Best listening experience is on Chrome, Firefox or Safari. Subscribe to Fed Tech Talk’s audio interviews on Apple Podcasts or PodcastOne.
By now, most people involved in the world of information technology have read articles about artificial intelligence (AI) or have even attempted to deploy solutions based on machine learning and AI.
In fact, it has gotten to the maturation level that people are starting to complain about weaknesses. For example, some will label AI with three words: Brittle, opaque, and shallow. There are stories going around that AI is reflecting gender bias.
Becky Fair is the CEO of Thresher and she joined host John Gilroy on this week’s Federal Tech Talk to address many of these issues.
For example, when it comes to memes and neologisms, and other verbal variations it is tough to get AI up to speed with idiosyncratic expressions that change rapidly. What if you were charged with understanding fluctuating memes in a wide variety of languages?
Fair has had experience in a wide area of developing data for machine learning. Her intelligence background has given her the ability to understand the whole package of understanding data in a critical situation.
During the interview, Fair talked about the importance of training data before you start relying on AI as a critical component of your data infrastructure.
From her experience, she has learned that the key to making AI productive is to have labels assigned in a manner that will allow deeper understanding of the system. This concept will resonate with veteran software developers – back in the day, the phrase was “garbage in, garbage out.”