My guess is that an “almost” AGI that is 90% correct and is almost 90% as productive as a human (even though it has many odd quirks) will happen by 2027. An AI like this that will be able to double check it’s own work will be enough to radically change the world.