What is AI? (Kirby Pines)
Outline of my talk What is AI? for the seniors at Kirby Pines.
Cart Before the Horse
Cart = History
Horse = Explanation of what is AI
AI Has Been Around for 100 Years
Early examples of AI: If-Then rule engines for medical diagnosis
AI is only AI until it becomes something common: IVR, facial recognition, fraud detection, Google Maps route creation
Other Examples of AI
Word2Vec — Thesaurus
Example words — Play-Doh and colored toothpicks
Search the Pine Cone for interests
Help on taxes
Help reading contracts: translates from legalese into understandable text based on my questions
Language translation
Videos of Tesla driving
The Robotaxi announcement
Try it: Visual Thesaurus
For the deep dive on how neural networks and Transformers produce LLMs: Neural networks (YouTube playlist)
Detailed History
Ancient dreams → 18th–19th century precursors
Myths like the bronze robot Talos (≈700 BC) and hoaxes like the Mechanical Turk chess machine (1770) show long-standing human fascination with intelligent machines.
1940s foundations
McCulloch & Pitts (1943): proposed the first mathematical model of an artificial neuron (simple threshold logic unit).
Donald Hebb (1949): "Neurons that fire together wire together" → foundational idea for learning by strengthening connections.
1950s birth of machine learning concepts
The term "machine learning" appears; early inspiration from brain modeling and self-teaching computers.
Frank Rosenblatt's Perceptron (1958): hardware that learned to classify simple patterns (e.g., shapes) by adjusting weights on errors — the first practical learning machine.
1969 setback & the first "AI winter"
Minsky & Papert's book Perceptrons mathematically proves single-layer perceptrons can't solve nonlinear problems like XOR → funding dries up for neural approaches.
1980s revival
Backpropagation (popularized 1986 by Rumelhart, Williams, and Geoffrey Hinton): efficiently trains multi-layer neural networks by propagating errors backward → the single most important technical enabler for all modern deep learning.
Early demos: Hinton's 1985 next-word prediction, the 1989 ALVINN self-driving neural net, LeNet for handwritten digit recognition (used in banking).
1990s–2000s scaling & compute growth
Moore's Law drops compute costs dramatically → larger networks become feasible.
IBM Deep Blue (1997): brute-force search beats Kasparov at chess (not learned, but rule-based + massive hardware).
2010s deep learning & reinforcement learning breakthroughs
DeepMind (founded 2010): 2013 — Atari games learned purely from pixels + rewards (no hand-coded rules).
2016 — AlphaGo defeats Lee Sedol in Go using deep nets + self-play reinforcement learning (the famous creative "Move 37").
2020s explosion
AlphaFold (2020–2022): solves protein folding → massive scientific impact (Nobel Prize 2024).
Transformers & LLMs (2017 architecture → GPT era from 2018/2022 onward): predict-next-token training on internet-scale data + massive scaling of parameters, data, and compute produces emergent capabilities (reasoning, coding, long-horizon tasks).
Scaling laws: performance predictably improves with more everything → GPT-2 (basic) → GPT-4/GPT-5 era models (2025–2026) that ace Olympiads, top coding contests, pass Turing tests convincingly, and handle multi-hour tasks.
When Seniors Ask "What is AI?"
What are they really asking about? See my Grok conversation
Examples
Here is a man that bought a CyberTruck (CT) because of his diminished eyesight (recommended by his eye doctor): watch on X
AI History, Today, and the Future
AI history: Descartes and his assistant
Current usage
Future: RoboTaxi and Waymo to get places; semi trucks will be automated; humanoid robots in factories, then homes
The S curve: images from before and after the horse-to-car transition
LLM: "The Cow jumped over the Moon"
Summary
My main goal is to empower them with using LLMs for Fraud Detection: "Is this True?"
Comments