● LIVE HISTORY · AI DOCUMENTARY

The Long Road to Intelligence

From a humble floor-sweeping robot that bumped into walls, to large language models that write poetry and solve equations — this is the full, unfiltered story of Artificial Intelligence.

75+
Years of AI History
2
AI Winters Survived
GPT-4
Current Frontier
Future Potential
Chapter 01

Before "AI" Had a Name

Long before silicon chips and neural networks, humans dreamed of mechanical minds. The roots of Artificial Intelligence stretch back centuries — to clockwork automata, analytical engines, and the first flickering ideas that machines might one day think.

In 1763, Swiss watchmaker Pierre Jaquet-Droz built The Writer — a mechanical boy automaton that could actually dip a quill in ink and write out text. It had no intelligence, but it planted a seed: can a machine do what a human does?

In 1837, Charles Babbage designed the Analytical Engine — a mechanical computer with a programmable memory and processor. Ada Lovelace wrote what is considered the first computer program for it. Though never fully built, it was the conceptual grandfather of every computer ever made.

By the 1940s, mathematician Alan Turing asked the most profound question in the history of technology: "Can machines think?" His 1950 paper "Computing Machinery and Intelligence" proposed the Turing Test — still discussed today. The age of AI had begun in the mind, before it was born in the lab.

  • ⚙️ 1763 — Jaquet-Droz's "The Writer" automaton — first mechanical human behaviour
  • 🔢 1837 — Babbage's Analytical Engine — programmable mechanical computation
  • 🧠 1943 — McCulloch & Pitts model the first artificial neuron in mathematics
  • ♟️ 1950 — Alan Turing's "Computing Machinery and Intelligence" — birth of AI theory
  • 🎓 1956 — The Dartmouth Conference coins the term "Artificial Intelligence"
⚙️ ⚙️ ⚙️
1956
AI officially named
KEY PIONEERS
Alan Turing · John McCarthy
Marvin Minsky · Claude Shannon
Ada Lovelace · Charles Babbage
Chapter 02

The Roomba & the Age of
Practical Robots

Not all AI starts with grand ambitions. Sometimes it starts with a small, round disc that bumps into furniture and vacuums your floor. The iRobot Roomba, launched in 2002, brought artificial intelligence into millions of ordinary homes for the first time — and the world was never quite the same.

🤖

iRobot Roomba — 2002 : AI Comes Home

The Roomba wasn't just a vacuum cleaner. It was a behaviour-based autonomous robot — meaning it didn't follow a pre-mapped path but made real-time decisions based on sensor input. Bump into a wall? Turn. Detect a cliff edge? Reverse. Dirt detected? Slow down and clean more thoroughly.

This is reactive AI — the simplest form, yet profoundly impactful. Over 30 million Roombas have been sold globally. It proved that AI didn't need to be a supercomputer — it could be a $300 disc that made your life genuinely easier. Later models gained SLAM navigation (Simultaneous Localisation and Mapping), building live maps of rooms using laser sensors — the same technology used in self-driving cars.

SENSOR AI

Bump & React Intelligence

The first Roombas used simple infrared and contact sensors. When the robot hit something, it would rotate a random angle and continue. No map, no plan — just reactive behaviour. Surprisingly effective and the foundation of modern robot navigation.

MAPPING AI

SLAM Navigation (2015+)

Modern Roombas use Visual SLAM — cameras and lasers build a live map of your home while simultaneously tracking the robot's position. This is the same fundamental technology used in autonomous vehicles and drone navigation.

CONNECTED AI

Smart Home Integration

Today's Roombas connect to Wi-Fi, learn room names, respond to voice commands via Alexa and Google Assistant, and learn your cleaning schedule automatically. A child of the 2000s, grown up in the era of IoT and machine learning.

BEYOND ROOMBA

The Robot Explosion

Roomba opened the floodgates. Boston Dynamics' BigDog (2005), Toyota's ASIMO humanoid, warehouse robots at Amazon, surgical robots in hospitals — all descendants of the same idea: machines that sense, decide, and act in the real world.

Step by Step

The Complete AI Timeline

From the first hand-coded rules to self-teaching neural networks — follow every major leap in the 75-year journey of Artificial Intelligence.

