# Dartmouth and the AI Winters

> How AI began at a summer meeting in 1956. What the first learning machine could do, and why the money for AI dried up twice.

**Course:** [AI Basics](https://learn.benibix.com/learn/ai-foundations)  
**Time:** 12 min

**After this lesson you can:**

- Say when and where AI began as a field
- Explain what the perceptron could and could not do
- Describe what causes an AI winter

## How AI began, and why it froze twice

AI became a field of study at a summer meeting at Dartmouth College in 1956. The plan, written in 1955, asked for ten people and two months. It said every part of learning could, in principle, be described so exactly that a machine could copy it. The plan also gave the field its name.

The people behind it, among them John McCarthy, Marvin Minsky and Claude Shannon, led AI for decades. They did not build a working AI that summer. They mostly swapped ideas. But hopes ran high, and some said a machine would match a person within a generation.

In 1958, Frank Rosenblatt showed the perceptron, one of the first machines that learned from examples. It added up its inputs, each with its own weight, and said yes if the total was big enough. When it got an answer wrong, it moved its weights a little. The machine he built soon after saw through 400 light sensors.

A single perceptron has a hard limit. It can only split things with a straight line. In 1969, Minsky and Seymour Papert wrote a book showing simple tasks it could never learn. Interest in learning machines dropped for years, though later networks with many layers got past that limit.

An AI winter is a time when money and interest in AI drop sharply. It comes when promises run far ahead of results. The first began in the mid 1970s. In 1973, a report for the UK government said AI had not met its grand claims, and funding was cut there and in the US.

The second winter began in the late 1980s. Firms had spent heavily on expert systems and the special computers that ran them. The systems cost a lot to keep up and broke on new cases. In 1987 the market for those computers fell apart, and AI became a word to avoid.

People often think AI research stopped in a winter. It did not. Work went on with far less money, and some of the ideas that later won, like networks with many layers, grew in those cold years. The lesson of the winters is about hype: promise less than you can show.

![A timeline from 1950 to 2022, with marks for the Turing test in 1950, Dartmouth in 1956, expert systems in the 1980s, deep learning in 2012, the transformer in 2017 and ChatGPT in 2022. Two shaded bands mark the AI winters between the bursts of hope.](https://learn.benibix.com/diagrams/ai-history-timeline.svg)

### Example: Jack's bank buys an expert system

In 1985, Jack's bank buys an expert system that judges loans with 2,000 rules. Two staff are hired just to keep the rules up to date.

Each new kind of loan needs dozens of new rules, and some clash with old ones. By 1990 the bank turns it off. Thousands of firms had the same story, and that is what the second winter looked like.

**Words in this lesson:** [Dartmouth Workshop](https://learn.benibix.com/term/ai/dartmouth-workshop), [Perceptron](https://learn.benibix.com/term/ai/perceptron), [AI Winter](https://learn.benibix.com/term/ai/ai-winter)

**Question:** What causes an AI winter?

**Answer:** Promises that run far ahead of results, so funders pull out. Both winters followed big claims that the tools could not meet. Funders felt let down and stopped paying, though some research went on.

---

Source: https://learn.benibix.com/learn/ai-foundations/dartmouth-and-the-ai-winters
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
