MIDCA figure
Figure 1: The Metagocnitive Integrated Dual-Cycle Architecture (MIDCA)

MIDCA consists of action-perception cycles at both the cognitive (i.e., object) level and the metacognitive (i.e., meta-) level (see Figure 1). The output side of each cycle consists of intention, planning, and action execution, whereas the input side consists of perception, interpretation, and goal evaluation. A cycle selects a goal and commits to achieving it. The agent then creates a plan to achieve the goal and subsequently executes the planned actions to make the domain match the goal state. The agent perceives changes to the environment resulting from the actions, interprets the percepts with respect to the plan, and evaluates the interpretation with respect to the goal.

At the object level, the cycle achieves goals that change the environment (i.e., ground level). At the meta-level, the cycle achieves goals that change the object level. That is, the metacognitive perception components introspectively monitor the processes and mental state changes at the cognitive level. The action component consists of a meta-level controller that mediates reasoning over an abstract representation of the object level cognition.


Figure 2: User interface for Ronin

Ronin is a community question answering (cQA) platform for people with diverse background to share information and knowledge (see Figure 2). The traditional FAQ archives constructed by the experts or companies where anyone can ask and answer questions on any topic, and people seeking information are connected to those who know the answers. In cQA, the systems can directly return answers to the queried questions instead of a list of relevant documents, thus providing an effective alternative to the traditional ad-hoc information retrieval.

In Ronin we first build measure the similarity between concepts in our questions and questions in our case base using a Word2Vec neural network. We then incorporate task information using a semantic net similarity algorithm we call NetSim. Netsim is based on the algorithm described in Sentence Similarity Using A Semantic Net and Corpus Statistics (Yuhua Li et al). These two similarity measures together enable Ronin to better retrieve related dialogues from its case base and use them as a template when guiding the user through their current dialogue.


Figure 3: The Mission Driven Learning Environment (MiDLE)

The Mission Driven Learning Environment (MiDLE) is a simulation platform currently under development for the Air Force Research Lab (see Figure 3). The purpose of this project is to provide the foundation for a testbed that can provide essential training for pilots in the process of learning how to operate a remotely piloted aircraft. At the center of this training is the notion of commander’s intent. Commander’s intent is traditionally represented by a concise natural language utterance. Its purpose is to provide structure and guidance for a mission by describing the end state, the key tasks to accomplish, and the reasons why. We capture this intent as a strategy and represent it as a set of goals and their priorities and constraints over time. As an example, a very simple commander’s intent text may read: “Survey the surrounding environment, respond appropriately to emergency situations, and above all else protect the platform.” Notice that there is not a detailed procedure to follow, but a list of goals to be accomplished. The use of commander’s intent is valuable because of the magnitude and unpredictable nature of modern warfare. Internally commander’s intent is represented using qualitative process models (QPM). The MiDLE environment infers the trainee’s goal priorities and then compares these to the priorities in the QPM. Feedback is given appropriately, and subsequent lessons are chosen on the basis of this comparison.