This can be either because one posits a fundamental limitation on self-conceptualization of systems, or a practical one on human biological or social intelligence.
By observing another small consciousness? That's like looking in the mirror. The traditional viewpoint of metaphysics, which studies consciousness, is that there are many such tiny mirrors everywhere, and all these mirrors reflect the one true consciousness that tries to understand itself by looking into these mirrors. Once it reaches this understanding, all the mirrors will go flat, will merge into one surface and will effectively disappear.
The "cannot escape" part is where physics disagrees with metaphysics. The latter believes that there's 'absolute existence' that creates and uses reality as a crutch or a mirror to learn walking.
If there was an AI, it would encounter the same problem: it exists in some indescribable form, but it cant touch itself, it cant see itself, it cant understand itself. So it would create a 'reality' - also a fictional thing, but with properties such as inertia, reaction and invariance (the three laws of Newton), and then it would constrain itself with that reality, so it would feel real. After that it would start building fictional sandcastles and understand itself.
