There has been a lot of debate about the state of AI in recent weeks on Twitter (see #AIDebate). A lot of it was about naming. It occurred to me that very opini

There has been a lot of debate about the state of AI in recent weeks on Twitter (see #AIDebate). A lot of it was about naming. It occurred to me that very opinionated people had no idea what people in the community are actually doing, and so the whole excercise seemed like a learning experience for them, which seems to be a good thing. The goals of AI (which Google trends classifies as “field of study” - I think that captures it quite well) haven’t changed in decades - to reverse engineer the miracle of human cognition. A certain number of people (like the teams of Yoshua Bengio or Demis Hassabis) have the clear mission to work on just that. The progress in this area is much slower than the perception of the last 5-10 years would suggest. It was just that work from the 90s and 2000s were put to test and quickly outperformed other approaches - symbolic or what we now call “classic machine learning” (e.g. in speech recognition, image classification/detection, machine translation, Information retrieval). All these areas had important and valuable applications in industry and have sucked up a lot of money. But this is only a tiny part of what human cognition entails. Areas around memory, reasoning, consciousness etc. are completely unsolved. Where are we on a scale of 0 to 1000 of solving the problem? Perhaps somewhere between 20 and 50, nobody knows. AI is a north star, and it is a weird development that people have started to call it “AI” again (it felt totally weird about 3-4 years ago when this happened). So, I think the field is still rightly called “AI”. Call the current state of it “system 1”, “differential programming”, “deep learning” or whatever.