software agents in support of human argument mapping
TRANSCRIPT
Software Agents in Support of Human Argument Mapping
1 http://creativecommons.org/licenses/by-nc/2.0/uk
3rd International Conference on Computational Modelling of Argument Desenzano del Garda, Italy, 8-10 Sept. 2010
Simon Buckingham Shum
Knowledge Media Institute Open University
Jack Park
Knowledge Media Institute Open University
Maarten Sierhuis
NASA Ames Research Center Technical University of Delft Carnegie Mellon University SV
Matthew Brown
Carnegie Mellon University SV University of Utah
overview the challenge + vision
background: IBIS, Compendium, Brahms
progress to date: human/agent argument mapping + multiagent simulation @NASA
new work: Brahms agent-enabling Compendium
Brahms IBIS-agent simulation of dialogue
future work
2
Current tools &
practices for discourse and
problem analysis
Argumentation Theory
COMMA research
?
Our challenge as an applied research discipline
semiformal bridge
Current tools &
practices for discourse and
problem analysis
Argumentation Theory
COMMA research
into logics
? ?
A Human-Centred Computing strategy
Annotation Hypertext
Visualization e-Deliberation
e-Learning UX design
The vision: Computer-Supported Collaborative Argumentation integrated into Work Systems
Agents
Simplified subset of discourse
Modelling & Simulating
Work Systems
Humans Discourse in
authentic work systems
Affordances & Services
The vision: Computer-Supported Collaborative Argumentation integrated into Work Systems
Agents Simplified subset of
discourse
Modelling & Simulating
Work Systems
Humans Discourse in
authentic work systems
Affordances & Services
Fraught with politics, emotion, pressure, information overload, competing agendas, high expertise but poor
argumentation skills and low tolerance of new ICT.
e.g. cases where Compendium has been used: redesigning federal airspace; environmental protection policy; improving Shuttle launch procedures; HIV/AIDS
prevention strategy; participatory urban planning
Help manage attention, coordination and reasoning in a
dynamic environment with information overload
9
Compendium Java application:
Nodes can be embedded in multiple maps, can be specialized with Tags, and can link to source documents
>80,000 downloads by >59,000 unique IP numbers Active user community and small developer community
visual hypermedia for managing the connections between ideas formally and informally
10
Real-time dialogue/argument mapping
Jeff Conklin, developer of gIBIS and QuestMap,
& Dialogue/Issue Mapping methods www.cognexus.org
Tim van Gelder, developer of Rationale & bCisive & Argument Mapping methods www.austhinkconsulting.com
Issue/Argument Mapping emerges from the labs
13
Online Deliberation: Emerging Tools Workshop Online Deliberation 2010, Leeds UK (30 June – 2 July) www.olnet.org/odet2010
ESSENCE: E-Science, Sensemaking & Climate Change
www.projects.kmi.open.ac.uk/essence
Brahms agent-based work practice modelling
& simulation system 16
Work is like a symphony, Well rehearsed, but always different
Work Practice Modeling
Groups & Agents work as activities beliefs trigger work
Collaboration between Agents agents react to and interact with
other agents same time/same place same time/different place different time/same place different time/different place
Work Practice Modeling (cont/d)
Tools & Artifacts tools used in activities artifacts created in activities
Environment/Geography agents have a location artifacts have a location detecting real-world facts
Communication is situated the means of communication
depends on the situation impacts efficiency of work
18
Group = Student, Agent = Alex
Geography = Berkeley, CA
Belief = Alex is hungry
Activity = Eating
Workframe = When hungry go eat
Object = Money, Debit card, ATM
Thoughtframe = If no money go to the ATM machine
Brahms Agent Environment http://www.agentisolutions.com
Composer for building models
Compiler for compiling models
Virtual Machine for simulating models
Agent Viewer for viewing simulations
Mission Control Center, International Space Station: Brahms multiagent Orbital Communications Adaptor Mirroring System [24] Sierhuis, et al., AAMA Conf. 2009
21
23
NASA Mobile Agents Field Trials: Simulating an Earth/Mars work system [16, 25]
http://projects.kmi.open.ac.uk/coakting/nasa (view interactive IBIS maps in Safari browser)
24
NASA Mobile Agents Field Trials: Simulating an Earth/Mars work system
Scientist (Mars)
Scientist (Earth)
Scientist (Mars)
Scientist (Earth)
Agents (Mars)
Compendium used as a collaboration medium with both humans + agents, reading + writing IBIS maps
Real time “Dialogue Mapping” of NASA science team deliberation (using graphical IBIS, in Compendium)
25
26
NASA Mobile Agents Field Trials Compendium activity plans for surface exploration, constructed by scientists on ‘Earth’, interpreted by software agents on ‘Mars’
The Compendium nodes and relationships in this plan were interpreted by Brahms software agents for monitoring and coordinating astronaut and robot activity during surface explorations.
