dither, noise;  nothing new

noise is not always bad

Nothing here is original

Not to make a point, but merely illuminate a perspective

Stochastic Sampling in Anti-Aliasing

"spurious low frequencies are replaced by random noise
which is easier for our visual system to ignore"
Generally and demonstably untrue;   a broadband noise will likely contain energy that aliases with content of interest, provoking coarser low frequency Stochastic Moiré.

Dither for other than quantization (thresholding) artifact mitigation

Dithering to mitigate quantization artifacts is generally accepted and understood.
Seemingly less appreciated (and forgotton) are
benefits of dither for improved linearization and image quality.
During WWII, British Airborne Interception Radar operators noted improved signals
in the presence of vibrations, e.g. during flights.
Many radars deliberately employ dithering.
Similar benefits are obtained in optical image capture by increased illumination diffusion,
with an opposite extreme example being laser speckle.

bad noise

Practically, masking noise interferes with signals,
which are understood to be information encoded with some recognizable pattern.
Aesthetically, noise distracts from enjoyment of art.  Whether noise can impact enjoyment without also having compromised artistic signals is not addressed here.

Non-masking noise

displaced: in frequency band, time or space

Synapses and nervous impulses.

The human brain is informed by impulses, transmitted among synapses.
To a first approximation, intensities correlate to impulse rates AKA time densities.
Arguably, this dependence on time rate of synapse impulses
prevents us from time-traveling (or at least from sensing it).

Human hearing is based on fluid motion detection by stereocilia of the inner ear.

To a first approximation, these behave like tuning forks,
with different stereocilia responsible for hearing different frequencies.

If the ear were any more sensitive it would detect the random movements of air molecules.

Presumably, detecting molecular movements provides no net survival benefit,
but why do we not hear them?

The brain soon learns to discount them.

Synaptic impulses from molecular movements are masked by other randomness.

Perhaps over areas of ear drums, random molecular vibrations average out, with tympanic membranes acting as low-pass filters with cutoff below audibility (no appreciable stereocilia sensitivities below about 20Hz, where their signals may already be masked by other synapsic impulses provoked by body movements).

perhaps there is lash/hysteresis/lost motion in connections between inner and outer ear.
In which case, as with magnetic tape bias, some non-masking noise may improve audibility of frequencies other than those to which human hearing is most sensitive:

also showing spectral masking (raised threshold of audibility) by 1kHz tone;
masking is also temporal:

.

Can ultrasonic dither increase audibility?

This paper addresses noise-shaping to optimally mask CD audio quantization artifacts,
which is not the same as whether human hearing is improved
in environments with nominally non-masking noise, e.g. by mild wind in:

fields of grass

forests

If real, such effects might help explain preferences for DSD over PCM.

5dB s/n penalty for DSD dithering approximates that for PCM quantization dither

Listen for alternative dithering artifacts (or not)

Advances on Calculating Effective Dither for Audio Signals
Data-Based Moving Horizon Estimation for Linear Discrete-Time Systems
Sonically Optimized Noise Shaping Techniques (ExtraBit)
ExtraBit Mastering Processor