🎓
1956 — Dartmouth, USA

Birth of Artificial Intelligence

John McCarthy, Marvin Minsky, Claude Shannon, and others gather at Dartmouth College for a landmark summer workshop. For the first time, the field is formally called "Artificial Intelligence." The optimism is electric — researchers predict human-level AI within 20 years. (They were slightly off.)

"Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." — Dartmouth Proposal, 1956
🧩
1956–1974 — The Golden Era

Logic, Symbols & Early Chatbots

The first wave of AI uses symbolic logic and hand-coded rules. Programs like ELIZA (1966) — the first chatbot — simulate a therapist by matching keywords. SHRDLU (1970) understands natural language in a limited "blocks world." The Logic Theorist proves mathematical theorems. Governments pour funding in. The future looks limitless.

ELIZA shocked users into believing they were talking to a real therapist — demonstrating how easily humans anthropomorphise machines.
🥶
1974–1980 — The First AI Winter

Promises Unfulfilled · Funding Dries Up

The gap between promises and reality becomes painfully obvious. Language translation is far harder than expected. Problem-solving doesn't scale. The Lighthill Report (1973) in the UK brutally critiques AI progress, triggering massive funding cuts by the British government. The US follows. Researchers scatter to other fields. AI enters its first long, cold silence.

The term "AI Winter" captures the bleak atmosphere — frozen budgets, abandoned projects, and a field that had over-promised and under-delivered.
🤖
1980–1987 — Expert Systems Boom

AI Returns with a Business Hat On

Expert systems — AI that encodes the knowledge of human specialists into rule-based programs — ignite a commercial revival. MYCIN diagnoses blood infections better than some doctors. XCON configures computer systems at DEC, saving $40 million a year. Japan launches the audacious Fifth Generation Computer Project. The AI industry hits $1 billion in annual revenue by 1985.

Expert systems proved AI could have real commercial value — but they were brittle, expensive to maintain, and couldn't learn from new data.
🥶
1987–1993 — The Second AI Winter

The Expert Systems Collapse

Expert systems are expensive to build, brittle to maintain, and can't adapt to changing knowledge. The market collapses. Apple and IBM personal computers outperform the specialised Lisp machines made for AI. DARPA cuts funding again. Japan's Fifth Generation project quietly winds down. A second, deeper winter settles in. The phrase "AI" becomes almost embarrassing to use in research circles.

Many researchers simply renamed their work "machine learning," "neural networks," or "pattern recognition" to avoid the toxic AI brand.
♟️
1997 — A Historic Moment

Deep Blue Defeats Garry Kasparov

IBM's Deep Blue becomes the first computer to defeat a reigning world chess champion — Garry Kasparov — in a six-game match. It evaluates 200 million positions per second. The world is stunned. But Deep Blue doesn't "understand" chess — it uses brute-force search and hand-crafted evaluation functions. It cannot learn or generalise. Still, it's a pivotal cultural moment that announces: machines can beat the best human minds at their own games.

Kasparov accused IBM of cheating — claiming a human must have intervened in game two. IBM denied it. The controversy added to the drama.
📊
1990s–2000s — Machine Learning Rises

Learning from Data, Not Rules

A quiet revolution: instead of hand-coding rules, AI researchers start training systems directly on data. Support Vector Machines, decision trees, Naive Bayes classifiers power spam filters, credit scoring, and early recommendation systems. Google's PageRank algorithm (1998) is machine learning in disguise. Amazon begins recommending products. The internet generates unprecedented amounts of training data — and AI learns to use it.

Every time you marked an email as spam in 2002, you were training an AI. Millions of users collectively taught machines to recognise junk mail.
🧠
2012 — The Deep Learning Revolution

AlexNet Rewrites Everything

A Toronto team — Geoffrey Hinton, Alex Krizhevsky, and Ilya Sutskever — enter their deep neural network AlexNet into the ImageNet competition. It wins by a margin so enormous that the AI field collectively gasps. Error rate drops from 26% to 15.3% overnight. Suddenly, deep learning is all anyone talks about. The GPU computing revolution and vast datasets make neural networks not just possible but unstoppable. The modern AI era begins right here.

