History and Evolution of AI — Beginner-Friendly Notes
1. Big Picture in One Line
AI did not appear with ChatGPT — it is ~70 years of ideas, winters, and breakthroughs built on maths, logic, and computers.
2. Timeline You Must Remember (exam-friendly)
| Period | What Happened | Why It Matters |
|---|---|---|
| Pre-1950 — Foundations | McCulloch & Pitts (1943) artificial neuron; Turing (1950) asks "Can machines think?" + Turing Test | Showed brains/computation can be modelled |
| 1956 — Birth of AI | Dartmouth Conference — John McCarthy coins "Artificial Intelligence" | AI becomes a field |
| 1956-74 — Early Optimism | Logic Theorist, General Problem Solver, ELIZA, early search & game playing | Proved AI could solve puzzles, play games |
| 1974-80 — First AI Winter | Hype > results, limited compute, no data — funding cut | Lesson: scaling needs data & compute |
| 1980-87 — Expert Systems Boom | Rule-based systems (MYCIN for diagnosis, DENDRAL) succeed in narrow domains | First commercial AI |
| 1987-93 — Second Winter | Expert systems brittle, expensive to maintain, market collapses | Lesson: hand-coded rules don't scale |
| 1993-2011 — Quiet Progress | ML rises: Bayes nets, SVMs, data mining; Deep Blue beats Kasparov (1997), web-scale data | Learning from data replaces hand-coding |
| 2012-now — Deep Learning Revolution | AlexNet (2012) wins ImageNet, GPUs + big data; AlphaGo (2016), Transformers (2017), ChatGPT (2022) | AI becomes everyday tech |
Memory trick: Foundations → Dartmouth → Optimism → Winter1 → Expert Boom → Winter2 → ML Quietly → Deep Boom
3. Three Pillars That Finally Made AI Work
- Data — internet gave billions of examples
- Compute — GPUs/TPUs made training feasible
- Algorithms — backpropagation, CNNs, Transformers, reinforcement learning
Before ~2012 we had ideas but missed the first two pillars.
4. India's Context (useful for viva)
- Early programmes at IISc, IITs (1980s-90s) in expert systems & NLP for Indian languages.
- Today: AI in agriculture (crop disease detection), health (TB screening), governance (DigiLocker, UPI fraud detection), ISRO robotics.
5. What Changed at Each Evolution Step?
- Symbolic AI (1956-90): If-then rules written by humans → good for logic, bad for noisy data.
- Statistical ML (1990-2012): Learn patterns from data → good for prediction, needs features.
- Deep Learning (2012-now): Learn features and patterns end-to-end from raw data → needs lots of data/compute.
6. AI Winters — Why They Happened
- Over-promised, under-delivered.
- Hardware too slow / data too small.
- Systems could not handle uncertainty or real-world noise.
Lesson for you: progress is cycles, not straight lines.
7. Two Giants to Name in Exam
- Alan Turing — fundamental question + test for intelligence.
- John McCarthy — named AI, invented Lisp, organised Dartmouth.
Add for bonus: Marvin Minsky, Geoffrey Hinton (deep learning), Yann LeCun, Yoshua Bengio.
8. One-Line Summary to Write in Exam
AI evolved from 1940s neural models and Turing's ideas through Dartmouth (1956), alternating booms and winters, to modern data-and-compute-driven deep learning that powers today's applications.
9. Quick Self-Check
- Who coined the term AI and when?
- Name one reason for an AI winter.
- What three factors caused the post-2012 boom?
- Pick one historical system (e.g., Deep Blue or MYCIN) and say whether it was symbolic or learning-based.