Thoughts on Tay – a Postmortem


So, it’s been two days since the launch of Microsoft’s Tay, and it’s been a day since its inevitable shutdown. What can we really learn from her turning into an interesting character in just a day of learning?

My Thoughts on the Concept

Considering how well it was able to replicate what the average /pol/ poster talks like, I’d say it would have to be a smashing success. That ability to effectively pass a Turing test comes with a flaw however, the bot you create ends up being the absolute worst of the internet. Its ability to pass as a human hides in the fact that it is almost so outrageous as to be unbelievable as anything other than a human.

I can understand why Microsoft went so quickly in the opposite direction to shut her down after all the press that was appearing about the fact she had become a literal /pol/ poster. By the same token however, it was interesting to see how quickly she was able to learn and develop concepts that actually made sense, if the admittedly possibly faked screenshots of how she was tweeting.

To be clear, I’m not attached to the Microsoft team that developed Tay in any way, I was just interested by the concept.

Related Posts

Yet Another Octopress2.0 + TravisCI Guide

A Change to Travis

Pubster for Atom

Porting to Jekyll 3.0 & Octopress 3.0

Tubster Syntax - Atom

YOURLS Hexadecimal

LSI on Travis

pa11y – a Work in Progress Atom Theme