This is ironic because the exact same thing could have been said about Neural Networks back in the 90s. Just because there aren't any practical applications now of "mathy ML", it does not mean it's a dead end.