Hand-coding world knowledge in formal rules failed, motivating machine learning from data

Early AI tried the knowledge-base approach: hard-coding facts about the world in formal languages so a computer could reason over them with logical inference rules (e.g. the Cyc project). None of these projects succeeded, because people struggle to devise formal rules with enough complexity to accurately describe the messy real world. This difficulty is what motivated machine learning: rather than being told everything, an AI system should acquire its own knowledge by extracting patterns from raw data. Understanding this failure is the key motivation for the whole learning-from-data paradigm — it explains why modern AI learns rather than being programmed with facts.

Examples

Cyc, a large hand-entered database of formal statements, famously failed on a simple story about a person shaving — the rules could not capture enough real-world nuance. Machine learning sidesteps this by learning patterns from examples instead of enumerating rules.

Assessment

Explain why the hand-coded knowledge-base approach to AI failed, and describe how machine learning’s learn-from-data strategy addresses that specific failure.