The AI hypes and the reality of AI capacity

Artificial Intelligence or AI: Everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it (check out Dan Ariely for the original).

In this post I try to describe development in AI in context of its past and current hype cycles along with the actual AI capacity.

AI, which is a subfield of computer science that deals with the simulation of intelligent behaviours in machines, has its origins in the 1950’s. In 1956, John McCarthy organized the Dartmouth conference, inviting leading researchers to discuss ideas around intelligent machines. The term ”Artificial Intelligence” was coined during that conference. What then followed, was a time during which fundamental breakthroughs were expected in AI, and therefore a lot of research funding was invested, and accordingly a lot of AI research was conducted. The excitement around AI lasted until around 1974 when the levels of expectations on AI, and actual achievements did not match – and a phase of disillusionment followed. This phase between 1974-1982 is known as the AI winter. Research funding was withdrawn and accordingly research in AI decreased substantially.

From 1983 until today, AI is experiencing a revival (first gradually and since 2011 rapidly). This revival is mainly due to the advancement in machine learning methods applied to areas such as image analysis and natural language processing. This advancement can be attributed to today’s fast computers on which huge neural networks can process huge data sets. This technology trigger lead to a renewed enthusiasm around AI, similar to the one in the 1950’s. The advancements of AI research have mainly been demonstrated in games where machines beat humans (e.g. Chess, Go), natural language processing, self-driving cars, and medical diagnosis (i.e. medical data analysis).

Now let’s look at hype cycles. The Gartner hype cycle illustrates how technology hypes are triggered by technological advancement, leading to a peak of inflated expectations which is followed by a drop into a trough of disillusionment. After this follows a slope of enlightenment and eventually a plateau of productivity where expectations adapt to the reality of the technological capacity.


Gartner hype cycle

If I sketch AI’s historical and current development I need two Gartner hype cycles. AI’s first hype was between 1956 and 1974 and a second between 2011 and 2018 (follow the black line in the sketch below). We see that expectations rise, fall and flatten. If the Gartner hype cycle is correct, we will experience a second AI winter in the near future.

Now I will expand the sketch with a curve showing the progress of actual development of AI technology (follow the red line in the sketch):

The red line illustrates that a jump in increased AI capacity causes expectations to inflate. On the way up to the peak, expectations rise faster than the actual AI capacity. During the AI winter, technology still developed, but at a slower pace (since research funding was substantially decreased) until the two lines converged on the plateau of productivity. The breakthroughs in AI the last couple of decades caused the current peak of inflated expectations, which, after a possible second AI winter, will converge with the actual AI capacity. But first we should be prepared to put on some warm winter clothes to survive the approaching winter! Brrrr.


Check out the Gartner Hype Cycle for Emerging Technologies 2018.

If you want to learn more about the AI hype cycles, read Alok Aggarwal’s excellent assessments about the first AI hype and the current AI hype.




0 Kommentarer

Lämna en kommentar


E-postadressen publiceras inte. Obligatoriska fält är märkta *