DARPA & KR: Closing the Loop
Ron Brachman
Director, Information Processing Technology OfficeDefense Advanced Research
Projects Agency
Approved for Public Release, Distribution Unlimited
SWANS Conference, 4/7/2005 2Approved for Public Release, Distribution Unlimited
DARPA Background
Created in 1958 in response to Sputnik launch
Agency mission:To maintain the technological superiority of the U.S.
military and prevent technological surprise from harming our national security
by sponsoring revolutionary, high-payoff research that bridges the gap between fundamental
discoveries and their military use.
SWANS Conference, 4/7/2005 3Approved for Public Release, Distribution Unlimited
DARPA’s Role in Science and Technology
10B -
NEAR MID FAR
Scie
nce
& T
echn
olog
y $
(FY0
3)
Service S&T5B -
Fundamental Research, Discovery, System Concept Invention
DARPA
6.1 Fundamental
Research
6.2 Applied Research/
Exploratory Development
6.3A Advanced
Technology Development
Types of R&DFunding
(illustrative/informal)
SWANS Conference, 4/7/2005 4Approved for Public Release, Distribution Unlimited
DARPA’s IT Legacy
Workstations,Windows
Workstations,Workstations,WindowsWindows
Internet/NGI Optical, ATM,
Sonet, TCP/IP, Packet
Switching, HTTP
Internet/NGI Internet/NGI Optical, ATM, Optical, ATM,
Sonet, TCP/IP, Sonet, TCP/IP, Packet Packet
Switching, HTTPSwitching, HTTP
Search EnginesSearch EnginesSearch Engines
TrainedResearchers
TrainedTrainedResearchersResearchers
DARPA is credited withbetween a third and a half of all the major innovationsin computer science and technology.”
What Will Be, Michael Dertouzos, Harper Collins
Linear AlgebraLibraries
Linear AlgebraLinear AlgebraLibrariesLibraries
Natural Language & Speech Processing
Natural Language & Natural Language & Speech ProcessingSpeech Processing
PCs, Graphical Interfaces, Icons, Time Sharing,
Firewalls
PCs, Graphical Interfaces, PCs, Graphical Interfaces, Icons, Time Sharing, Icons, Time Sharing,
FirewallsFirewalls
RISC – Microprocessors,Parallel Processing RISC RISC –– Microprocessors,Microprocessors,Parallel Processing Parallel Processing
SWANS Conference, 4/7/2005 5Approved for Public Release, Distribution Unlimited
DARPA’s AI Impact on DoD
Speech and Language Understanding and Translation
PhraselatorTIDES/EARS (automated processing of Arabic dispatches)LCS (Listen-Communicate-Show) spoken language interaction system
SWANS Conference, 4/7/2005 6Approved for Public Release, Distribution Unlimited
DARPA’s AI Impact on DoD
Planning SystemsDART (Dynamic Analysis Replanning Tool)ACPT (Air Campaign Planning Tool)
Decision Support Systems
CPOF (Command Post Of the Future)
Robotic & Autonomous Systems
PackBot
SWANS Conference, 4/7/2005 7Approved for Public Release, Distribution Unlimited
DARPA’s AI Impact on DoD
Simulation/TrainingTacAir-Soar
Image/Signal Understanding
BCAMS (Bosnian Cantonment Area Monitoring System)ASF (Automated Signal Filtering)
Intelligent Agent-Based Systems
NCOT (Network-Centric Operating Technology)
SWANS Conference, 4/7/2005 8Approved for Public Release, Distribution Unlimited
A Working Definition of Intelligence
The ability to acquire, through experience, models of the world (including other entities and self), and use them productively to solve novel problems and deal successfully with unanticipated circumstancesAcquiring models = learning
Including: by observation, exploration and experiment, teaching and coaching, reading
Using models = reasoningIncluding: “mental simulation,” hypotheticals, plausible inference, logical thinking
Will yield robust, adaptable, transparent, supervisable, autonomous systems
SWANS Conference, 4/7/2005 9Approved for Public Release, Distribution Unlimited
IPTO’s Focus
Developing Cognitive Systems:
Systems that know what they’re doing
A cognitive system is one thatcan reason, using substantial amounts of appropriately represented knowledgecan learn from its experience so that it performs better tomorrow than it did todaycan explain itself and be told what to docan be aware of its own capabilities and reflect on its own behavior can respond robustly to surprise
