SINGAPORE - O’Reilly Media, the world’s largest provider of artificial intelligence (AI) surveys, recently announced an annual survey of AI usage among organizations in which it found that the percentage of respondents saying they had AI projects in production was 26 percent, unchanged from the previous year. 

Equally disappointing, the percentage of those saying they were not using AI rose to 31 percent from 13 percent, with a corresponding reduction in the percentage in the middle (evaluating and considering AI) to 43 percent from 61 percent. The 2021 results were surprisingly like 2020’s. Has that little really changed in the application of AI to enterprise problems? 

It was not supposed to be this way. Just a few years ago, many consultancies were forecasting US$15 trillion in economic gains for 2030. Turing Award winner Geoffrey Hinton said in 2016: “We should stop training radiologists now, it's just completely obvious within five years deep learning is going to do better than radiologists.” And in 2020, even after it had become clear that his first prediction was clearly wrong, he said: “Deep learning is going to be able to do everything.”

So far, none of these predictions have come true. Instead, it is O’Reilly Media’s survey results which appear closer to the truth. Forrester, another large consulting firm, also reported that the AI market was only US$17 billion in 2020, or almost 1,000 times smaller than it was supposed to be by 2030. The number of radiology jobs has actually