Cognitive Science Initiative encapsulates expertise

Currently, Sandia reports that just 100,000 concepts in their model's semantic network can yield near carbon-copy fidelity between what their model predicts and what the individuals do in real life. The semantic network holds the knowledge of an individual in a directed graph, the vertices of which represent his or her concepts, and the edges of which represent semantic relations between these concepts.

"Before we had automated concept acquisition we had to get a psychologist to do an interview--but even if the person was available for interviewing, the psychologist had to cross reference every concept with every other concept--an n-squared intractable problem which severely limited the number of concepts we could have in a model," said Wagner. "But today our automated methods can have a hundred thousand concepts, which the computer might have to take a lot of time to process, but at least it's now a tractable problem."

One of the first accomplishments for which the new initiative hopes to get a notch on its belt is improving the training experience of new military recruits. The theory is that if Sandia National Laboratories can create a high-enough fidelity model of a new recruit and an expert war fighter, then they can use those models to optimize the training experience for new recruits.

"Our hypothesis is that if we build a cognitive model of a novice and we compare it to the cognitive model of an expert, then we will be able to tailor the training experiences for each individual. It won't be a one-size-fits-all training experience anymore," said Wagner. "We will be able to pinpoint a person's weaknesses and identify what experiences they need to have to make their cognitive model look more like the cognitive model of an expert."

On the sensor side of development, Sandia is also planning to validate the scientific foundation of their models by doing their own basic research in neuroscience, starting with the creation of specialized MEMS, nanoscale and super-conducting sensors to non-invasively monitor neural behavior.

"We have one project already funded using nanocrystals to perform sub-cellular imaging," said Wagner. "We also want to build wearable sensors using high-temperature super-conducting quantum interference devices [Squids] that do real time monitoring of the brain's electrical activity using MEG [magnetoencephalography]."

Sandia also plans to add a natural language interface to the system so that users can interact with experts--ask their advice using natural language. Ultimately they want to build an encyclopedia of cognitive models--expert personas that can be consulted merely by visiting their web sites. Milestones along the way to the encyclopedia of expert personas will include a "learning by analogy" capability slated for final testing within a year. The team also plans to announce memory and knowledge milestones along the way to its grand vision of encapsulated expertise on-demand.