax that allows expression of such complex goals when appropriate or when desired, not that every goal is itself necessarily complex. It is contemplated that this ability is provided through an interagent communication language having the necessary syntax and semantics. In one example, the goals may take the form of compound goal expressions composed using operators similar to those employed by PROLOG, that is, the comma for conjunction, the semicolon for disjunction, the arrow for conditional execution, etc. The present invention also contemplates significant extensions to PROLOG syntax and semantics. For example, one embodiment incorporates a "parallel disjunction" operator indicating that the disjuncts are to be executed by different agents concurrently. A further embodiment supports the specification of whether a given sub-goal is to be executed breadth-first or depth-first. A further embodiment supports each sub-goal of a compound goal optionally having an address and/or a set of parameters attached to it. Thus, each sub-goal takes the form Address:Goal::Parameters where both Address and Parameters are optional. An address, if present, preferably specifies one or more agents to handle the given goal, and may employ several different types of referring expression: unique names, symbolic names, and shorthand names. Every agent has preferably a unique name, assigned by its facilitator, which relies upon network addressing schemes to ensure its global uniqueness. Preferably, agents also have self-selected symbolic names (for example, "mail"), which are not guaranteed to be unique. When an address includes a symbolic name, the facilitator preferably takes this to mean that all agents having that name should be called upon. Shorthand names include 'self' and 'parent' (which refers to the agent's facilitator). The address associated with a goal or sub-goal is preferably always optional. When an address is not present, it is the facilitator's job to supply an appropriate address. The distributed execution of compound goals becomes particularly powerful when used in conjunction with natural language or speech-enabled interfaces, as the query itself may specify how functionality from distinct agents will be combined. As a simple example, the spoken utterance "Fax it to Bill Smith's manager." can be translated into the following compound ICL request: oaa_Solve((manager('Bill Smith', M), fax(it,M,[ ])), [strategy(action)]) Note that in this ICL request there are two sub-goals, "manager('Bill Smith',M)" and "fax(it,M,[ ])," and a single global parameter "strategy(action)." According to the present invention, the facilitator is capable of mapping global parameters in order to apply the constraints or advice across the separate sub-goals in a meaningful way. In this instance, the global parameter strategy(action) implies a parallel constraint upon the first sub-goal; i.e., when there are multiple agents that can respond to the manager sub-goal, each agent should receive a request for service. In contrast, for the second sub-goal, parallelism should not be inferred from the global parameter strategy(action) because such an inference would possibly result in the transmission of duplicate facsimiles. Refining Service Requests In a preferred embodiment of the present invention, parameters associated with a goal (or sub-goal) can draw on useful features to refine the request's meaning. For example, it is frequently preferred to be able to specify whether or not solutions are to be returned synchronously; this is done using the reply parameter, which can take any of the values synchronous, asynchronous, or none. As another example, when the goal is a non-compound query of a data solvable, the cache parameter may preferably be used to request local caching of the facts associated with that solvable. Many of the remaining parameters fall into two categories: feedback and advice. Feedback parameters allow a service requester to receive information from the facilitator about how a goal was handled. This feedback can include such things as the identities of the agents involved in satisfying the goal, and the amount of time expended in the satisfaction of the goal. Advice parameters preferably give constraints or guidance to the facilitator in completing and interpreting the goal. For example, a solution_limit parameter preferably allows the requester to say how many solutions it is interested in; the facilitator and/or service providers are free to use this information in optimizing their efforts. Similarly, a time_limit is preferably used to say how long the requester is willing to wait for solutions to its request, and, in a multiple facilitator system, a level_limit may preferably be used to say how remote the facilitators may be that are consulted in the search for solutions. A priority parameter is preferably used to indicate that a request is more urgent than previous requests that have not yet been satisfied. Other preferred advice parameters include but are not limited to parameters used to tell the facilitator whether parallel satisfaction of the parts of a goal is appropriate, how to combine and filter results arriving from multiple solver agents, and whether the requester itself may be considered a candidate solver of the sub-goals of a request. Advice parameters preferably provide an extensible set of low-level, orthogonal parameters capable of combining with the ICL goal language to fully express how information should flow among participants. In certain preferred embodiments of the present invention, multiple parameters can be grouped together and given a group name. The resulting high-level advice parameters can preferably be used to express concepts analogous to KQML's performatives, as well as define classifications of problem types. For instance, KQML's "ask_all" and "ask_one" performatives would be represented as combinations of values given to the parameters reply, parallel_ok, and solution_limit. As an example of a higher-level problem type, the strategy "math_problem" might preferably send the query to all appropriate math solvers in parallel, collect their responses, and signal a conflict if different answers are returned. The strategy "essay_question" might preferably send the request to all appropriate participants, and signal a problem (i.e., cheating) if any of the returned answers are identical. Facilitation In a preferred embodiment of the present invention, when a facilitator receives a compound goal, its job is to construct a goal satisfaction plan and oversee its satisfaction in an optimal or near optimal manner that is consistent with the specified advice. The facilitator of the present invention maintains a knowledge base that records the capabilities of a collection of agents, and uses that knowledge to assist requesters and providers of services in making contact. FIG. 7 schematically shows data structures 700 internal to a facilitator in accordance with one embodiment of