Since the early version of IBM Watson in 2011 won over the world champion in Jeopardy!, it is clear that a computer system can answer a sort of question better than humans: all thanks to the so called cognitive computing.
Since 2011, IBM researchers and developers have been continuously improving Watson and expanding its capabilities. In parallel, other IT giants established their cognitive systems and platforms (e.g., Microsoft Cognitive Services). Currently, many institutions are piloting new ways they can use Watson in their daily business. The spectrum of application ranges from chatbot or call-center support to highly specialized medicine and pharmacy applications: research institutions are exploring ways to utilize Watson for better handling of archive data, whereas physicians are offered support at the interpretation of patients' X-ray images. Potential benefits from using cognitive computing include significantly broader analysis and efficiency. However, Watson is not a plug-and-play machine: everything from project selection and goals specification up to execution and roll-out needs to be done very carefully. Only through actually launching a project, one can explore whether the identified goal was realistic given the data, the company structure, the resources, and the abilities of Watson. The same holds for the projects, IMRG is currently approaching.
Our long-term interest is to improve the co-located interpersonal collaboration by introducing a speech-based interaction with a computer and with databases into a collaborative setting, such as the career advisory service or doctor-patient encounters. Optimally, seamless conversational practices emerge over time, in which Watson (or an alternative) takes on the role of a secretary, transcript writer, desk researcher, assistant, creative insider, or, even, devil's advocate. Which of those roles are acceptable in the envisioned scenario is also part of the research endeavor: Do advisors accept an assistant that can potentially counter their statements? Do advisees trust an automatically generated protocol? How does talking to a machine feels like, when conducted in an advisory setting (as opposite to single-user scenario)? Many questions await deeper investigations - and finding answers to them is of great importance for the future of practical application of cognitive computing.
In parallel, we explore how companies set up and execute Watson projects. In particular, we seek for successful patterns and practices so that Watson's adoption in business can be easier. We hypothesize that the Watson projects, despite their strong research (and, consequently case-specific) nature, share features which make them likely to fail or succeed. This issue has been an object of a multi-case study and observations of our own efforts.