The Study · DoNotEnergize · AI history

AI Winter: When the Promises Outran the Results

An excellent question. An uncomfortable meeting. Join SunJewel and Patch in the Mansion workshop to ask what happens when the promises keep growing—and the results struggle to keep up.

AI Winter · The full film

6:31 · Narrated by SunJewel · DoNotEnergize

Watch the full film on YouTube (opens in a new tab)

How Hard Can a Road Trip Be?

An AI Search Problem · 0:37 · A complete idea from the workbench

Three stops. Six possible orders. Thirty stops? The little car has a much bigger problem—and someone still needs a bathroom.

Three ideas to keep

1. A demonstration has conditions

Early AI programs achieved real results in logic, games, and symbolic tasks. Success inside a carefully prepared problem did not establish that a program could handle every unfamiliar situation. Ask what was tested—and what changes outside that setting.

2. More choices can change the problem

Visiting three distinct stops once each, from a fixed starting point, gives 3! = 6 possible orders. Thirty stops give 30! ≈ 2.65 × 10³²: about 265 nonillion orders.

That counts possible visit orders; it does not mean every route planner must try them all. In our example, a road closure rules out a branch, then a useful clue guides the next choice. Guiding a search can avoid wasted work; a helpful clue alone does not guarantee the best route.

3. The work did not disappear

The 1973 Lighthill report questioned AI’s wider ambitions. In the United States, researchers also faced pressure to justify funding. Later, expert systems could be useful within their prepared territory, but their rules needed care as the world changed.

At the 1984 AAAI panel, researchers debated a possible backlash from inflated expectations. They did not agree that collapse was inevitable. Different parts of AI felt the chill at different times, and important research continued.

Enthusiasm can open the laboratory door. Evidence helps keep the lights on.

Bring these questions back to the workbench

What can this system actually do?
Under what conditions?
What happens when those conditions change?
How much work does it take to keep it useful?

Companion notes, not a verbatim transcript. The Mansion workshop, characters, road-trip map, budget meeting, and DARPA office scenes are illustrative or fictional. AI-generated scenes are not archival footage. The film’s YouTube description includes archival-image credits and production notes.