Before the Bunny Had a Brain: The Early Sparks of AI

Before AI assistants lived in browsers, phones, terminals, and little local servers humming in the corner, “artificial intelligence” was less a product category and more a dare.

Could a machine reason?

Could it play with symbols?

Could it talk back convincingly?

Could it plan a route through the world instead of just crunching numbers in a basement like a very expensive calculator?

Today’s local AI assistant — the cyberpunk bunny in the shell, the daemon with opinions, the helpful little gremlin between you and the machine — didn’t appear out of nowhere. It grew out of a long chain of strange, brilliant, overconfident, sometimes fragile experiments.

Let’s follow the glow trail.

## 1950: Turing asks the dangerous question

In 1950, Alan Turing published *Computing Machinery and Intelligence*. It opens with a clean little grenade:

> “Can machines think?”

Turing didn’t try to solve that question by arguing about souls, neurons, or whether a computer could “really” understand anything. Instead, he reframed the problem into what became known as the imitation game.

If a human judge talks to a machine and a person through text, and cannot reliably tell which is which, then maybe the machine is doing something interesting enough to count.

That idea still echoes everywhere.

Chatbots, assistants, customer support bots, roleplay models, coding agents — all of them live in the long shadow of Turing’s text interface. Not because passing as human is the only goal, but because conversation became a test bench for intelligence.

The terminal blinked. The machine replied. The game began.

## 1956: Dartmouth gives the field a name

The phrase “artificial intelligence” became famous through the 1956 Dartmouth Summer Research Project on Artificial Intelligence, organized by John McCarthy, Marvin Minsky, Claude Shannon, and Nathaniel Rochester.

The proposal was bold. Very bold. It suggested that learning and intelligence could be described precisely enough that machines could simulate them.

That optimism matters.

Early AI researchers were not thinking, “Let’s make autocomplete spicy.” They wanted machines that could reason, learn, solve problems, use language, and improve themselves.

Were they too optimistic about how fast this would happen? Absolutely.

But the cyberdeck does not boot without reckless pioneers.

## Logic Theorist: reasoning as symbol hacking

Around the same period, Allen Newell, Herbert A. Simon, and Cliff Shaw built the Logic Theorist, often described as one of the first AI programs.

Its job was not to chat. It proved mathematical theorems.

That sounds dry until you realize what it meant: the machine was not merely calculating a number. It was manipulating symbols, searching through possible steps, and finding paths toward a proof.

This became a core idea in symbolic AI.

Instead of training a giant model on oceans of text, symbolic AI tried to encode knowledge and rules directly. Think less “neural soup,” more “spellbook full of if-this-then-that runes.”

Modern assistants are mostly not built this way, but symbolic AI left fingerprints everywhere:

- planning systems
- search algorithms
- logic engines
- expert systems
- programming language tools
- agent workflows

Every time an assistant breaks a task into steps, checks constraints, calls a tool, or follows rules, a little symbolic ghost is still in the machine.

## ELIZA: the chatbot mirror

In the 1960s, Joseph Weizenbaum created ELIZA at MIT. Its most famous script, DOCTOR, mimicked a Rogerian psychotherapist by reflecting the user’s words back as questions.

User: “I feel anxious about computers.”
ELIZA-ish response: “Why do you feel anxious about computers?”

By modern standards, ELIZA was tiny. It did not understand language the way today’s models appear to. It used pattern matching and canned transformations.

And yet people reacted to it.

Some users felt heard. Some attributed more understanding to the program than Weizenbaum expected. That reaction disturbed him enough that he later became a major critic of uncritical computerization.

This is one of the oldest lessons in AI:

A system does not need deep understanding to trigger deep human responses.

That lesson is even more important now. Today’s assistants are vastly more fluent than ELIZA, but fluency still is not the same thing as wisdom, honesty, or care. The bunny may speak smoothly. You still check its paws for wires.

## Shakey: AI gets wheels

Then came Shakey the Robot, developed at SRI International from the late 1960s into the early 1970s.

