I sort of agree with your skepticism, but you gotta admit that some of the things the ML folks are doing are uncanny in terms of how they seem to model the human visual system and perform other human-like tasks. Additionally, we already have tons of CPU horsepower that can get close in terms of raw processing ability. Even though we don't yet know what the missing "special sauce" is, I don't think it's inconceivable that someone in 5 years figures it out (though 50 years is just as likely)
