The Policy Enabled Next geNeration Internet (PENNI) project aims to lay the technical foundations for a policy, trust, and transparency enabled decentralised
Next Generation Internet (NGI) that can support intelligent agents that act on behalf of humans.
Our hypothesis being that decentralized applications that are able to understand constraints (in the form of goals, preferences, norms, and usage restrictions)
together with trust and transparency mechanisms (in the form of usage control and fact checking) will give citizens more control over how their data is utilised
and will foster trustworthiness in NGI applications, such as intelligent agents.
In this document, we propose a benevolent and trustworthy agent architecture where trust and transparency are first class citizens and demonstrate its effectivess
with the help of a motivating use case scenario.
We sincerely thank Karen Sollins for her feedback on the initial draft of this document.
Introduction
This document describes a benevolent and trustworthy agent architecture that serves as a blueprint for developing well meaning and dependable intelligent software web agents.
We start by describing the various components that emcompass the proposed benevolent and trustworthy agent architecture. Following on from this we discuss the cross-cutting role of
policies, transparency, and compliance checking. We subsequently discuss how requests are handled by the proposed architecture. Finally we demonstrate the effectivess of the proposed architecture
by revisiting the original intelligent software web agent motivating scenario with a particular focus on underlining the need for benevolent and trustworthy agents and
how this is reflected in the proposed architecture.
Aims of the Specification
The overarching aim is to descibe the proposed benevolent and trustworthy agent architecture in detail and to demonstrate how it can be used to support various
intelligent software web agents tasks. Although policies, transparency, and compliance checking are paramount from a benevolence and trust perspective, these functions cannot be
discussed in isolation and thus it is necessary to describe how these functions fit with core agent architecture components.
The examples throughout the document are serialized as [[turtle]].
The examples presented here make use if the following prefixes:
Before going into details on how the Benevolent and Trustworthy Agent (BTA) Architecture can be used to model information, booking, and scheduling requestes we first provide an overview
of the BTA components. .
Benevolent and Trustworthy Agent (BTA) Architecture
Interface Component
The Interface Component (IC) is responsible for interacting with humans, agents, and networked devices.
Sensors are used to perceive the world (e.g., keyboards, cameras, microphones, network ports, etc.).
Actuators are used to perform actions (e.g., screens, printers, headphones, network ports, etc.).
Controller Component
The Controller Component (CC) is responsible for interpreting perceptions, managing internal processes and tasks, and interpreting actions.
Perception Interpreter (PI) parses perceptions received via sensores and forwards requests to the Execution Engine.
Execution Engine (EE) devises execution plans that leverage the deliberative components and executes these plans.
Initiate Process triggers the start of a reasoning or querying process.
Monitor/Control Process ensures that internal reasoning or querying is working as expected and if not take corrective action.
Action Interpreter (AI) is responsible for parsing actions forwarded by the Execution Engine and sending commands to the respective actuator.
Deliberative Component
The Deliberative Component (DC) encompasses the agents knowledge, reasoning, and querying functionality.
The Goal Encoding parses requests and generates goals.
The Reasoning Engine (RE) is responsible for performning reasoning tasks.
Goal Based Reasoning is used for problem solving tasks.
Constraint Selection is used to retrieve constraints that are relevant for a given context.
Compliance Checking involves checking if a policy or log entry is subsumbed by some other policy.
Transparency Provision provides transparncy with respect to the agents behaviour.
The Utility Function helps the agent to make decisions and solve conflicts.
The Goal Assessment verifies if goals have been satisfied and if not generates new goals.
Knowledge Base Component
The Knowledge Base (KB) is responsible for persisting the agents knowledge about itself and its environment.
Goals are used to store the agents current goals.
Beliefs represent the knowledge that the agent believes to be true.
Processes are functions that the agents can execute.
Index/Registery of external knowledge bases and services.
Policies are associated with agent things and services.
Preferences inform agent actions.
Norms specify regulatory or social requirements relating to agents or multi-agent systems.
Logs encode requests received, actions taken, and responses provided.
Querying Component
The Querying Component (QC) allows the agent to execute queries against both internal and external knowledge bases.
