It seems the article was published in 2017, but adding that as a suffix to the submission title would look awkward, I suppose. > "2029 is the consistent date I

It seems the article was published in 2017, but adding that as a suffix to the submission title would look awkward, I suppose. > "2029 is the consistent date I have predicted for when an AI will pass a valid Turing test and therefore achieve human levels of intelligence." Assuming he (and the rest of us) are still around in 7 years, we should be able to congratulate him on being right, or criticise him for not being even close. I think that a couple more advances like GPT-3 that are focused on conversational interaction would be enough to get us there. Ironically, if E2E encryption for chat apps weren't so common, companies like Facebook and Google would probably have enough data for training such an AI already. The missing step, though, is probably the ability for an AI to model the world based on input it receives. If you tell an AI that you're an only child, and then a minute later you talk about meeting your brother for lunch tomorrow, it won't pick you up on your inconsistency, whereas a human probably would, because they build a world model from the facts they read and notice the contradiction. I'm not sure how to train an AI to build world models, but perhaps a manually created corpus of "common sense" could be used to bootstrap some sort of adversarial machine learning process fed with untagged data. That's probably even harder than it sounds, but presumably once someone comes up with some clever way to achieve it, everyone will say it's just applying a simple rule and a lot of statistics, so there's no real intelligence in the AI's conversations.