I've been working in machine learning for a relatively long time (~15 years), and most of what I've seen does not pass my AI smell test: "do the results amaze m

I've been working in machine learning for a relatively long time (~15 years), and most of what I've seen does not pass my AI smell test: "do the results amaze me, even though I built it and know how it works?" I am confident true AI, by any reasonable definition, will pass this test, so I think it's a valid question to ask (to be clear I love machine learning, I just think its tight relationship with the term "AI" is an historical error). Deep neural networks on visual and audio data are passing my smell test. At my last job I helped design and build neural networks to process music data. The results were shockingly good: the networks could tell us a very wide variety of information about a song at an accuracy that exceeded human experts, including genre and mood. This is in a domain that most people think of as very human ("understanding" and "feeling" music), which is what made the results so impressive. The goalposts will continue to move, but an inflection point has been hit in the last few years. I believe we'll see at least one or two more inflection points in the next several decades as we move closer and closer to "strong" conscious AIs that are functionally above human intelligence.