the present invention. Consider the function of a Agent Registry 702 in the present invention. Each registered agent may be seen as associated with a collection of fields found within its parent facilitator such as shown in the figure. Each registered agent may optionally possess a Symbolic Name which would be entered into field 704. As mentioned elsewhere, Symbolic Names need not be unique to each instance of an agent. Note that an agent may in certain preferred embodiments of the present invention possess more than one Symbolic Name. Such Symbolic Names would each be found through their associations in the Agent Registry entries. Each agent, when registered, must possess a Unique Address, which is entered into the Unique Address field 706. With further reference to FIG. 7, each registered agent may be optionally associated with one or more capabilities, which have associated Capability Declaration fields 708 in the parent facilitator Agent Registry 702. These capabilities may define not just functionality, but may further provide a utility parameter indicating, in some manner (e.g., speed, accuracy, etc), how effective the agent is at providing the declared capability. Each registered agent may be optionally associated with one or more data components, which have associated Data Declaration fields 710 in the parent facilitator Agent Registry 702. Each registered agent may be optionally associated with one or more triggers, which preferably could be referenced through their associated Trigger Declaration fields 712 in the parent facilitator Agent Registry 702. Each registered agent may be optionally associated with one or more tasks, which preferably could be referenced through their associated Task Declaration fields 714 in the parent facilitator Agent Registry 702. Each registered agent may be optionally associated with one or more Process Characteristics, which preferably could be referenced through their associated Process Characteristics Declaration fields 716 in the parent facilitator Agent Registry 702. Note that these characteristics in certain preferred embodiments of the present invention may include one or more of the following: Machine Type (specifying what type of computer may run the agent), Language (both computer and human interface). A facilitator agent in certain preferred embodiments of the present invention further includes a Global Persistent Database 720. The database 720 is composed of data elements which do not rely upon the invocation or instantiation of client agents for those data elements to persist. Examples of data elements which might be present in such a database include but are not limited to the network address of the facilitator agent's server, facilitator agent's server accessible network port list, firewalls, user lists, and security options regarding the access of server resources accessible to the facilitator agent. A simplified walk through of operations involved in creating a client agent, a client agent initiating a service request, a client agent responding to a service request and a facilitator agent responding to a service request are including hereafter by way of illustrating the use of such a system. These figures and their accompanying discussion are provided by way of illustration of one preferred embodiment of the present invention and are not intended to limit the scope of the present invention. FIG. 8 depicts operations involved in instantiating a client agent with its parent facilitator in accordance with a preferred embodiment of the present invention. The operations begin with starting the Agent Registration in a step 800. In a next step 802, the Installer, such as a client or facilitator agent, invokes a new client agent. It will be appreciated that any computer entity is capable of invoking a new agent. The system then instantiates the new client agent in a step 804. This operation may involve resource allocations somewhere in the network on a local computer system for the client agent, which will often include memory as well as placement of references to the newly instantiated client agent in internal system lists of agents within that local computing system. Once instantiated, the new client and its parent facilitator establish a communications link in a step 806. In certain preferred embodiments, this communications link involves selection of one or more physical transport mechanisms for this communication. Once established, the client agent transmits it profile to the parent facilitator in a step 808. When received, the parent facilitator registers the client agent in a step 810. Then, at a step 812, a client agent has been instantiated in accordance with one preferred embodiment of the present invention. FIG. 9 depicts operations involved in a client agent initiating a service request and receiving the response to that service request in accordance with a preferred embodiment of the present invention. The method of FIG. 9 begins in a step 900, wherein any initialization or other such procedures may be performed. Then, in a step 902, the client agent determines a goal to be achieved (or solved). This goal is then translated in a step 904 into ICL, if it is not already formulated in it. The goal, now stated in ICL, is then transmitted to the client agent's parent facilitator in a step 906. The parent facilitator responds to this service request and at a later time, the client agent receives the results of the request in a step 908, operations of FIG. 9 being complete in a done step 910. FIG. 10 depicts operations involved in a client agent responding to a service request in accordance with a preferred embodiment of the present invention. Once started in a step 1000, the client agent receives the service request in a step 1002. In a next step 1004, the client agent parses the received request from ICL. The client agent then determines if the service is available in a step 1006. If it is not, the client agent returns a status report to that effect in a step 1008. If the service is available, control is passed to a step 1010 where the client performs the requested service. Note that in completing step 1010 the client may form complex goal expressions, requesting results for these solvables from the facilitator agent. For example, a fax agent might fax a document to a certain person only after requesting and receiving a fax number for that person. Subsequently, the client agent either returns the results of the service and/or a status report in a step 1012. The operations of FIG. 10 are complete in a done step 1014. FIG. 11 depicts operations involved in a facilitator agent response to a service request in accordance with a preferred embodiment of the present invention. The start of such operations in step 1100 leads to the reception of a goal request in a step 1102 by the facilitator. This request is then parsed and interpreted by the facilitator in a step 1104. The facilitator then proceeds to construct a goal satisfaction