On the Limits of AI…

By Deane Barker
AI Summary

This post explores the significance of understanding the Winograd Schema Challenge in artificial intelligence. The author examines the implications of this challenge for machine understanding and emphasizes the need for nuanced comprehension in AI systems, highlighting ongoing debates in the field.

Original Location
This content was originally posted to LinkedIn. Link to Original Content

Optimizely is doing a lot of research around AI, so I’ve been recently introduced to a “Winograd Challenge.” This is a phrase that’s very hard for AI to understand due to ambiguity – some of these are invented purposely to test or tune AI.

Example: “The trophy doesn’t fit into the brown suitcase because it’s too small.”

Grammatically, you’d think the trophy was too small. But a human understands the context of something going “into” a suitcase, and it wouldn’t make sense for the trophy to be “too small,” so we reverse it in our heads and understand that it’s actually the suitcase that’s too small.

Another example: “I was trying to open the lock with the key, but someone had filled the keyhole with chewing gum, and I couldn’t get it out.”

As humans, the only thing that makes sense to “get out” is the key, not the keyhole, regardless of how grammar dictates.

Both of these examples are from “The Creativity Code,” which I just finished. It’s an interesting book about the limits of AI, and how it may or may not emulate creativity in humans.