Cognitive

Part 1 · Foundations

AI, Machine Learning, and Deep Learning

AI is the big goal. Machine learning is the main way we reach it, deep learning is a powerful kind of machine learning, and generative AI is what deep learning made possible.

3 min readBuilds on Defining Intelligence

Machine learning infers a rule from labeled examples instead of being programmed with one.

Key concepts

  • Symbolic AI uses hand-written rules. It is clear and exact, but brittle in messy situations.
  • Machine learning learns rules from examples.
  • Deep learning uses many-layered neural networks and excels with images, sound, and language.
  • Generative AI creates new content rather than only classifying existing content.
  • Symbolic AI and machine learning are siblings, and modern systems often combine them.

How the main approaches relate. Each step narrows the one before it, with symbolic AI as a separate branch.

Figure 2.1. How the main approaches relate. Each step narrows the one before it, with symbolic AI as a separate branch.
Deep learning stacks many layers of simple units.

Formal view and evidence

  • Established. The field was named in the 1955 proposal for the 1956 Dartmouth workshop. Symbolic and connectionist traditions have alternated in prominence.
  • Debated. Sutton's "bitter lesson" argues general methods that scale with computation beat encoded human knowledge. Neuro-symbolic researchers argue structure remains necessary for reliable reasoning.
  • Open problem. How best to combine learned representations with verifiable symbolic reasoning.

Common misconception

Common belief

"AI and machine learning mean the same thing."

In practice

A hand-coded 1985 chess program is AI, but it did not learn from data.

Key takeaways

  • AI is the broad field, and machine learning is one approach within it.
  • Deep learning is machine learning with many-layered neural networks.
  • Modern systems often combine learned models with explicit rules.

Explore next