Building machines that can read and write

Description

Individuals have access to more information than all previous generations combined, but more information doesn’t always result in more knowledge. Finance, media, legal, research, and intelligence analysts across every vertical spend countless hours collecting, deduplicating, reading, and summarizing thousands of documents to find hidden insights. Due to human attention and memory limitations, analysts often reduce the corpus of data they analyze or skim past important pieces. These human limitations lead analysts to miss the context that could mean closing a legal case faster, making money in a new investment, or saving a life on the front line.

After seeing these problems firsthand, Primer has deployed breakthroughs in AI and natural language processing to create machines that can read and write, augmenting human analysts. Sean Gourley dives into the technical aspects of building these machines, including multidocument abstractive summarization, multidocument event detection, classification, and unsupervised entity extraction as well as advanced fundamental lower-level technologies like novel information detection and Chinese character segmentation.