Look IBM for the last 6 years (21 quarters) has been working on a total joke. The company has been shrinking nonstop under the brain-dead ignorant leadership of Ginni Rometty because AI is just not ready for Primetime in 98% of all fields. The only place AI can replace human expertise is in places where you don't care one Whit about the quality of the result, for example language translation at the level of a 12 year old translator. The only reason people accept that product from Google is because it's free. Same thing for categorizing your bad photos. The only reason AI can flourish is because there is really no one willing to annotate photos a thousand times cheaper like a computer can do it with image recognition. but if you know something about image recognition you will realize that the recognizers make horrible mistakes all the time Nerural networks get really popular every 25 years and then about 5 or 10 years later people realize they STILL can't do anything they promised. In the 1960s Marvin Minsky at MIT killed this research with a book called Perceptrons. In the late 1980s one level networks got popular for optimization (hopfield-tank networks) and died out in about five to eight years later. Now it's the two level networks and deep learning but this perpetually returning overhyped party guest will only last another few years and people will realize that self-driving cars still crash into concrete barriers and that neural networks are not the answer to practically anything. Currently we have neural networks that can play chess and go better than humans and that has been worked on for 60 years. That is all.
Seems like a pessimistic view of it. Perceptron and dual layer neural networks are hardly state of the art. Google Translate works better than ever before, thanks to neural networks built on actual translations.
The bottom line is that many applications simply don't have enough data to effectively train an accurate ML algorithm. Improving performance on existing problems requires more data, and progress in the academic field has limited incremental effects.
Never heard of it.