I'm not convinced that the ABS graph shows that productivity and earnings were closely coupled before or during the 90s. As it says in the graph title, they've set 1991 as a starting origin (setting Productivity equal to Earnings), so it doesn't imply the two were already as closely coupled as they look. They only appear so close because the graph sets 1991 as the common point to compare both axes.

To demonstrate, I've edited the graph to show what would happen if they made that same graph start from 2009. I've done this by copying the orange line up (and colouring it red) so that both lines begin at the same spot in 2009 instead of 1991. And just like the 1991 line, they appear to match each other for a few years - apart from one major dip around 2016, they align very closely for the first 10 years just like in the full 1991 graph.

But we know from your original ABS graph that the wages were already significantly diverging from productivity by 2009. So, I suspect that if we had a longer graph, then we'd learn that wages were already decoupled from productivity in the decades before 1991, but at the very least this graph doesn't imply close coupling existed in the past and shows evidence of regular uncoupling.

Maybe we should use AI to train them :D

Sure. Although like all tools, AI can only help if used properly. It's not a panacea, and it can't replace most training techniques by itself. Similarly, we can't just "use the internet" to train them or "use books" to train them.