plan in a next step 1106. In steps 1108 and 1110, respectively, the facilitator determines the required sub-goals and then selects agents suitable for performing the required sub-goals. The facilitator then transmits the sub-goal requests to the selected agents in a step 1112 and receives the results of these transmitted requests in a step 1114. It should be noted that the actual implementation of steps 1112 and 1114 are dependent upon the specific goal satisfaction plan. For instance, certain sub-goals may be sent to separate agents in parallel, while transmission of other sub-goals may be postponed until receipt of particular answers. Further, certain requests may generate multiple responses that generate additional sub-goals. Once the responses have been received, the facilitator determines whether the original requested goal has been completed in a step 1118. If the original requested goal has not been completed, the facilitator recursively repeats the operations 1106 through 1116. Once the original requested goal is completed, the facilitator returns the results to the requesting agent 1118 and the operations are done at 1120. A further preferred embodiment of the present invention incorporates transparent delegation, which means that a requesting agent can generate a request, and a facilitator can manage the satisfaction of that request, without the requester needing to have any knowledge of the identities or locations of the satisfying agents. In some cases, such as when the request is a data query, the requesting agent may also be oblivious to the number of agents involved in satisfying a request. Transparent delegation is possible because agents' capabilities (solvables) are treated as an abstract description of a service, rather than as an entry point into a library or body of code. A further preferred embodiment of the present invention incorporates facilitator handling of compound goals, preferably involving three types of processing: delegation, optimization and interpretation. Delegation processing preferably supports facilitator determination of which specific agents will execute a compound goal and how such a compound goal's sub-goals will be combined and the sub-goal results routed. Delegation involves selective application of global and local constraint and advice parameters onto the specific sub-goals. Delegation results in a goal that is unambiguous as to its meaning and as to the agents that will participate in satisfying it. Optimization processing of the completed goal preferably includes the facilitator using sub-goal parallelization where appropriate. Optimization results in a goal whose interpretation will require as few exchanges as possible, between the facilitator and the satisfying agents, and can exploit parallel efforts of the satisfying agents, wherever this does not affect the goal's meaning. Interpretation processing of the optimized goal. Completing the addressing of a goal involves the selection of one or more agents to handle each of its sub-goals (that is, each sub-goal for which this selection has not been specified by the requester). In doing this, the facilitator uses its knowledge of the capabilities of its client agents (and possibly of other facilitators, in a multi-facilitator system). It may also use strategies or advice specified by the requester, as explained below. The interpretation of a goal involves the coordination of requests to the satisfying agents, and assembling their responses into a coherent whole, for return to the requester. A further preferred embodiment of present invention extends facilitation so the facilitator can employ strategies and advice given by the requesting agent, resulting in a variety of interaction patterns that may be instantiated in the satisfaction of a request. A further preferred embodiment of present invention handles the distribution of both data update requests and requests for installation of triggers, preferably using some of the same strategies that are employed in the delegation of service requests. Note that the reliance on facilitation is not absolute; that is, there is no hard requirement that requests and services be matched up by the facilitator, or that interagent communications go through the facilitator. There is preferably support in the agent library for explicit addressing of requests. However, a preferred embodiment of the present invention encourages employment the paradigm of agent communities, minimizing their development effort, by taking advantage of the facilitator's provision of transparent delegation and handling of compound goals. A facilitator is preferably viewed as a coordinator, not a controller, of cooperative task completion. A facilitator preferably never initiates an activity. A facilitator preferably responds to requests to manage the satisfaction of some goal, the update of some data repository, or the installation of a trigger by the appropriate agent or agents. All agents can preferably take advantage of the facilitator's expertise in delegation, and its up-to-date knowledge about the current membership of a dynamic community. The facilitator's coordination services often allows the developer to lessen the complexity of individual agents, resulting in a more manageable software development process, and enabling the creation of lightweight agents. Maintaining Data Repositories The agent library supports the creation, maintenance, and use of databases, in the form of data solvables. Creation of a data solvable requires only that it be declared. Querying a data solvable, as with access to any solvable, is done using oaa_Solve. A data solvable is conceptually similar to a relation in a relational database. The facts associated with each solvable are maintained by the agent library, which also handles incoming messages containing queries of data solvables. The default behavior of an agent library in managing these facts may preferably be refined, using parameters specified with the solvable's declaration. For example, the parameter single_value preferably indicates that the solvable should only contain a single fact at any given point in time. The parameter unique_values preferably indicates that no duplicate values should be stored. Other parameters preferably allow data solvables use of the concepts of ownership and persistence. For implementing shared repositories, it is often preferable to maintain a record of which agent created each fact of a data solvable with the creating agent being preferably considered the fact's owner. In many applications, it is preferable to remove an agent's facts when that agent goes offline (for instance, when the agent is no longer participating in the agent community, whether by deliberate termination or by malfunction). When a data solvable is declared to be non-persistent, its facts are automatically maintained in this way, whereas a persistent data solvable preferably retains its facts until they are explicitly removed. A further