Voice assistants. Movie recommendations. Car insurance. Facial recognition software. All of these things use artificical intelligence and machine learning in one way or another. For the most part, we don't feel the need to question why we get a particular movie recommendation on whatever streaming application we use. However, when you start moving into applications that have the potential for more serious outcomes, there becomes a need for understanding that "why."

This is where explainable AI comes in. Unlike traditional programming, AI and ML algorithms are trained using batches of data; not a set of specific instructions. So when we look to understand exactly why we're told some rom-com is a 97% match for us, it's not always clear why that's the case. With things like autonomous vehicles (which are developed primarily with AI/ML algorithms) increasingly becoming a reality, comprehending why that car stopped when it did becomes much more imperitive.

This week, Roberto Stelling and Adriana Prado, two researchers from the Office of the CTO Research Office, join host Kelly Lynch to talk through exactly what explainable AI is, why it is such an important factor in the future development of algorithms, and how, once we can fully trust it, AI has the potential to positively impact our lives.

In this episode:

• Bridging the gap between fear and trust in AI (01:19)
• Review of traditional programming vs. AI (01:45)
• Re-imaginging traditional programming vs. AI in the context of a paper shredder (02:30)
• Difficulty understanding outputs from AI and ML algorithms (04:20)
• Why it would be impossible to write a program for an autonomous vehicle (05:02)
• The best definition for explainable AI (06:36)
• A clear example of the need for explainability through the lens of autonomous cars (07:13)
• Explainable AI as Responsible AI (07:59)
• Is explainability only going to help the internals of AI algorithms? (09:07)
• How voice assistants are becoming more and more regionally aware (10:20)
• Reaching a new level of trust with AI (11:30)