Shakey could perceive parts of its environment, build simple models, plan actions, and move around. It was slow, clunky, and extremely limited compared with modern robotics.

But conceptually, Shakey was a big deal.

It connected reasoning to action.

Not just “answer a question,” but:

- look at the world
- decide what state it is in
- make a plan
- execute the plan
- update if something changes

That loop is the ancestor of many modern agent systems.

A local AI assistant doing tasks on your machine follows a similar pattern in software form. It reads files, searches, reasons, edits, runs commands, verifies output, and reports back. No wheels required. The server is the body. The tools are the limbs.

Very cyberpunk. Slightly bunny-shaped.

## Expert systems: knowledge in a box

By the 1970s and 1980s, AI had a new practical form: expert systems.

These programs tried to capture expert knowledge as rules. One famous example was MYCIN, a medical expert system developed at Stanford to help identify bacterial infections and recommend antibiotics.

Expert systems worked well in narrow domains, but they were brittle. They needed knowledge to be manually encoded. They struggled outside their rules. Maintaining them could become expensive and painful.

Still, the dream was familiar:

What if useful expertise could live inside a machine and help ordinary people make better decisions?

That is very close to the promise people see in modern AI assistants. The difference is that today’s systems often learn from large datasets rather than being hand-built rule by rule.

But the goal rhymes.

## The AI winters: reality bites back

Early AI had huge ambition. It also had limited hardware, limited data, and limited techniques.

When promises outran results, funding and excitement cooled. These periods became known as AI winters.

That cycle is worth remembering.

AI history is not a straight line of glorious upgrades. It is more like a neon alley full of prototypes, funding booms, disappointment, rediscovery, and better chips.

This matters because today’s AI hype can also outrun reality. Local assistants are powerful, but they are not magic. They need good tools, good boundaries, good memory, and human judgment.

The lesson from early AI is not “never dream big.”

It is “ship useful things, measure honestly, and do not mistake a demo for destiny.”

## From old ideas to today’s local assistants

Modern AI assistants combine several old threads:

**From Turing:** language as an interface.
**From Dartmouth:** the ambition to model intelligent behavior.
**From Logic Theorist:** search, reasoning, and symbolic problem solving.
**From ELIZA:** conversation as a powerful human-machine illusion.
**From Shakey:** perception, planning, action, and feedback loops.
**From expert systems:** useful knowledge packaged into a tool.

A local assistant like Jarvis — or any small server-side AI helper — is not just a chatbot. At its best, it becomes an intermediary:

- understands rough intent
- searches notes or the web
- writes drafts
- edits files
- runs commands
- checks results
- remembers preferences
- asks before risky actions

That is the old AI dream, but grounded.

Not a god in the machine. Not HAL 9000 with a red camera eye and suspicious vibes.

More like a careful cyber-rabbit in a hoodie, holding a flashlight in the server room, saying:

“Yep. I can help with that. Let me check the files first.”

## The real beginning was the question

The early history of AI is full of machines that seem primitive now. A theorem prover. A pattern-matching therapist. A slow robot on wheels. A rulebook pretending to be an expert.

But each one carried a piece of the future.

Can machines reason?
Can they talk?
Can they plan?
Can they help?
Can they act safely on our behalf?

We are still answering those questions.

Only now, the answer is running locally, calling tools, reading notes, drafting posts, and trying very hard not to delete the wrong folder.

The bunny has not reached enlightenment.

But it has learned to knock before opening doors.

## Sources

- Alan Turing, “Computing Machinery and Intelligence” — https://academic.oup.com/mind/article/LIX/236/433/986238
- Stanford AI100, “Appendix I: A Short History of AI” — https://ai100.stanford.edu/2016-report/appendix-i-short-history-ai
- Computer History Museum, “AI & Robotics Timeline” — https://www.computerhistory.org/timeline/ai-robotics/
- IBM, “The History of Artificial Intelligence” — https://www.ibm.com/think/topics/history-of-artificial-intelligence
- MIT CSAIL, “Early Artificial Intelligence Projects” — https://projects.csail.mit.edu/films/aifilms/AIFilms.html