preferred embodiment of present invention supports an agent library through procedures by which agents can update (add, remove, and replace) facts belonging to data solvables, either locally or on other agents, given that they have preferably the required permissions. These procedures may preferably be refined using many of the same parameters that apply to service requests. For example, the address parameter preferably specifies one or more particular agents to which the update request applies. In its absence, just as with service requests, the update request preferably goes to all agents providing the relevant data solvable. This default behavior can be used to maintain coordinated "mirror" copies of a data set within multiple agents, and can be useful in support of distributed, collaborative activities. Similarly, the feedback parameters, described in connection with oaa_Solve, are preferably available for use with data maintenance requests. A further preferred embodiment of present invention supports ability to provide data solvables not just to client agents, but also to facilitator agents. Data solvables can preferably created, maintained and used by a facilitator. The facilitator preferably can, at the request of a client of the facilitator, create, maintain and share the use of data solvables with all the facilitator's clients. This can be useful with relatively stable collections of agents, where the facilitator's workload is predictable. Using a Blackboard Style of Communication In a further preferred embodiment of present invention, when a data solvable is publicly readable and writable, it acts essentially as a global data repository and can be used cooperatively by a group of agents. In combination with the use of triggers, this allows the agents to organize their efforts around a "blackboard" style of communication. As an example, the "DCG-NL" agent (one of several existing natural language processing agents), provides natural language processing services for a variety of its peer agents, expects those other agents to record, on the facilitator, the vocabulary to which they are prepared to respond, with an indication of each word's part of speech, and of the logical form (ICL sub-goal) that should result from the use of that word. In a further preferred embodiment of present invention, the NL agent, preferably when it comes online, preferably installs a data solvable for each basic part of speech on its facilitator. For instance, one such solvable would be: solvable(noun(Meaning, Syntax), [ ], [ ]) Note that the empty lists for the solvable's permissions and parameters are acceptable here, since the default permissions and parameters provide appropriate functionality. A further preferred embodiment of present invention incorporating an Office Assistant system as discussed herein or similar to the discussion here supports several agents making use of these or similar services. For instance, the database agent uses the following call, to library procedure oaa_AddData, to post the noun 'boss', and to indicate that the "meaning" of boss is the concept 'manager': oaa_AddData(noun(manager, atom(boss)), [address(parent)]) Autonomous Monitoring with Triggers A further preferred embodiment of present invention includes support for triggers, providing a general mechanism for requesting some action be taken when a set of conditions is met. Each agent can preferably install triggers either locally, for itself, or remotely, on its facilitator or peer agents. There are preferably at least four types of triggers: communication, data, task, and time. In addition to a type, each trigger preferably specifies at least a condition and an action, both preferably expressed in ICL. The condition indicates under what circumstances the trigger should fire, and the action indicates what should happen when it fires. In addition, each trigger can be set to fire either an unlimited number of times, or a specified number of times, which can be any positive integer. Triggers can be used in a variety of ways within preferred embodiments of the present invention. For example, triggers can be used for monitoring external sensors in the execution environment, tracking the progress of complex tasks, or coordinating communications between agents that are essential for the synchronization of related tasks. The installation of a trigger within an agent can be thought of as a representation of that agent's commitment to carry out the specified action, whenever the specified condition holds true. Communication triggers preferably allow any incoming or outgoing event (message) to be monitored. For instance, a simple communication trigger may say something like: "Whenever a solution to a goal is returned from the facilitator, send the result to the presentation manager to be displayed to the user." Data triggers preferably monitor the state of a data repository (which can be maintained on a facilitator or a client agent). Data triggers' conditions may be tested upon the addition, removal, or replacement of a fact belonging to a data solvable. An example data trigger is: "When 15 users are simultaneously logged on to a machine, send an alert message to the system administrator." Task triggers preferably contain conditions that are tested after the processing of each incoming event and whenever a timeout occurs in the event polling. These conditions may specify any goal executable by the local ICL interpreter, and most often are used to test when some solvable becomes satisfiable. Task triggers are useful in checking for task-specific internal conditions. Although in many cases such conditions are captured by solvables, in other cases they may not be. For example, a mail agent might watch for new incoming mail, or an airline database agent may monitor which flights will arrive later than scheduled. An example task trigger is: "When mail arrives for me about security, notify me immediately." Time triggers preferably monitor time conditions. For instance, an alarm trigger can be set to fire at a single fixed point in time (e.g., "On Dec. 23rd at 3 pm"), or on a recurring basis (e.g., "Every three minutes from now until noon"). Triggers are preferably implemented as data solvables, declared implicitly for every agent. When requesting that a trigger be installed, an agent may use many of the same parameters that apply to service and data maintenance requests. A further preferred embodiment of present invention incorporates semantic support, in contrast with most programming methodologies, of the agent on which the trigger is installed only having to know how to evaluate the conditional part of the trigger, not the consequence. When the trigger fires, the action is delegated to the facilitator for execution. Whereas many commercial mail programs allow rules of the form "When mail arrives about XXX, [forward it, delete it, archive it]", the possible actions are hard-coded and the user must select from a fixed set. A further