What Are Agents?

  • Operate on their own
    • Independent thread of control
    • Only interact with humans when needed
  • Proactive/Reactive
  • Intelligent
    • Capture expert knowledge
    • Learn and adapt to situations, if required

Why Agents?

  • More tasks are automated through computer systems
  • Massive amounts of unstructured and dynamic data
  • More untrained users in the world today
  • Technologically complex world
  • Fit well in decision support environments; mimic human processes
  • Humans and computers perform better combined than alone

What Are Agents, Really?

  • Proxies for human Subject Matter Experts
  • Software components capable of observing and pursuing a set of user-defined objectives based on:
    • Rules, action semantics
    • Reasoning, formal logics (Bayesian, Dempster-Schaefer, Subjective Logic)
    • Belief/disbelief algebra over domain ontologies
    • Knowledge representation (KR) of subject matter
  • Software that executes “autonomously” and acts/interacts “rationally” within the constraints of the situation
  • Software that transforms raw data into actionable knowledge within the semantics of a problem-domain

Similar and Complementary

  • Humans use experience and intuition
  • Humans rely on sensors and systems
  • Humans learn and adapt easily
  • Humans recognize patterns
  • Humans are biased and may become victim to social fallacies

  • Agents use heuristics, optimization, numerical algorithms
  • Agents interface with other systems
  • Agents are impartial and reliable
  • Agent systems are complex and not yet tested operationally

What Can Agents Do?

  • Autonomously monitor and alert the operator
  • Perform repetitive, tedious, or time-consuming tasks on behalf of the human
  • Pro-actively consider and evaluate alternatives
  • Coach and advise the novice users
  • Mimic, adapt and learn from the human users
  • Act as the “functional glue” for integrating disparate systems
  • Provide a secure trusted environment
    • Information integrity
    • Contract enforceability
    • Flexibility, Scalability

 

Spectrum of Agents

  • Approximates large class of cognitive services.
  • Approximates some cognitive services. Amplifies human decision making in some circumstances.
  • Limited Decision Support, System of Systems not purposefully designed as agents.

 

 

AEDGE Agent Evolution

  • Agents initially developed to provide simple behaviors in simulation environments
  • Agents trained to be “Virtual Instructor” for training simulations
  • Agents progress to point of replacing missing members of training team
  • Agents applied to live exercises and deployment
  • Agents self-analyze and learn/improve
  • Agents deployed to live fire environments
  • 21CSI has engaged in all of the above