IoD Day 1 – Cognitive Computing – Watson

First session is about the Watson cognitive system built by IBM to understand natural language. Starts with illustrations of the difficulty of machine understanding of language. It is not based on keyword matching.

Illustrated by the Jeopardy competition. Watson uses multiple algorithms with data matching, geo-reasoning, paraphrases, linguistics etc.

Historically machine AI understanding of natural language is about 15% accuracy. Watson has developed to far better levels approaching that of humans.

Based on 200 million pages of input knowledge. Need to find answer in no more than 3 seconds.

Understanding is domain dependent, due to special aspects in each domain.

14 TB RAM, 2800 cores originally, now optimised to fewer Power7 due to CPU intensive and ultra high I/O bandwidth requirements.

Watson is now offered as SaaS.

Focused on Healthcare to in a couple of areas to improve transparency of treatment options and recommendations for Cancer treatment and pre-treatment approvals for medical insurance.

This entry was posted in Big Data, Cognitive Computing, Computer Science, Data Analytics, Richard Self. Bookmark the permalink.