preferred embodiment of present invention, the consequence of a trigger may be any compound goal executable by the dynamic community of agents. Since new agents preferably define both functionality and vocabulary, when an unanticipated agent (for example, a fax agent) joins the community, no modifications to existing code is required for a user to make use of it—"When mail arrives, fax it to Bill Smith." The Agent Library In a preferred embodiment of present invention, the agent library provides the infrastructure for constructing an agent-based system. The essential elements of protocol (involving the details of the messages that encapsulate a service request and its response) are preferably made transparent to simplify the programming applications. This enables the developer to focus functionality, rather than message construction details and communication details. For example, to request a service of another agent, an agent preferably calls the library procedure oaa_Solve. This call results in a message to a facilitator, which will exchange messages with one or more service providers, and then send a message containing the desired results to the requesting agent. These results are returned via one of the arguments of oaa_Solve. None of the messages involved in this scenario is explicitly constructed by the agent developer. Note that this describes the synchronous use of oaa_Solve. In another preferred embodiment of present invention, an agent library provides both intraagent and interagent infrastructure; that is, mechanisms supporting the internal structure of individual agents, on the one hand, and mechanisms of cooperative interoperation between agents, on the other. Note that most of the infrastructure cuts across this boundary with many of the same mechanisms supporting both agent internals and agent interactions in an integrated fashion. For example, services provided by an agent preferably can be accessed by that agent through the same procedure (oaa_Solve) that it would employ to request a service of another agent (the only difference being in the address parameter accompanying the request). This helps the developer to reuse code and avoid redundant entry points into the same functionality. Both of the preferred characteristics described above (transparent construction of messages and integration of intraagent with interagent mechanisms) apply to most other library functionality as well, including but not limited to data management and temporal control mechanisms. Source Code Appendix Stored on a Compact Disc in Parent Application Ser. No. 09/225,198, U.S. Pat. No. 6,851,115 Source code for version 2.0 of the OAA software product is included as an appendix hereto, and is incorporated herein by reference. The code includes an agent library, which provides infrastructure for constructing an agent-based system. The library's several families of procedures provide the functionalities discussed above, as well as others that have not been discussed here but that will be sufficiently clear to the interested practitioner. For example, declarations of an agent's solvables, and their registration with a facilitator, are managed using procedures such as oaa_Declare, oaa_Undeclare, and oaa_Redeclare. Updates to data solvables can be accomplished with a family of procedures including oaa_AddData, oaa_RemoveData, and oaa_ReplaceData. Similarly, triggers are maintained using procedures such as oaa_AddTrigger, oaa_RemoveTrigger, and oaa_ReplaceTrigger. The provided source code also includes source code for an OAA Facilitator Agent. The source code appendix is offered solely as a means of further helping practitioners to construct a preferred embodiment of the invention. By no means is the source code intended to limit the scope of the present invention. Illustrative Applications To further illustrate the technology of the preferred embodiment, we will next present and discuss two sample applications of the present inventions. Unified Messaging A further preferred embodiment of present invention incorporates a Unified Messaging application extending the Automated Office application presented previously herein with an emphasis on ubiquitous access and dynamic presentation of the information and services supported by the agent community. The agents used in this application are depicted in FIG. 12. A hypothetical example of realistic dialog using a preferred embodiment of the present invention can provide insight into how systems may preferably be built using the present invention. In this scenario, the user, with only a telephone as an interface, is planning a trip to Boston where he will soon give a presentation. Capitalized sentences are phrases spoken by the user into the telephone and processed by a phone agent 452. Responses, unless otherwise indicated, are spoken by the system using text-to-speech generation agent 454. 1.1 Welcome to SRI International. Please enter your user ID and password. <User enters touchtone ID and password>
Every user interface agent 408, including the telephone agent 452, should know the identify of its user. This information is used in resolving anaphoric references such as "me" and "I", and allows multiple user interfaces operated by the same user to work together. 1.2 WHAT IS TODAY'S SCHEDULE? Here is today's schedule for Adam Cheyer:
If the user is operating both a graphical user interface and a telephone, as described in conjunction with the Automated Office application, the result of this spoken request is to display a calendar window containing the current schedule. In this case, with no graphical display available, the GEN_NL agent 1202 is tasked to produce a spoken response that can be played over the phone. GEN_NL shares the same dynamic vocabulary and phrasal rules as the natural language parser DCG_NL 426, and contains strategies for producing responses to queries using either simple or list-based multimedia utterances. 1.3 FIND FRIDAY'S WEATHER IN BOSTON. The weather in Boston for Friday is as follows:
In addition to data accessible from legacy applications, content may be retrieved by web-reading agents which provide wrappers around useful websites. 1.4 FIND ALL NEW MAIL MESSAGES. There are 2 messages available. Message 1, from Mark Tierny, entitled "OAA meeting." 1.5 NEXT MESSAGE Message 2, from Jennifer Schwefler, entitled "Presentation Summary." 1.6 PLAY IT. This message is a multipart MIME-encoded message. There are two parts. Part 1. (Voicemail message, not text-to speech): Thanks for taking part as a speaker in our conference. The schedule will be posted soon on our homepage. 1.7 NEXT PART Part 2. (read using text-to-speech): The presentation home page is http://www . . . . 1.8 PRINT MESSAGE Command executed. Mail messages are no longer just simple text documents, but often consist of multiple subparts containing audio files, pictures, webpages, attachments and so forth. When a user asks to play a complex email message over the telephone, many different agents may be implicated in the translation process, which would be quite different given the request "print it." The challenge is to develop a system which will enable agents to cooperate in an extensible, flexible manner that alleviates explicit coding of agent interactions for every possible input/output combination. In a preferred embodiment of the present invention, each agent concentrates only on what it can do and on what it knows, and leaves other work to be delegated to the agent community. For instance, a printer agent 1204, defining the solvable print(Object,Parameters), can be defined by the following pseudo-code, which basically says, "If someone can get me a document, in either POSTSCRIPT or text form, I can print it.".
