Computer Science

Computer Science Colloquium

Im Rahmen des Informatik-Kolloquiums, das von den Instituten des Fachbereichs Informatik, der Österreichischen Gesellschaft für Informatik (ÖGI), der Arbeitsgemeinschaft für Datenverarbeitung (ADV) sowie der Österreichischen Computergesellschaft (OCG) abgehalten wird, spricht

John Prager

IBM T.J. Watson Research Ctr. & IBM Watson Health, Yorktown Heights, NY USA

über das Thema:

IBM Watson: Building a Question-Answering system to Beat Humans on TV

Zeit: Mon 30.5.2016, 15:45, 90 Minuten
Ort: HS 18


In February of 2011, IBM's computer system called Watson participated in the US game show Jeopardy! on national television against two former champions - and won. Jeopardy! is a questionanswering challenge, which, despite including elements of speed and betting, is at its core about finding answers to questions and having a good confidence estimate about those answers. Question-answering (QA) is a sub-field of Artificial Intelligence, and Watson is acknowledged to represent the current state-of-the-art. Watson is able to answer some obscure general-knowledge questions that average people have difficulty with, so in that very limited and technical sense is super-human. On the other hand it makes regular mistakes, usually in regard to questions about common sense or common human experience (of which it has none). This talk will examine what went into Watson, to what extent it actually understands the domain it operates in, and where the technology is headed.


John Prager has been working in technical fields related directly or indirectly to Question Answering for most of his professional career. Most recently, while at the IBM T.J. Watson Research Center he has been part of the Watson project, a system that played (and won) the Jeopardy! TV quizshow game. He has been involved in both the algorithms area, concentrating on question analysis and wordplay, and strategy. He is still involved with Watson, as it is being adapted to clinical decision support in the health-care domain. Previously, he led IBM?s successful entries in the TREC-QA