The Sentences Computers Can't Understand, But Humans Can

Summary by Krater

Tom Scott explains the Winograd schema, a test for natural language processing that evaluates whether a machine can determine pronoun referents based on context and commonsense reasoning, and discusses why AI still struggles with it.

From the video

Answers: What is the Winograd schema and why is it difficult for AI and natural language processing?

• Winograd schema
• Natural language processing
• Referents and pronouns
• Commonsense reasoning in AI
• Machine learning text generation

What it concludes

• Resolving ambiguous pronouns in sentences requires real-world experience and commonsense reasoning about objects and their physical properties.
• Current natural language processing systems and AI models struggle with context because they rely on word collections rather than broad knowledge and understanding.
• Artificial language processing remains roughly ten years away from achieving human-like comprehension.