In the above example, since an email message is the salient document, the mail agent 442 will receive a request to produce the message as POSTSCRIPT. Whereas the mail agent 442 may know how to save a text message as POSTSCRIPT, it will not know what to do with a webpage or voicemail message. For these parts of the message, it will simply send oaa_Solve requests to see if another agent knows how to accomplish the task. Until now, the user has been using only a telephone as user interface. Now, he moves to his desktop, starts a web browser 436, and accesses the URL referenced by the mail message. 1.9 RECORD MESSAGE Recording voice message. Start speaking now. 1.10 THIS IS THE UPDATED WEB PAGE CONTAINING THE PRESENTATION SCHEDULE. Message one recorded. 1.11 IF THIS WEB PAGE CHANGES, GET IT TO ME WITH NOTE ONE. Trigger added as requested. In this example, a local agent 436 which interfaces with the web browser can return the current page as a solution to the request "oaa_Solve(resolve reference(this, web_page, [ ],Ref),[ ])", sent by the NL agent 426. A trigger is installed on a web agent 436 to monitor changes to the page, and when the page is updated, the notify agent 446 can find the user and transmit the webpage and voicemail message using the most appropriate media transfer mechanism. This example based on the Unified Messaging application is intended to show how concepts in accordance with the present invention can be used to produce a simple yet extensible solution to a multi-agent problem that would be difficult to implement using a more rigid framework. The application supports adaptable presentation for queries across dynamically changing, complex information; shared context and reference resolution among applications; and flexible translation of multimedia data. In the next section, we will present an application which highlights the use of parallel competition and cooperation among agents during multi-modal fusion. Multimodal Map A further preferred embodiment of present invention incorporates the Multimodal Map application. This application demonstrates natural ways of communicating with a community of agents, providing an interactive interface on which the user may draw, write, gesture or speak. In a travel-planning domain illustrated by FIG. 13, available information includes hotel, restaurant, and tourist-site data retrieved by distributed software agents from commercial Internet sites. Some preferred types of user interactions and multimodal issues handled by the application are illustrated by a brief scenario featuring working examples taken from the current system. Sara is planning a business trip to San Francisco, but would like to schedule some activities for the weekend while she is there. She turns on her laptop PC, executes a map application, and selects San Francisco. 2.1 [Speaking] Where is downtown? Map scrolls to appropriate area. 2.2 [Speaking and drawing region] Show me all hotels near here. Icons representing hotels appear. 2.3 [Writes on a hotel] Info? A textual description (price, attributes, etc.) appears. 2.4 [Speaking] I only want hotels with a pool. Some hotels disappear. 2.5 [Draws a crosscut on a hotel that is too close to a highway] Hotel disappears 2.6 [Speaking and circling] Show me a photo of this hotel. Photo appears. 2.7 [Points to another hotel] Photo appears. 2.8 [Speaking] Price of the other hotel? Price appears for previous hotel. 2.9 [Speaking and drawing an arrow] Scroll down. Display adjusted. 2.10 [Speaking and drawing an arrow toward a hotel] What is the distance from this hotel to Fisherman's Wharf? Distance displayed. 2.11 [Pointing to another place and speaking] And the distance to here? Distance displayed. Sara decides she could use some human advice. She picks up the phone, calls Bob, her travel agent, and writes Start collaboration to synchronize his display with hers. At this point, both are presented with identical maps, and the input and actions of one will be remotely seen by the other. 3.1 [Sara speaks and circles two hotels] Bob, I'm trying to choose between these two hotels. Any opinions? 3.2 [Bob draws an arrow, speaks, and points] Well, this area is really nice to visit. You can walk there from this hotel. Map scrolls to indicated area. Hotel selected. 3.3 [Sara speaks] Do you think I should visit Alcatraz? 3.4 [Bob speaks] Map, show video of Alcatraz. Video appears. 3.5 [Bob speaks] Yes, Alcatraz is a lot of fun. A further preferred embodiment of present invention generates the most appropriate interpretation for the incoming streams of multimodal input. Besides providing a user interface to a dynamic set of distributed agents, the application is preferably built using an agent framework. The present invention also contemplates aiding the coordinate competition and cooperation among information sources, which in turn works in parallel to resolve the ambiguities arising at every level of the interpretation process: low-level processing of the data stream, anaphora resolution, cross-modality influences and addressee. Low-level processing of the data stream: Pen input may be preferably interpreted as a gesture (e.g., 2.5: cross-out) by one algorithm, or as handwriting by a separate recognition process (e.g., 2.3: "info?"). Multiple hypotheses may preferably be returned by a modality recognition component. Anaphora resolution: When resolving anaphoric references, separate information sources may contribute to resolving the reference: context by object type, deictic, visual context, database queries, discourse analysis. An example of information provided through context by object type is found in interpreting an utterance such as "show photo of the hotel", where the natural language component can return a list of the last hotels talked about. Deictic information in combination with a spoken utterance like "show photo of this hotel" may preferably include pointing, circling, or arrow gestures which might indicate the desired object (e.g., 2.7). Deictic references may preferably occur before, during, or after an accompanying verbal command. Information provided in a visual context, given for the request "display photo of the hotel" may preferably include the user interface agent might