A Cognitive Assistant for Cancer Team Science

2015-08-18T14:29:44Z (GMT) by Paul Thompson
Douglas Englebart advocated the use of computers for “augmented cognition” in 1962. Even earlier artificial intelligence researchers had debated whether computers should be artificially intelligent in the same way as humans, or as engineered systems that would perform intelligently, but not necessarily in ways recognizably human. Development of a cognitive assistant is in keeping with Englebart’s concept of augmented cognition. Recently IBM developed Watson, a question answering system that was able to outperform two human Jeopardy champions. IBM Watson is being used in several medical research settings for various tasks. For example, Sloan Kettering Cancer Institute developed a Watson cognitive assistant for diagnosing and treating cancer. 

A cognitive assistant can provide recommendations. Two recommender system projects were presented at the 3rd National VIVO Conference in 2012, which could automatically assemble a team for team science based on meta-data available in the VIVO system. VIVO and its companion system, eagle-i, together represent researchers and their publications and resources in biomedical research laboratories. The concept of a reference interview has been discussed for decades in the field of librarianship. This concept has significant overlap with that of the role of a cognitive assistant. A patron consulting a reference librarian at a library often does not have a precise formulation of an information need. Through a discussion with the patron the reference librarian eventually determines what the patron’s information need is. Several years ago Syracuse University had a virtual reference librarian project. Another library science concept worth noting in this connection is that of the Anomalous State of Knowledge, or ASK. In the 1970s Belkin and Oddy built a system based on this concept. A patron often does not know what he or she is seeking. He or she has a mental map, but there is a gap, or anomaly which needs to be filled by knowledge to be discovered in documents or other knowledge resources. Often the patron cannot formulate such an inchoate need, but a mixed-initiative interaction system, much like the reference librarian in a human context, can help the patron clarify his, or her, need. Our project can also build on ideas from these and other biomedical Watson projects. 

A Watson cognitive computing project relies on the data which is fed into Watson. Other biomedical research projects feed biomedical publications, medical records, and other biomedical knowledge into Watson. Our project will ingest similar data, but we will also take advantage of human curation applied to such data by biomedical librarians and other information specialists, e.g., references sources such as UptoDate. Instead of treating all journal articles equal as sources of biomedical knowledge, more weight will be given to sources such as meta reviews or systematic reviews. Furthermore, we will develop a collaborative user modeling frame work for information retrieval, or question answering, which will treat the queries of collaborating team scientists as additional data sources for Watson. Although our approach can support biomedical science teams more generally, we will consider the particular use case of a team of specialists working together to solve a particular difficult problem in cancer research.