Strategic research directions in artificial intelligence
Download
1 / 10

Strategic Research Directions in Artificial Intelligence - PowerPoint PPT Presentation


  • 91 Views
  • Uploaded on

Strategic Research Directions in Artificial Intelligence. AFRL/IF Workshop 26-27 June 2003 Ithaca, NY Jared Freeman, Ph.D. Principal Cognitive Scientist Aptima. Aptima’s Work. Training Systems  Decision Support LSI to assess IMINT text reports

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Strategic Research Directions in Artificial Intelligence' - oakley


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
Strategic research directions in artificial intelligence

Strategic Research Directions in Artificial Intelligence

AFRL/IF Workshop

26-27 June 2003

Ithaca, NY

Jared Freeman, Ph.D.

Principal Cognitive Scientist

Aptima


Aptima s work
Aptima’s Work

  • Training Systems  Decision Support

    • LSI to assess IMINT text reports

    • SVM + IBIS rationale models to assess PSYOPS plans

    • Cognitive models + Speech-to-text + NLP to train F-15 & AWACS

    • Speech-to-text + SVM to categorize MOUT comms

    • NLP + HF taxonomies to categorize usability reports

  • Formal Modeling of Organizations

    • Optimization + dynamic clustering to define smallest, fastest human team for a mission given a technology suite

    • Social network modeling of terrorist organizations

  • Military domain analysis

    • Intel, AOC, AWACS, Navy Air wing, urban warfare…


Method
Method

  • Participants

    • Head researcher in an AF training lab

    • Chief of AF intelligence training

    • Leading researcher in design of organizational structure and process

    • Engineer heading DoD human-centered engineering firm

    • Cognitive scientist with 25 years experience in military systems development

    • Others…

  • Question

    • What directions in AI research will serve the AF over the next decade?


Hot topics acquiring data for modeling
Hot Topics: Acquiring Data for Modeling

  • Title: Data mining for human / team models

  • Objective:

    • Use extant data to accelerate development & validation of behavioral and cognitive models

      • Simple: track id

      • Complex: PSYOPS planning

    • Develop and validate performance metrics

  • Challenges:

    • Data acquisition from proprietary, independent, classified systems

    • Data pre-processing

    • Hypothesis generation & testing


Hot topics modeling decision makers
Hot Topics: Modeling Decision Makers

  • Title: Cognitive / Behavioral / Organizational Models

  • Objectives:

    • Increase efficiency w/ synthers (vs. humans) in

      • Large scale experiments, especially at low levels in organizations

      • Individual training of strategic skills (vs. procedural)

    • Analyze strategic expertise at high levels, e.g., JFACC (2-3 star)

    • Test new organizational architectures & processes (TTPs)

    • Test effects of technology insertion in these closed loop systems

    • Develop models of communicative & social behavior

  • Challenges:

    • Efficient development of models from top (theory) & bottom (data)

    • Modeling strategic (vs. procedural) skill

    • Modeling interaction of cognition  decision support

    • Learning from users (batch and real time)


Hot topics modeling decision makers1
Hot Topics: Modeling Decision Makers

  • Title: Organizational management

  • Objectives:

    • Help teams adapt on the fly

      • Speed – when to accelerate

      • Tasking – what tasks to drop

      • Process – what process to adapt

      • Structure – when and how to adapt architecture

  • Challenges:

    • Dynamic organizational models

      • Organizational adaptation

      • Organizational evolution

    • Empirical validation


Hot topics modeling content
Hot Topics: Modeling Content

  • Title: Multi-domain information fusion

  • Objectives:

    • Fuse hard intel (SIGINT) with soft intel (HUMINT, open source)

  • Challenges:

    • Knowledge acquisition re: organizations with multiple experts, multiple data sources

    • Modeling interaction of domain specialists


Hot topics modeling content1
Hot Topics: Modeling Content

  • Title: Ontologies

  • Objectives:

    • Improve the psychological validity of ontologies

  • Challenges:

    • Human categorization is

      • Dynamic: Generative (Barselou)

      • Continuous: From prototypes (robins) to exceptions (penguins) (Rosch)

    • Ontologies as defined in computer science are

      • Static

      • Dichotomous


Hot topics modeling reasoning
Hot Topics: Modeling Reasoning

  • Title: Explaining reasoning

  • Objectives:

    • Justify training assessments & diagnoses: feedback

    • Justify advice to varied specialists, e.g., explain ROE decisions to

      • AWACS Air Weapons Officer

      • AWACS lawyers: Airborne Cmd Element + JAG (in AOC)

  • Challenges:

    • Customize language per user or level of expertise

    • Customize granularity or style of reasoning (psychological validity per paradigm, per level of expertise)

    • Ascertaining when explanations are weak, models are wrong


Hot topics modeling reasoning1
Hot Topics: Modeling Reasoning

  • Title: Supporting reasoning (vs. replacing it)

  • Objectives:

    • “The greatest benefit AI can bring to me … is the ability to know which question to ask.”

    • “The one thing that I note is the glaring emphasis on replacing the "man in the loop" almost to the exclusion of providing performance support…there will always have to be a man in the loop. the AI capability is only an assist…not the answer.”

  • Challenges:

    • Modeling of question asking, critical thinking

    • Dialogue modeling


ad