Generate Query prepares a query that can be executed against local or remote knowledge bases.
Query Engine is responsible for handling both local and remote queries.
Execute Query can execute queries against local and remote knowledge bases.
Process Response prepares responses.
Task Environment
The agents Environment is a complex space that consists of a variety of different networking technologies, devices, information sources,
applications, and both human and artificial agents.
Agents interact with humans, agents, and networked devices in their environment.
Humans interact with humans, agents, and networked devices in their environment.
Networked Devices interact with humans, agents, and networked devices in their environment.
Performance Assessment is a mechanism used to evaluate the effectivess of an agent.
Request Types
This section describes the general structure of Request messages used to interact with benevolent and trustworthy agents and to inform the corresponding
agent behaviour. Although several request types (e.g., InformationRequest, BookingRequest, SchedulingRequest, TaskRequest)
and the corresponding agent behaviour are described in detail, these request types serve to demonstrate the effectivess of the proposes architecture and are by no means exhaustive.
Benevolent and Trustworthy Agent (BTA) Message Model
Information Request
An information request specifies the information required and where it may be obtained.
Goal: Obtain and return the requested information.
Performance Measure: The infromation has been provided.
Message Structure:
RequestType is set to InfoRequest for this message.
Source the entity that initiated the request.
Destination the entity responsible for fulfilling the request.
Type denotes the information that needs to be retrieved.
Provider denotes where the information may be found.
Constraint represents knowledge that is needed to fulfil the request.
Credential represents the Credential that are needed to fulfil the request.
This section describes how agents handle various request types. Although the internal behaviour for several request types (e.g., InformationRequest,
BookingRequest, SchedulingRequest, TaskRequest) is described in detail, these request types serve to demonstrate the effectivess of the proposes architecture and are by no means exhaustive.
Benevolent and Trustworthy Agent (BTA) Message Model
The following internal behaviour is common to all tasks.
Request Agnostic Agent Behaviour:
The Interface Component Sensors module ReceiveRequest(string request) event handler forwards the request to the Controller Component.
The Controller Component PerceptionInterpreter module InterpretRequest(string request) function parses the request, extracts the message,
and calls the ExecutionEngine module.
The Controller Component ExecutionEngine module CreateProcess(string message) calls the request specific prepare process function.
The Controller Component ResponseInterpreter module PrepareResponse(string response) function prepares a response and calls Interface Component.
The Interface Component Actuator module SendResponse(string response) event handler forwards the response to originator.
Information Request
An information request specifies the information required and where it may be obtained.
The Controller Component ExecutionEngine CreateProcess module Init(string message) function initiates an information create process.
The Controller Component ExecutionEngine CreateProcess module InformationRequest(string message) function calls the following function:
The Controller Component ExecutionEngine CreateProcess module GetSources(string message) retrieves references for the sources to be queried.
The Controller Component ExecutionEngine CreateProcess module PrepareQuery(string message) prepares a query to be executed against the sources.
The Controller Component ExecutionEngine CreateProcess module InformationRequest(string request) function calls the ControlProcess request handler.
The Controller Component ExecutionEngine ControlProcess module Init(string message) function initiates an information control process.
The Controller Component ExecutionEngine ControlProcess module InformationRequest(string action) function calls the following function:
The Controller Component ExecutionEngine ControlProcess module InformationRequest.QueryExecution(string action) executes a query using a query engine. i>
The Controller Component ExecutionEngine ControlProcess module InformationRequest(string request) function calls the prepare response request handler.
Booking Request
A booking request specifies the service required and possible service providers.
The Controller Component ExecutionEngine CreateProcess module Init(string message) function initiates a booking create process.
The Controller Component ExecutionEngine CreateProcess module BookingRequest(string message) function calls the following function:
The Controller Component ExecutionEngine CreateProcess module BookingRequest.GetSources(string message) retrieves references for the sources to be queried.
The Controller Component ExecutionEngine CreateProcess module BookingRequest.PrepareQuery(string message) prepares a query to be executed against the sources.
The Controller Component ExecutionEngine CreateProcess module BookingRequest(string request) function calls the ControlProcess request handler.
