Defining AGI as computers being better than experts at their specific tasks isn’t AGI, it’s just AI. The “G” means general. And it means one neural network with generalized and generalizable intelligence. It won’t happen, ever. Not unless and until the machine is able to intentionally contradict its node weights and still carry on a cogent conversation.

The Turing test will soon be beaten (maybe it already has), but that test is flawed. It’s not a test for AGI, it’s a test for how best to fool average people. Not a very high bar.