determine that only one hotel is currently visible on the map, and therefore this might be the desired reference object. Database queries preferably involving information from a database agent combined with results from other resolution strategies. Examples are "show me a photo of the hotel in Menlo Park" and 2.2. Discourse analysis preferably provides a source of information for phrases such as "No, the other one" (or 2.8). The above list of preferred anaphora resolution mechanisms is not exhaustive. Examples of other preferred resolution methods include but are not limited to spatial reasoning ("the hotel between Fisherman's Wharf and Lombard Street") and user preferences ("near my favorite restaurant"). Cross-modality influences: When multiple modalities are used together, one modality may preferably reinforce or remove or diminish ambiguity from the interpretation of another. For instance, the interpretation of an arrow gesture may vary when accompanied by different verbal commands (e.g., "scroll left" vs. "show info about this hotel"). In the latter example, the system must take into account how accurately and unambiguously an arrow selects a single hotel. Addressee: With the addition of collaboration technology, humans and automated agents all share the same workspace. A pen doodle or a spoken utterance may be meant for either another human, the system (3.1), or both (3.2). The implementation of the Multimodal Map application illustrates and exploits several preferred features of the present invention: reference resolution and task delegation by parallel parameters of oaa_Solve, basic multi-user collaboration handled through built-in data management services, additional functionality readily achieved by adding new agents to the community, domain-specific code cleanly separated from other agents. A further preferred embodiment of present invention provides reference resolution and task delegation handled in a distributed fashion by the parallel parameters of oaa_Solve, with meta-agents encoding rules to help the facilitator make context- or user-specific decisions about priorities among knowledge sources. A further preferred embodiment of present invention provides basic multi-user collaboration handled through at least one built-in data management service. The map user interface preferably publishes data solvables for elements such as icons, screen position, and viewers, and preferably defines these elements to have the attribute "shareable". For every update to this public data, the changes are preferably automatically replicated to all members of the collaborative session, with associated callbacks producing the visible effect of the data change (e.g., adding or removing an icon). Functionality for recording and playback of a session is preferably implemented by adding agents as members of the collaborative community. These agents either record the data changes to disk, or read a log file and replicate the changes in the shared environment. The domain-specific code for interpreting travel planning dialog is preferably separated from the speech, natural language, pen recognition, database and map user interface agents. These components were preferably reused without modification to add multimodal map capabilities to other applications for activities such as crisis management, multi-robot control, and the MVIEWS tools for the video analyst. Improved Scalability and Fault Tolerance Implementations of a preferred embodiment of present invention which rely upon simple, single facilitator architectures may face certain limitations with respect to scalability, because the single facilitator may become a communications bottleneck and may also represent a single, critical point for system failure. Multiple facilitator systems as disclosed in the preferred embodiments to this point can be used to construct peer-to-peer agent networks as illustrated in FIG. 14. While such embodiments are scalable, they do possess the potential for communication bottlenecks as discussed in the previous paragraph and they further possess the potential for reliability problems as central, critical points of vulnerability to systems failure. A further embodiment of present invention supports a facilitator implemented as an agent like any other, whereby multiple facilitator network topologies can be readily constructed. One example configuration (but not the only possibility) is a hierarchical topology as depicted in FIG. 15, where a top level Facilitator manages collections of both client agents 1508 and other Facilitators, 1504 and 1506. Facilitator agents could be installed for individual users, for a group of users, or as appropriate for the task. Note further, that network work topologies of facilitators can be seen as graphs where each node corresponds to an instance of a facilitator and each edge connecting two or more nodes corresponds to a transmission path across one or more physical transport mechanisms. Some nodes may represent facilitators and some nodes may represent clients. Each node can be further annotated with attributes corresponding to include triggers, data, capabilities but not limited to these attributes. A further embodiment of present invention provides enhanced scalability and robustness by separating the planning and execution components of the facilitator. In contrast with the centralized facilitation schemes described above, the facilitator system 1600 of FIG. 16 separates the registry/planning component from the execution component. As a result, no single facilitator agent must carry all communications nor does the failure of a single facilitator agent shut down the entire system. Turning directly to FIG. 16, the facilitator system 1600 includes a registry/planner 1602 and a plurality of client agents 1612-1616. The registry/planner 1604 is typically replicated in one or more locations accessible by the client agents. Thus if the registry/planner 1604 becomes unavailable, the client agents can access the replicated registry/planner(s). This system operates, for example, as follows. An agent transmits a goal 1610 to the registry planner 1602. The registry/planner 1604 translates the goal into an unambiguous execution plan detailing how to accomplish any sub-goals developed from the compound goal, as well as specifying the agents selected for performing the sub-goals. This execution plan is provided to the requesting agent which in turn initiates peer-to-peer interactions 1618 in order to implement the detailed execution plan, routing and combining information as specified within the execution plan. Communication is distributed thus decreasing sensitivity of the system to bandwidth limitations of a single facilitator agent. Execution state is likewise distributed thus enabling system operation even when a facilitator agent fails. Further embodiments of present invention incorporate into the facilitator functionality such as load-balancing, resource management, and dynamic configuration