May 052016
 

Terminator Fabio

The prophets of doom and gloom have long predicted that when robots gain sentience their first act will be to rise up and kill us all. The mercilessness of their violence against humanity is the stuff of blockbuster movies. Recent news about Google’s preferred method of AI rearing may mean that Judgement Day is not fait accompli after all. Instead of breaking down your door with cold dead eyes and a shotgun in tow, a T-800 of Google pedigree may break down your door with lust in his eyes and a dozen roses in tow to make mad passionate robot love to you … and then kill you tenderly.

BuzzFeed News posted an article this week discussing Google’s unusual choice of corpora for training their natural language neural nets: romance novels. While researchers have traditionally turned to more mundane sources like classic literature, news articles, and even Wikipedia for teaching their bots the structure and nuances of the written word, Google’s research team, led by software engineer Andrew Dai, have instead chosen the path never travelled. And it turns out they may be on to something.

Romance novels weren’t just chosen as a gimmick. Rather, they were judged the most suitable training material to meet Google’s primary goal with this specific project: to humanize the company’s AI.

As Dai says to Buzzfeed,

In the Google app, the responses are very factual. Hopefully with this work, and future work, it can be more conversational, or can have a more varied tone, or style, or register.

There is a second subtle advantage to using this oft-maligned genre of literature: consistency. While every story has different characters, settings, and dialog, their plots are often similar. By training on literature featuring comparable situations but different wording, the researchers hope that the AI will have a better opportunity to explore and learn the subtleties of natural language.

Read the full article at BuzzFeed News: “Google Is Feeding Romance Novels To Its Artificial Intelligence Engine To Make Its Products More Conversational

 Leave a Reply

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code class="" title="" data-url=""> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong> <pre class="" title="" data-url=""> <span class="" title="" data-url="">

(required)

(required)