Geoffrey Hinton, called the "Godfather of Deep Learning," later resigned from Google in 2023 to speak freely about AI's potential dangers — a remarkable arc.
🎮
2016–2017 — AlphaGo & Reinforcement Learning

Mastering the Hardest Human Game

DeepMind's AlphaGo defeats 18-time world champion Lee Sedol 4–1 at the ancient game of Go. Unlike chess, Go has more possible positions than atoms in the observable universe — brute force is impossible. AlphaGo learns through deep reinforcement learning — playing millions of games against itself. Its move 37 in Game 2 is declared the most creative move ever played by a Go player. Humans are awed, and a little unsettled.

AlphaGo Zero (2017) learned with zero human knowledge — starting from random moves, it surpassed all previous versions in 40 days. Self-taught superhuman intelligence.
💬
2017–2020 — The Transformer Revolution

"Attention Is All You Need"

Google researchers publish a paper with a deceptively modest title: "Attention Is All You Need" (2017). It introduces the Transformer architecture — a new way for neural networks to process language by paying "attention" to relationships between all words simultaneously. This paper becomes the most influential AI publication of the decade. BERT (2018) and GPT-2 (2019) demonstrate that language models trained on vast text data can write, summarise, and reason in startlingly human ways.

The Transformer paper has been cited over 100,000 times. It is arguably the most impactful research paper of the 21st century so far.
🌍
Nov 2022 — The World Changes

ChatGPT Reaches 100 Million Users in 60 Days

OpenAI releases ChatGPT on November 30, 2022. It becomes the fastest product in history to reach 100 million users — Instagram took 2.5 years; TikTok took 9 months. ChatGPT does it in 60 days. Suddenly, AI isn't for labs and tech companies — it's for everyone. Writers, teachers, doctors, students, farmers, programmers — all find immediate, practical uses. The race to the future is officially on.

The day ChatGPT launched, Google reportedly declared an internal "code red" and mobilised its entire AI workforce. The most consequential product launch since the iPhone.
🚀
2023–2025 — The Intelligence Explosion

GPT-4, Gemini, Claude, Llama & Beyond

GPT-4 passes the bar exam in the top 10%. Google Gemini achieves human-expert performance on 57 subjects. Anthropic's Claude focuses on safety and nuanced reasoning. Meta's Llama democratises AI with open-source models. AI generates photorealistic images (DALL-E 3, Midjourney, Stable Diffusion), composes music, writes code, discovers new proteins for medicine, and assists in scientific research at every level. The pace of progress is unlike anything in technological history.

In 2024, AlphaFold 3 predicted the structure of virtually every known protein — a problem biologists had chased for 50 years. AI is now doing real science.
Dark Chapters

The Two AI Winters

Understanding why AI failed — twice — is just as important as understanding how it succeeded. These periods of collapse and disillusionment ultimately made modern AI stronger, more grounded, and more useful.

❄️ First AI Winter (1974–1980)

The first wave of AI research collapsed under the weight of its own optimism. The promises were extraordinary; the results were disappointing.

  • Computers were far too slow for the algorithms being designed
  • Machine translation quality was embarrassingly poor despite early hype
  • The Lighthill Report (1973) declared most AI research a failure
  • Systems that worked on toy problems failed completely on real ones
  • The "combinatorial explosion" — problems grew exponentially harder at scale
  • US and UK governments slashed research funding dramatically

❄️ Second AI Winter (1987–1993)

Just as commercial AI was booming, the expert systems market collapsed — taking billions of dollars and thousands of jobs with it.

  • Expert systems cost millions to build and were expensive to update
  • They couldn't learn from new data — required manual reprogramming
  • Desktop computers outperformed expensive specialised AI hardware
  • Japan's Fifth Generation project failed to deliver its ambitious goals
  • DARPA cut the Strategic Computing Initiative budget severely
  • The word "AI" became commercially toxic — researchers rebranded work
What didn't kill AI made it unimaginably stronger. Every winter planted seeds that bloomed into a revolution.
— The paradox of AI's two dark ages
Landmark Moments

Moments That Changed Everything

Across 75 years, certain moments stand apart — events so dramatic they permanently shifted the trajectory of AI and, with it, human civilisation.