of agent locations and numbers, using (for example) any of the topologies discussed. Other embodiments incorporate into a facilitator the ability to aid agents in establishing peer-to-peer communications. That is, for tasks requiring a sequence of exchanges between two agents, the facilitator assist the agents in finding one another and establishing communication, stepping out of the way while the agents communicate peer-to-peer over a direct, perhaps dedicated channel. Further preferred embodiments of the present invention incorporate mechanisms for basic transaction management, such as periodically saving the state of agents (both facilitator and client) and rolling back to the latest saved state in the event of the failure of an agent. Distributed Agents in a Highly Mobile, Ambient Computing Environment In another preferred embodiment of the present invention an application of the collaborative OAA architecture is provided which addresses the post-desktop, mobile/ubiquitous computing environment. The present invention addresses the highly mobile computing environment by incorporating elements such as: GPS agents, speech recognition (including other hands-free UI, multi-modal UI), and opportunistic connectivity among meeting participants (utilizing docked or IR-linked PDA's in addition to Internet sites), by using autonomous service-providing electronic agents associated with available resources, such as meeting resources. Cooperative Agents and Recognition Systems for Drivers and Passengers One embodiment of the present invention discloses a human-machine interface for a mobile computing environment such as a car environment or a portable computing device. The present invention provides a natural interface that allows, for example, a driver to interact with the navigation system, control electronic devices, and communicate with the rest of the world much as is possible in the office environment. Passengers are able to use the system to watch TV, browse the net or play games in their private spaces. The system is configurable including choice of languages, and voice output. In addition, speaker recognition technology can be used for resetting preferences and for security. In addition, the present invention enables intelligent collaboration among agents including user interface agents for providing an ambient interface well suited for the mobile environment as just described, as well as location-aware agents providing current positional information through technologies such as Global Positioning System ("GPS"). Such collaboration yields powerful results greatly enhancing the mobile user's experience, as will now be described and illustrated. New technology such as Global Positioning System (GPS), wireless phones, wireless internet, and electronic controls are currently available in cars to improve the way people drive and manage the time spent in automobiles. The present invention manages this heavy flow of data and keeps the cognitive load as low as possible for the driver by providing a small speech-enabled, touch display device that provides a combination of several interfaces. The interfaces are the front end of the OAA framework, discussed above, which allows a community of intelligent agents to work together to achieve user goals. Natural language and multimodal interpretation agents are utilized to build the multimodal system. These are some of the key agents that recognize and extract the meaning of human signals, such as speech or gestures. According to one embodiment of the present invention, different interfaces are used to complete various task. The user can select tabs using both speech or deictic gestures. Each panel provides its own vocabulary and set of commands in addition to the main commands that allow navigation between the tabs. Navigation System FIG. 17 is an illustration showing a navigation panel 1700 in accordance with one embodiment of the present invention. The panel depicted in FIG. 17 contemplates an automobile-based environment, but practitioners of the relevant arts will recognize that with appropriate modification useful embodiments can be created along similar lines for other mobile computing environments such as portable computers, PDA's etc. As discussed above, Multimodal Maps 1702 allow users to navigate maps naturally and query associated databases using speech, handwriting, and 2D gestures in a synergistic fashion utilizing a pen computer. A preferred embodiment of the present invention enables effective collaboration between and among the user interface agents providing Multimodal Map functionality and one or more agents providing current user location information such as GPS. For example, using a similar interface as in FIG. 17 (but preferably replacing the pen with the finger for convenience, in a car), the driver or the passengers can interact with the navigation system, and the system can in turn take advantage of GPS information to plan a route and get location-specific information from local or remote databases displayed on the map. For example, in response to a user voice command such as "I want to go to SRI on Ravenswood Avenue in Menlo Park," one or more GPS agents, map database agents (and/or online search agents), and multimodal map interface agents would preferably collaborate to ascertain the user's current location and display a suggested route from there to the desired destination. and/or by speaking the suggested directions using. The system preferably guides the car along the chosen route using both the multimodal map display (e.g., to support further queries such as "show me the restaurants along the way") and text-to-speech output 1704 for speaking the suggested driving directions aloud. The interactions among all the agents belonging to the system enable a great degree of proactivity from the system. For example, the system would preferably include agents wrapping various operational sensors of a car, such as the fuel level. Triggers might be set, in accordance with the present invention, causing the system to sound an alert signal when low fuel level is detected. One or more GPS agents, database search agents (perhaps for searching the internet online by wireless communication, or alternatively by searching an on-board digital atlas), and multimodal map interface agents would collaborate to display a multimodal map highlighting the nearest gas station. As another illustration, this time outside the automobile setting, consider a user equipped with a handheld computing device enhanced in accordance with the present invention, who is located on a large corporate or academic campus and needs to proceed on foot to his/her next meeting. The user might ask the device "Show me how to get to my next meeting." In a preferred embodiment of the present invention, the user's request—as processed by natural language agent(s)—would lead to collaboration (preferably in accordanc | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||