Future of Work?

Rob Toews is one of my fav follows on Twitter. In this article, he makes a leap of faith in comparing the current state of AI (or Coginitive Automation) to the 'Centaur Chess' phase of Chess.

Quite a provocative article and an entertaining read. He quotes Geoff Hinton saying that world should stop training Radiologists. In spite of all the investments, hype and effort in the area of machines learning in radiology, this paper shows that in case of Covid19, it pretty much fell flat. Summary of the paper is out of thousands of 'research papers or approaches' written, this team whittled down 62 models for a systematic review of their results in being able to detect Covid19 using CXR or CT images. How many were found robust enough for potential clinical use? Zero. It doesnt prove or disprove anything but the fact that real world deployment of AI and ML requires careful and painful collection of annotated data, with limited biases for training the models.

Meaningful value from models in real world setting comes on the basis of real world data collected and prepared by talented data wranglers. That is one skill which doesnt scale easily, no matter how many future of work articles are published in leading business magazines.