AI's terrible twos: When AI does what we taught it

Thinking through some of the recent challenges she’s seen with ethics and AI, Jana Eggers was reminded of the “terrible twos” in infant development—when your toddler clings close to your side but also runs off on independent streaks. The AI works well, even surprising you with “maturity”…until it doesn’t. And as with toddlers, the challenges you encounter vary, from relatively easy challenges, like teaching a chatbot stop words, to medium-hard challenges, like improving efforts to tackle bias, where you have known actions and tools, to difficult challenges, like trying to make clear abstract concepts like curse words or death.

Jana covers wide-ranging and important themes to think about as you raise your AI systems. You’ll also learn how to detect possible “cute” behavior early that develops into bad behavior later.