The key component of an effective performance support system is a memory of expert knowledge and experience which enables the system to provide useful guidance to its users as they work to achieve their goals.
People share their knowledge in many ways, most often by talking – and often in the form of stories that make a point, contain something worth remembering, or offer advice or knowledge relevant to future decisions. People know how to tell stories, and they naturally comprehend stories. The phenomenon is the same for descriptions of everyday human experiences and for much of the advice conveyed by experts.
Our Experts Telling Relevant Advice (EXTRA) approach delivers context-specific expert advice and experience, typically in the form of short, well-told stories, when they are of value to a person solving a difficult problem.
In addition to providing the right kind of content, EXTRA tackles another challenge problem solvers often face - just-in-time access to the right expert when his or her advice will be useful. In the past, advice was limited to who you know. The medical advice someone received, for example, was only as good as the doctor they personally had access to. Technology expands our networks and can potentially give us access to the best and brightest. It gives us an opportunity to seek the opinions of multiple experts, expanding the range of content we have access to. It gives us a deeper exposure to content by showing us experts who agree with each other, others of whom disagree with each other, and still others who can potentially expand the scope of the conversation providing new perspectives. Geography, personal network, access to the "right" people - EXTRA eliminates these constraints. EXTRA ensures that as they make decisions, users have equal access to the best experts in a field.
Four problems arise in creating such a knowledge repository:
To build an online repository of expert experience, we must work with experts who are in a position to tell good and relevant stories. We interview people who are respected for their expertise, and we work with them to capture their best stories on video. A good story is a vivid and memorable telling of personal experience that makes a valuable point and typically lasts no more than two minutes.
Indexing is the foundation of intelligence in an EXTRA system. Information relevant to human problem solving needs cannot simply be indexed according to key words, nor can it be alphabetical, organized by speaker, or topic, or any other of the standard indexing schemes. The indexing scheme must be the one that has been in use for millennia, namely the one that humans use, unconsciously, to store and find stories in their own minds with no obvious effort.
Each story in EXTRA is indexed with conceptual elements that represent the point of a story as well as its surface features. These elements include participating entities, goals, plans, affordances, barriers, risk factors, and outcomes. The system can also combine the interplay among these elements, for example highlighting different themes that are contradictory or reinforcing. In fact, years of research and work on EXTRA have actually helped us achieve a breakthrough in thematic indexing allowing us to capture the complex interplay of multiple themes. These representations are based on decades of research in natural language processing and, more generally, artificial intelligence.
Given the appropriate indexing, relatively straightforward retrieval algorithms suffice to locate and recommend stories to the user. We view matching on thematic features as primary; a story must make the appropriate general point, or convey the appropriate lesson, to be worth telling. After the best possible thematic matches have been found (taking a particular use case into account; see below), contextual (i.e., surface) features serve to identify the thematically appropriate stories that the user will perceive as most directly relevant to his or her situation.
EXTRA’s interaction with a user is modeled on human conversation. If you’ve been diagnosed with a serious form of cancer and want to explore the range of treatment options, would you rather perform a Google-like search for them or talk to a panel of the world’s top oncologists, surgeons, and other medical specialists? Likely you’d strongly prefer expert advice and opinion. Like a good mentor/expert the EXTRA system reacts to situations it finds a user to be in and offers up relevant experiences.
This "conversation" is possible because an EXTRA story corpus is organized using principles of memory organization and memory retrieval derived from human cognition. When people hear a story, they are reminded of other stories that are similar in some way. Similar stories are connected in memory because they share both contextual (surface) and thematic (deep) features. Contextual and thematic "cues" that arise in a conversation remind the participants of additional relevant stories, facts to relate, et cetera. In other words, a conversation provides a rich, but focused, context for generating relevant remindings for retrieving and telling stories.
Beyond the notion of similar stories, we rely on a theory of conversational coherence to structure retrieved stories into logical conversational threads. When a user has heard a particular story only certain types of follow-up stories continue the conversation rather than shifting its focus. For example, a story might support another story, elaborate on it, present a conflicting point of view, provide a causal or correlational explanation, offer advice, et cetera. Variants of the matching algorithm, involving different relationships among thematic and contextual features are used to retrieve stories in such categories. More idiosyncratically, the user’s history while interacting with the system also affects stories that are offered.
EXTRA has been applied in domains as diverse as diagnostic medicine, pharmaceutical clinical trials, military tactics, business consulting, and software development.