Intelligence Archive
From research laboratories to the AI systems shaping information today.
1956
Artificial intelligence did not begin with ChatGPT.
In 1956, researchers gathered at Dartmouth College for the Dartmouth Summer Research Project on Artificial Intelligence. Organized under the leadership of John McCarthy, the workshop helped establish AI as a distinct field of research and introduced the term “artificial intelligence.”
Dartmouth — Where AI Was Born 1955 Dartmouth proposal (McCarthy et al.)
1960s–1970s
Early AI research was dominated by what is now often called symbolic AI.
Instead of learning from enormous datasets, systems operated largely through explicitly represented symbols, rules, logic, search and human-defined knowledge.
Think:
IF condition X AND condition Y THEN conclusion Z
Researchers developed systems for theorem proving, planning, problem solving, language experiments and early robotics. The dream was enormous. The computers were not.
Tier I — Rule-based intelligence
1970s–1980s
AI repeatedly encountered a problem that will sound surprisingly familiar today:
expectations exceeded capabilities.
Research promises were ambitious, computing resources were limited, and many systems struggled outside constrained environments.
Funding and enthusiasm declined during periods now commonly known as the AI winters.
This is worth including because the history of AI isn’t:
It contains repeated cycles of: breakthrough → hype → limitations → reduced investment → breakthrough.
1980s
This is where a 1980 starting point becomes particularly relevant.
Expert systems became one of the dominant commercial AI approaches.
Instead of attempting general intelligence, developers encoded specialist knowledge into systems containing rules and knowledge bases.
An expert system might attempt to reproduce parts of the reasoning process of:
Tier I — Rule-based intelligence
1990s
The center of gravity increasingly shifted.
Instead of telling machines every rule explicitly, researchers increasingly built systems capable of deriving patterns from data.
This is the conceptual shift from:
That distinction eventually becomes fundamental to modern AI.
Toward Tier III — Learned intelligence
1997
One of the most famous AI milestones arrived in May 1997.
IBM’s Deep Blue defeated reigning world chess champion Garry Kasparov in a six-game match under standard tournament controls. IBM says the system could evaluate roughly 200 million chess positions per second.
It was extraordinary — but Deep Blue wasn’t ChatGPT hiding in a refrigerator.
It was highly specialized.
Tier II — Specialized machine intelligence
2000s
Then three things began converging:
More data. More computing power. Better machine-learning techniques.
The internet produced enormous datasets.
GPUs provided increasingly useful computational power for machine learning.
Storage became cheaper. Cloud computing expanded.
Machine learning started moving from laboratories into everyday technology.
AI was already becoming embedded everywhere — long before most people started calling everything “AI.”
Tier III — Learned intelligence
2010s
Neural networks were not new, but increasing computing power, larger datasets and technical advances dramatically improved their usefulness.
Deep learning began producing major advances in:
Tier III — Learned intelligence
2016
Another symbolic moment came when Google DeepMind’s AlphaGo defeated Lee Sedol in Go.
This mattered because Go’s enormous search space made traditional brute-force approaches much less practical than chess.
AlphaGo combined neural networks with search and reinforcement-learning techniques.
The significance wasn’t merely: computer wins board game.
It demonstrated increasingly sophisticated learned decision-making.
2017
This deserves a giant marker in our timeline.
The transformer architecture fundamentally changed modern language AI.
Transformers made it possible to model relationships across sequences extremely effectively and eventually became the foundation for the modern large language model ecosystem.
From here we get the technological lineage leading toward systems such as GPT, Gemini, Claude, Llama and many others.
One model can begin performing many tasks, rather than being built for only one narrow function. That’s a massive conceptual transition.
Tier IV — Foundation models
2022
OpenAI released ChatGPT publicly in late 2022.
And something changed culturally.
AI stopped being something most people encountered primarily behind the scenes.
People could talk to it directly.
The interface to AI became natural language. That dramatically reduced the barrier between ordinary users and sophisticated AI systems.
Tier V — Generative AI
2023
The ecosystem accelerated.
Google launched Bard and expanded it internationally during 2023.
Anthropic expanded Claude.
Microsoft integrated generative AI into products and search experiences.
Meta released Llama-family models.
Perplexity pushed the idea of an AI-native answer/search engine.
xAI introduced Grok.
AI was no longer one interesting chatbot. It was becoming an ecosystem.
And that’s particularly important to AIB.
Google: introducing Bard Anthropic: Claude 2 Meta: Llama 2 xAI: Announcing Grok
2023–2024
Models increasingly stopped being text-only systems.
They began working across combinations of:
Google introduced Gemini in December 2023 as a multimodal model family spanning Ultra, Pro and Nano, and renamed the Bard consumer experience to Gemini in February 2024.
Tier V — Multimodal intelligence
Google: Introducing Gemini Google: Bard becomes Gemini (Feb 2024)
2024–2025
This is very important for AIB.
AI systems increasingly began combining generative models with live information retrieval.
Instead of relying only on knowledge represented within model parameters, systems could:
This is where traditional SEO starts colliding with GEO, AEO and AI visibility. The question is no longer merely where a website ranks. It becomes: which information does the AI retrieve, trust and use when constructing its answer?
Tier VI — Grounded AI
2025–2026
Google itself described Gemini 2.0 as a model for the “agentic era,” highlighting capabilities such as tool use and agent-like experiences.
Instead of only Prompt → Answer, AI increasingly operates like:
This is the world AIB is being built for. Because now an AI may not merely tell someone about a company. It may increasingly participate in researching, comparing, filtering and selecting options on that person’s behalf.
Tier VII — Agentic intelligence
An AIB educational map of capability — not a ranking of vendors.
| Tier | Era | Intelligence model | Example |
|---|---|---|---|
| Tier I | 1950s–80s | Symbolic / rule-based | Expert systems |
| Tier II | 1990s | Specialized computational AI | Deep Blue |
| Tier III | 2000s–10s | Machine learning | Recommendation and prediction systems |
| Tier IV | 2017+ | Foundation models | Transformer-based models |
| Tier V | 2022+ | Generative and multimodal AI | ChatGPT, Gemini, Claude |
| Tier VI | 2024+ | Grounded AI | Generative models plus live retrieval |
| Tier VII | 2025+ | Agentic AI | Models plus tools plus actions |
AIB Classification: These tiers are an educational framework created by AIB to explain changes in AI capabilities. They are not an industry-standard classification and eras overlap.
Citations are ordinary editorial links to primary sources. They are not partnerships, sponsorships, or endorsements.
For most of AI’s history, businesses did not need to worry about how an AI system represented their company.
Deep Blue wasn’t recommending an accountant.
An image classifier wasn’t comparing cybersecurity companies.
A spam filter wasn’t advising someone which hotel to book.
Modern AI systems are different. They increasingly sit between a question and a decision.
An AI agent may eventually perform portions of that research automatically.
That creates a new organizational question:
What information will the machine find about you when it investigates?
That is the environment AIB investigates.
History explains the ecosystem. An investigation shows where you sit inside it.