♟️

Deep Blue vs Kasparov

First AI to defeat a world chess champion in official match play. Proved computational brute force could beat human intuition in constrained domains.

1997
🎮

AlphaGo Beats Lee Sedol

First AI to defeat a 9-dan Go professional. Move 37 is called the greatest Go move ever played — spontaneously creative, not in any human training data.

2016
🖼️

AlexNet — Vision Breakthrough

Deep convolutional network halves the image recognition error rate. The moment the research world unanimously switched to deep learning.

2012
🤖

Transformer Architecture

"Attention Is All You Need" — the paper that enabled BERT, GPT, and every modern large language model. The DNA of current AI.

2017
💬

ChatGPT Goes Public

100 million users in 60 days. AI stops being a specialist tool and becomes a household name. The world's relationship with intelligence changes forever.

2022
🧬

AlphaFold Cracks Proteins

DeepMind predicts 3D structures of virtually all known proteins — solving a 50-year-old biology problem and transforming drug discovery overnight.

2020–24
🚗

Self-Driving Cars Arrive

Waymo, Tesla Autopilot, and others bring AI-powered driving to public roads. Autonomous vehicles log millions of miles. The age of robotic transportation begins.

2015–now
🎨

Generative AI Explosion

DALL-E, Midjourney, Stable Diffusion generate photorealistic images from text. Suno and Udio compose music. AI creativity challenges every assumption about art.

2022–24
Technical Deep Dive

How Deep Learning Actually Works

The engine behind modern AI is the artificial neural network — inspired by, but not identical to, the human brain. Understanding it demystifies the "magic" of contemporary AI systems.

NEURAL NETWORK — 25 NODES · 5 LAYERS

  • 1
    Input Layer
    Raw data enters — pixels from an image, words from text, sensor readings from a robot. Each neuron receives one piece of the input signal.
  • 2
    Hidden Layers
    Multiple layers of neurons each detect increasingly abstract patterns — edges → shapes → objects in vision; syllables → words → meaning in language.
  • 3
    Weights & Backpropagation
    The network learns by adjusting the "strength" of connections. Wrong answers trigger a backpropagation signal that nudges all weights slightly in the right direction.
  • 4
    Output Layer
    The final answer — "this is a cat" with 97% confidence, or the next word in a sentence, or the steering angle for an autonomous car.
  • 5
    Scale Changes Everything
    GPT-4 has an estimated 1.76 trillion parameters (adjustable weights). The human brain has ~100 trillion synapses. We're getting closer — and the gap is narrowing fast.
The Modern Era

The Large Language Model Revolution

In less than a decade, language models grew from curiosities to civilisation-altering tools. Here is the complete story of how AI learned to talk — and then to think.

2018

GPT-1 & BERT — Language Models Are Born

OpenAI releases GPT-1 (117M parameters) — proves unsupervised pre-training on text works. Google releases BERT — bidirectional understanding revolutionises search. The age of pre-trained language models begins.

117M params Transformer-based
2019

GPT-2 — "Too Dangerous to Release"

OpenAI initially refuses to release GPT-2 (1.5B parameters) citing misuse risks. It writes coherent multi-paragraph text that fools people into thinking a human wrote it. The first major public debate about AI safety and information hazards.

1.5B params Controlled release
2020

GPT-3 — The Leap That Stunned the World

175 billion parameters. GPT-3 writes essays, translates languages, debugs code, and generates poetry with minimal prompting. It demonstrates few-shot learning — the ability to learn new tasks from just a few examples. Developers build hundreds of applications on the API within months.

175B params API released publicly
2022

ChatGPT, Claude, Bard — AI for Everyone

OpenAI's ChatGPT uses RLHF (Reinforcement Learning from Human Feedback) to create a conversational AI that the public actually wants to use. Anthropic's Claude and Google's Bard follow. Suddenly every company needs an AI strategy.

