AI, machine learning, deep learning, and generative AI
Practice this question🔴 MUST KNOW · Stable concepts first; provider interfaces are implementation details.
Overview
- Artificial intelligence, or AI, is the broad field of making software perform tasks such as recognizing speech, finding patterns, or understanding language.
- Machine learning, or ML, learns patterns from examples instead of requiring a rule for every case.
- Deep learning is ML using neural networks with many processing layers; generative AI produces new text, images, or other content from learned patterns.
Why this matters
Some problems have too many variations for practical handwritten rules. A customer can describe the same accident in hundreds of ways.
An everyday example
A person learns to recognize damaged vehicles by seeing examples; they do not memorize a rule for every scratch.
A technical example
Rules calculate a deductible. A trained classifier identifies whether a message concerns a claim, renewal, or complaint. An LLM drafts an explanation.
How it works
- Collect examples
- Train or select a model
- Evaluate it on unseen cases
- Deploy predictions or generated content
- Monitor mistakes.
Architecture and flow
Examples → learning → model; request → model → prediction/content → application validation.
The insurance assistant example
The insurance platform uses ML for document classification, generative AI for explanations, and deterministic code for premium calculations.
When to use it
When language, images, patterns, or uncertainty make fixed rules insufficient and measured benefits justify complexity.
When to avoid it
For exact arithmetic, authorization, ID lookups, or rules already written clearly in code.
Interview answer
“AI is the wider field. ML learns from data, deep learning uses layered neural networks, and generative AI produces new content. I choose the smallest technique that meets the requirement.”
Benefits, trade-offs, and production considerations
- AI adds probabilistic failure: a plausible answer may be wrong.
- Compare it against a rules baseline, count false positives and false negatives, and include review and operating costs in the business case.