AES Paper: Hires is Audible
Unfortunately, there is not currently nor is there likely to be in the foreseeable future definitive proof that hires is audible. Scientists are subject to corruption just as much as any other human beings. Over the years many scientists have published papers which conclude that, for example; Smoking is not injurious to health, lead in gasoline is not harmful, man made climate change does not exist, etc. The consumer audio equipment and content industry has an existential interest in hires and will commission papers to refute evidence to the contrary and even accepted fact in order to maintain their existence, just as the tobacco, fossil fuel and other industries have.
Unfortunately, there is an added difficulty with hires; it's not such an obviously black and white issue as say smoking tobacco, which has always been injurious to health regardless of any marketing or even evidence to the contrary. Hires is/can be patently obvious! An example: The sound of an electric guitar is partially dependent on IMD, tones/frequencies generated by a guitar amplifier in the audible band from ultrasonic frequencies. We don't need to actually record these ultrasonic freqs, just the audible band IMD product (tones) which of course can be captured with a mic at 44.1kHz. However, there are a number of software guitar amps in common use which model the response of actual guitar amps and to do so accurately requires those ultrasonic freqs. In this case, a sample rate of say 96kHz is obviously distinguishable from a 44.1kHz sample rate as 44.1kHz cannot process ultrasonic freqs and will not contain those resultant essential tones in the audible band. Other examples include modelled compressors and limiters and some soft-synths, although in the latter case often because of programming priorities rather than ultrasonic IMD components and programming priorities is also responsible for differences in other areas of DSP, such as filters for example. Generally, all this adds up to relatively little (though potentially audible) difference as far as consumers are concerned but for those actually creating content the differences are more obvious because we can easily compare the performance of various digital processors at different sample rates and these days that means not just an elite few but a significant number of the general population.
Having said all this, it would appear that there could indeed be a discernible difference, however, the above describes the state of play a decade or so ago. Since then computing power has increased and DSP programming has advanced. It's now practical and indeed common/usual for DSP plugins to locally up-sample when there is need (with modelling plugins for example) and so the potential sound quality need to run sessions at 96kHz rather than 44.1kHz no longer exists and even the creators can no longer tell a difference.
G