100M users in 60 days RLHF training
2023–25

GPT-4, Gemini Ultra, Claude 3 — Frontier AI

GPT-4 is multimodal — it sees images and reads text. It passes the bar exam (top 10%), medical licensing, and GRE. Google Gemini Ultra achieves human-expert level on MMLU. Claude 3 and Claude 4 demonstrate extended reasoning. These are no longer narrow tools — they are approaching general-purpose reasoning systems.

Multimodal Expert-level reasoning Agentic capabilities
Where We Are Now

AI in 2024–2025: The Full Landscape

From floor-sweeping robots to systems that help design cancer drugs — here is what Artificial Intelligence is actually doing in the world right now.

💊

Medicine & Drug Discovery

AlphaFold has predicted 200 million protein structures. AI identifies cancer in scans with radiologist-level accuracy. DeepMind's AI discovered a new class of antibiotics. Drug discovery timelines shrinking from 12 years to 3.

💻

Software & Coding

GitHub Copilot writes 46% of code accepted by developers using it. AI passes Google software engineering interviews. Systems like Claude Code and Devin can execute multi-step programming tasks autonomously.

🌾

Agriculture & Environment

AI monitors crop health from satellite imagery. Precision agriculture drones apply fertiliser only where needed. Climate models powered by AI provide predictions with unprecedented accuracy. Food waste reduced by smart supply chain optimisation.

🎓

Education

AI tutors provide personalised learning at scale. Khan Academy's Khanmigo adapts to each student's pace. Language learning apps use AI conversation partners. Students in rural India access quality tutoring previously available only in cities.

🎨

Creative Arts

Midjourney, DALL-E 3, and Stable Diffusion generate professional-quality images from text. AI composes film scores, writes novels, and creates video. The boundaries of authorship and creativity are being fundamentally redefined.

🔬

Scientific Research

AI assists in nuclear fusion research at DeepMind. Mathematical proofs verified and discovered by AI. Telescope data analysed by AI finds new exoplanets. AI co-authors are appearing in peer-reviewed journals.

By the Numbers

The Scale of the AI Era

$200B+
Global AI Investment
in 2024
1.76T
Estimated GPT-4
Parameters
200M
Protein Structures
Predicted by AlphaFold
60 days
ChatGPT to
100M Users
30M+
Roombas Sold
Worldwide

AI Capability Growth — Key Benchmarks

Image Recognition Accuracy (ImageNet)99.7%
GPT-4 Bar Exam Score (percentile)90th
AI Protein Structure Accuracy95%
Code Written by GitHub Copilot (accepted)46%
Gemini Ultra — MMLU Human Expert Level90.0%
What Comes Next

The Road Ahead

From Roomba to GPT-4 in twenty years. What will the next twenty bring? Six transformations that researchers, engineers, and ethicists believe will define the coming era of intelligence.

01

Artificial General Intelligence

AGI — AI that matches human cognitive ability across all domains — is no longer science fiction. Estimates range from 5 to 30 years. Every major lab is racing toward it with different strategies and different safety philosophies.

02

Brain-Computer Interfaces

Neuralink and competitors are building direct neural interfaces. Within a decade, AI may be accessible not through keyboards or voice — but through thought itself. The boundary between human and artificial intelligence begins to blur.

03

AI in Every Device

The next Roomba will not just map your floor — it will understand your household routines, order supplies, and coordinate with your calendar. Every appliance, vehicle, and medical device will have embedded intelligence.

04

Autonomous Science

AI will design experiments, analyse results, form hypotheses, and publish papers with minimal human oversight. The pace of scientific discovery will accelerate by orders of magnitude. Diseases that took decades to understand may be solved in years.

05

AI Safety & Alignment

Ensuring AI systems remain aligned with human values is the defining challenge of our era. Anthropic, DeepMind Safety, and OpenAI alignment teams are racing to solve this before capability outpaces safety. The stakes could not be higher.

06

Global Access & Equity

The most important AI question is not "how powerful?" but "available to whom?" Bridging the AI divide — ensuring students in rural Bengal have the same AI tools as students in Silicon Valley — may be the moral imperative of the age.

The journey from a bumping robot on your kitchen floor to a system that writes poetry and discovers antibiotics took just twenty-two years. Imagine the next twenty-two.
— The Roomba to AGI arc