MMOOZZGG.com

Glossary

What is an AI winter?

In short

An AI winter is a period when disappointment with artificial intelligence leads to cuts in research money and a fall in interest in the field.

History of AI · AI winter

MMOOZZGG.com editorsChecked: 1 min read

For a child

Imagine the whole class started building a huge castle out of blocks. Everyone boasted that the towers would reach the ceiling. But the towers kept falling, the children got bored, and the teacher put the blocks back in the cupboard. Some time later, new pieces arrived, and the building started again. This happened several times to the scientists who teach computers to think. People called these cold pauses winters.

In plain words

These are the years when people almost stop believing in smart programs. The pattern is similar each time. First, scientists and journalists promise that machines will soon understand speech and translate as well as a person, and research gets generous funding. Then it turns out the tasks are much harder and the computers too weak. Governments and companies cut funding, and students move to other fields. A few years later a new approach appears, and interest returns. For example, in the mid-1980s there was a lot of noise around expert systems, and then the excitement gave way to a decline.

For business

For a manager, the history of AI winters works as a vaccine against inflated expectations. Each downturn started the same way: promises ran ahead of what the technology could really do, and projects started without a measurable goal. Today the same risk exists at the level of one company: you buy subscriptions, hold a presentation, and six months later nobody can name the effect in money. The defense is simple. Choose a process with a clear metric, count the costs of data and support, and decide in advance at which result the project closes. If the market cools on the technology again, a project with proven benefit keeps working, while a fashionable project without numbers closes first. You can do this calculation before the start with an AI solutions map.

For a specialist

An AI winter is a phase of falling funding and trust in artificial intelligence research. The first is usually linked to James Lighthill’s report “Artificial Intelligence: A General Survey”, written for the British Science Research Council and published in early 1973. Lighthill wrote about disappointment with the results of twenty-five years of work and said researchers had ignored the combinatorial explosion: the number of options to search grows so fast that methods from toy problems do not carry over to real ones. The phrase “AI winter” was already used among researchers in 1984: at the AAAI-84 conference, the panel “The Dark Ages of AI” discussed whether business and military expectations were too high and whether this would lead to a funding crash. The hype around expert systems in the mid-1980s and its decline are recalled by the participants of a panel on the history of expert systems in AI Magazine (2018). New data, more computing power and probabilistic methods brought the field out of each downturn.

Example

A director remembers how, a few years ago, the company bought a “smart” analytics system and switched it off a year later: nobody understood what result it should give. Before a new project with a language model, he first asks the team to name a metric and the threshold at which the pilot counts as a failure.

Related terms

Questions and answers

How many AI winters were there? Most people count two. The first came in the 1970s, after criticism in the British Lighthill report of 1973.

The second came when interest faded in expert systems, which peaked in the mid-1980s.

Will there be a new AI winter? Nobody knows for sure. Today AI already brings companies measurable results, but if expectations again run ahead of reality, funding may shrink.

Projects with a calculated benefit get through such downturns more calmly than others.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us