Copyright, 2004, RIACS/NASA Ames, Open University, Southampton University Not to be used without permission
RST-telecon-2005-04-11.i.avi 1:11:57
27
NASA Mobile Agents Field Trials Compendium science data map, generated by software agents, for interpretation by Mars+Earth scientists
The Compendium maps were autonomously created and populated with science data by Brahms software agents that use models of the mission plan, work process, data flow and science data relationships to create the maps.
Copyright, 2004, RIACS/NASA Ames, Open University, Southampton University Not to be used without permission
28
NASA testbed: Compendium-based photo analysis by geologists on ‘Mars’
Copyright, 2004, RIACS/NASA Ames, Open University, Southampton University Not to be used without permission
29
NASA testbed: Compendium scientific feedback map from Earth scientists to Mars colleagues
Copyright, 2004, RIACS/NASA Ames, Open University, Southampton University Not to be used without permission
Compendium-Brahms Use Cases
User generates query seeking nodes in remote map databases
Brahms VM accepts query Brahms VM broadcasts query Remote Brahms VM passes query to Compendium
Adaptor Compendium Adaptor queries database Compendium Adaptor returns query results to Brahms
VM Query result returned to calling agent User selects results Results added to user’s Dialogue/Argument Map
32
How to enable agents to conduct IBIS conversations?
The map ≠ the discussion for humans, but for agents, the map = the discussion
IBIS Dialogue Mapping benefits from human intelligence to take turns, summarise and link utterances — but agents can respond simultaneously and identically, potentially resulting in duplicate nodes
Thus, there is need for a facilitator agent to maintain the structure of the argumentation structure and ensure there are no duplicate nodes
IBIS Agent Interfaces
IBISParticipantAgent preArgumentationActivity()
Defines the actions taken by an agent before the argumentation begins, this may include the sending of the initial IBIS nodes that start the argumentation
postArgumentationActivity() Defines the actions taken by an agent after the argumentation has
concluded, this may include deciding the outcome of the argumentation
processQuestionNode(IBISNode node), processIdeaNode(IBISNode node), processProNode(IBISNode node), processConNode(IBISNode node)
Defines the actions taken by an agent when processing the various types of IBIS nodes, this may include the creation of new beliefs and/or responding with an IBIS node
IBIS Agent Interfaces (continued)
IBISFacilitatorAgent checkForDuplicate(IBISNode node)
Defines the process by which IBIS nodes are determined to be unique or duplicate
Collaborative Convective Forecast Product (CCFP)
As a consensus forecast created through online textchat by
meteorologists representing different organizations
Example CCFP, used for FAA strategic planning around severe weather.
AWC Forecaster leads, presenting their forecast for
discussion
Textchat shown to produce inefficient
dialogue, motivating agent simulation as
IBIS moves
CCFP Chat Scenario Setup
Simulation consists of: Agents representing CCFP chat participants (ZNY, ZDI,
AWCForecaster) Each agent has initial beliefs about the weather forecaster ZNY and ZDI have the ability to voice their disagreement with
AWCForecaster's initial forecast
Facilitator agent (CCFPFacilitatorAgent)
43
CCFP Chat Example: Brahms agents’ beliefs about WeatherEvent1 agent AWCForecaster memberof CCFPChatLeader{ initial_beliefs: (WeatherEvent1.name = "weather_event_1"); (WeatherEvent1.confidence = 1); (WeatherEvent1.growth = 1); (WeatherEvent1.tops = 1); (WeatherEvent1.coverage = 1); (WeatherEvent1.speed = 25); (WeatherEvent1.direction = 45); } agent ZID memberof CCFPChatParticipant{ initial_beliefs: (WeatherEvent1.name = "weather_event_1"); (WeatherEvent1.confidence = 1); (WeatherEvent1.growth = 1); (WeatherEvent1.tops = 2); (WeatherEvent1.coverage = 2); (WeatherEvent1.speed = 25); (WeatherEvent1.direction = 45); } agent ZNY memberof CCFPChatParticipant{ initial_beliefs: (WeatherEvent1.name = "weather_event_1"); (WeatherEvent1.confidence = 2); (WeatherEvent1.growth = 2); (WeatherEvent1.tops = 2); (WeatherEvent1.coverage = 1); (WeatherEvent1.speed = 25); (WeatherEvent1.direction = 45); }
The Vision… Agents as described, augmenting work by integrating discourse with work system models
47
Agents drawing on known constraints and arguments in a dynamic work practice environment, e.g.
the location of people or artifacts
the availability of resources or communication channels
the argumentation schemes on which decisions may depend
relevant other conversations/analyses
improved tools for filtering overwhelming information