Posts tagged Conceptnet
How to make a racist AI without really trying - ConceptNet Blog

Rob Speer

Perhaps you heard about Tay, Microsoft’s experimental Twitter chat-bot, and how within a day it became so offensive that Microsoft had to shut it down and never speak of it again. And you assumed that you would never make such a thing, because you’re not doing anything weird like letting random jerks on Twitter re-train your AI on the fly.

My purpose with this tutorial is to show that you can follow an extremely typical NLP pipeline, using popular data and popular techniques, and end up with a racist classifier that should never be deployed.

There are ways to fix it. Making a non-racist classifier is only a little bit harder than making a racist classifier. The fixed version can even be more accurate at evaluations. But to get there, you have to know about the problem, and you have to be willing to not just use the first thing that works.

Read More
Better, Less-Stereotyped Word Vectors -Conceptnet Blog

Bias and Disenfranchisement Conversational interfaces learn from the data they have been given, and all datasets based on human communication encode bias. In 2013, researchers at Boston University and Microsoft discovered what they characterized as “extremely sexist” patterns in “Word2Vec,” a commonly used set of data based upon three million Google News stories.18 They found, among other things, that occupations inferred to be “male” included Maestro, Skipper, Protégé and Philosopher, while those inferred to be female included Homemaker, Nurse, Receptionist and Librarian. This is more than a hypothetical risk for organizations; Word2Vec is used to train search algorithms, recommendation engines, and other common applications related to ad targeting or audience segmentation. Organizations building chatbots based on common data sets must investigate potential bias and design for it upfront to prevent alienating and disenfranchising customers and consumers. The good news is that these and other researchers are working on methods to audit predictive models for bias.

Read More