When it comes to AI, whether it’s Netflix’s “recommendations”, Amazon’s “other customers purchased”, the first few letters of a Google search, or self-driving cars; it is probably the most chameleon-like technology our generation has seen outside of the trojan.
Doing research with a client recently, I was alarmed to discover that error rates for image labeling, once measured at ~30% in 2010, have dropped to an error accuracy rate of less than 2.5% in 2018.
While impressive, I think we can all agree however that when AI gets it wrong, it’s really wrong. We’ve all had an interaction with a chatbot that is borderline comical, we’ve even gawked at a suggestion from an Amazon flash deal, right? We laugh it off and go about our day with little consideration about what’s behind the curtain.
What we can do
- Dig deep to understand the root cause
- Clarify the problem statement
- Gain consensus
- Provide clarity and simplify the remedy