The Controller Component ExecutionEngine ControlProcess module Init(string message) function initiates a booking control process.
The Controller Component ExecutionEngine ControlProcess module BookingRequest(string action) function calls the following function:
The Controller Component ExecutionEngine ControlProcess module BookingRequest.QueryExecution(string action) executes a query using a query engine.
The Controller Component ExecutionEngine ControlProcess module BookingRequest(string request) function calls the prepare response request handler.
Scheduling Request
A scheduleing request is a reasoning tasks that needs to be performed by the agent.
The Controller Component ExecutionEngine CreateProcess module Init(string message) function initiates a scheduling create process.
The Controller Component ExecutionEngine CreateProcess module SchedulingRequest(string message) function calls the following function:
The Controller Component ExecutionEngine CreateProcess module SchedulingRequest.GetSources(string message) retrieves references for the sources to be queried.
The Controller Component ExecutionEngine CreateProcess module SchedulingRequest.PrepareReasoningTask(string message) prepares a query to be executed against the sources.
The Controller Component ExecutionEngine CreateProcess module SchedulingRequest(string request) function calls the ControlProcess request handler.
The Controller Component ExecutionEngine ControlProcess module Init(string message) function initiates a scheduling control process.
The Controller Component ExecutionEngine ControlProcess module SchedulingRequest(string action) function calls the following function:
The Controller Component ExecutionEngine ControlProcess module SchedulingRequest.ExecuteReasoningTask(string action) inititates the reasoning engine.
The Controller Component ExecutionEngine ControlProcess module SchedulingRequest(string request) function calls the prepare response request handler.
Task Request
A task request is essentially a collection of actions (i.e., requests) that need to be executed.
The Controller Component ExecutionEngine CreateProcess module Init(string message) function initiates a task create process.
The Controller Component ExecutionEngine CreateProcess module TaskRequest(string message) function calls the following function:
The Controller Component ExecutionEngine CreateProcess module TaskRequest.GetActions(string message) function extracts the actions that need to be executed and inititates the execute
process for each action.
The Controller Component ExecutionEngine CreateProcess module TaskRequest(string request) calls the execute process request handler.
The Controller Component ExecutionEngine ControlProcess module Init(string message) function initiates a task control process.
The Controller Component ExecutionEngine ControlProcess module TaskRequest(string action) function calls the following function:
The Controller Component ExecutionEngine ControlProcess module TaskRequest.ExecuteActions(string action) function executes the actions and sends a response confirming the execution
is complete.
The Controller Component ExecutionEngine ControlProcess module TaskRequest(string request) function calls the prepare response request handler.
An Appointment Scheduling Usecase Scenario
The original semantic web vision was presented with the help of a motivating use case scenario, revolving around Pete and Lucy and their mother who has just found out that
she needs to attend regular physiotherapy sessions. Pete and Lucy ask their personal agents to prepare a physiotherapy appointment schedule such that they are able to share
the chauffeuring duties. Lucy's agent is tasked with finding a physiotherapist who: (i) is covered by their mothers insurance; (ii) has a trusted service rating of very good
or excellent; (iii) is located within a 20 mile radius of their mother's home; and (iv) has appointments that work with Lucy and Pete's busy schedules.
The workflow depicted in \emph{Figure}~\ref{fig:lucys-agent} can be summarised as follows:
Lucy requests that her agent devises a schedule considering the given constraints.
Lucy's agent consults with the doctor's agent in order to retrieve information relating to the prescribed treatment.
Lucy's agent consults with the insurance provider agent in order to find physiotherapists considering the given constraints.
Lucy's agent consults the various physiotheraphy agents in order to retrieve available appointment times.
Lucy's agent asks Pete's agent to give her access to his schedule.
Lucy's agent retrieves here own schedule.
Lucy's agent uses the available appointment times provided by the various physiotherapy service providers, together with Pete's schedule provided by his agent, and Lucy's own
schedule in order to prepare a physiotherapy appointment schedule considering available appointments and Pete's and Lucy's busy schedules.
Lucy's agent shares the proposed schedule with Lucy, and Pete's agent who in turn shares it with Pete.