SWANS Conference, 4/7/2005 10Approved for Public Release, Distribution Unlimited
Notional Anatomy of a Cognitive Agent
External EnvironmentExternal Environment
Communication(language,gesture,image)
Prediction,planning
Deliberative Processes
Reflective Processes
Reactive Processes
Perception Action
STM
Sensors Effectors
Other reasoning
LTM(knowledge base)
Concepts
SentencesCognitiv
eAgen
t
Affect
Attention
SWANS Conference, 4/7/2005 11Approved for Public Release, Distribution Unlimited
Notional Anatomy of a Cognitive Agent
External EnvironmentExternal Environment
Communication(language,gesture,image)
Prediction,planning
Deliberative Processes
Reflective Processes
Reactive Processes
Perception Action
STM
Sensors Effectors
Other reasoning
LTM(knowledge base)
Concepts
SentencesCognitiv
eAgen
t
Affect
Attention
LEARNING
SWANS Conference, 4/7/2005 12Approved for Public Release, Distribution Unlimited
Notional Anatomy of a Cognitive Agent
External EnvironmentExternal Environment
Communication(language,gesture,image)
Prediction,planning
Deliberative Processes
Reflective Processes
Reactive Processes
Perception Action
STM
Sensors Effectors
Other reasoning
LTM(knowledge base)
Concepts
SentencesCognitiv
eAgen
t
Affect
Attention
LEARNING
KRKR
SWANS Conference, 4/7/2005 13Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
QA2/QA3
- First-o- Theorem-proving
for Question-Answering
rder logic
1970 1980 1990 2000
SWANS Conference, 4/7/2005 14Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
QA2/QA3
Shakey
STRIPS, A*, Other AI planning
methods (Hierarchial,non-linear, etc.)
1970 1980 1990 2000
SWANS Conference, 4/7/2005 15Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
KL-ONE
QA2/QA3
Shakey
NIKLLOOM
Description Logics
1970 1980 1990 2000
SWANS Conference, 4/7/2005 16Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
Pilot’s Associate (SCI)
KL-ONE
QA2/QA3
Shakey
Cyc
Case-BasedReasoning
1970 1980 1990 2000
SWANS Conference, 4/7/2005 17Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
QA2/QA3
Shakey
Cyc
DART,Multi-agent
Planning
1970 1980 1990 2000
SWANS Conference, 4/7/2005 18Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
I3/DARPA Knowledge-Sharing Initiative
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
QA2/QA3
Shakey
Cyc
KIFKQML
Ontolingua
1970 1980 1990 2000
SWANS Conference, 4/7/2005 19Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
I3/DARPA Knowledge-Sharing Initiative
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
HPKB
RKF
QA2/QA3
Shakey
Cyc
ProjectHALO
1970 1980 1990 2000
SWANS Conference, 4/7/2005 20Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
I3/DARPA Knowledge-Sharing Initiative
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
HPKB
RKF
QA2/QA3
Shakey
CoABS
Cyc
CoABS GridFastC2AP
1970 1980 1990 2000
SWANS Conference, 4/7/2005 21Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
I3/DARPA Knowledge-Sharing Initiative
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
HPKB
RKF
DAML
QA2/QA3
Shakey
CoABS
Cyc
SemanticWeb
1970 1980 1990 2000
SWANS Conference, 4/7/2005 22Approved for Public Release, Distribution Unlimited
DARPA’s Investment in KR&R
I3/DARPA Knowledge-Sharing Initiative
ARPA-Rome Labs Planning Initiative (ARPI)
Pilot’s Associate (SCI)
KL-ONE
HPKB
RKF
DAML
QA2/QA3
Shakey
CoABS
REAL
PAL
Cyc
1970 1980 1990 2000
SWANS Conference, 4/7/2005 23Approved for Public Release, Distribution Unlimited
Key Challenges for the New (Old) Future
Dealing with the messiness of the real worldNon-logical statements (e.g., defaults, statistical assertions)UncertaintyFuzzy conceptsMixed representations (use each for what it’s best)Mismatch of vocabularies – integration, translationScale
Representation of and reasoning about actions, processes, flowsTemporal changeCreation and destruction of objects
Trust and transparencySharing derivations and explanations
Decrease barriers to knowledge creationAccessibility to the untrained, use of the massesAutomatic methods
Please send us a Program Manager!