This is interesting because a similar thought experiment that immediately occurred to me would be to consider a program that calculates digits of pi, sqrt(2), or some other transcendental number, given some user input --- basically, a calculator.

The program itself is almost certainly copyrightable, but what about its output?

These mazes feel like a similar idea to me - he even mentioned that this particular one is generated by a "fibonacci spiral", so there is likely a compact mathematical description of its vertices, and a PRNG (further thought experiment: what if the "randomness" was generated by a true RNG?) along with its seed that was used to determine where the lines appeared. It's an artistic work, but the majority of the "work" was done by an algorithm - he didn't manually draw all the lines.

Remember, copyright is about "provenance", or how you arrive at the result. If you get the same maze but you didn't copy from the original, it can't be a copyright violation. It's the copying that creates the problem. That said, there is a "minimum of creativity" that limits what you can copyright. https://en.wikipedia.org/wiki/Threshold_of_originality

It's evident to me that the current notions of the "Threshold of originality" lag far behind the capabilities of today's media. Just because something is mechanically produced shouldn't mean that it can't be copyrightable. That said, it's pretty foolish to copyright something like that. I just don't think I can determine that it would always be foolish to copyright such information, so the conservative choice would be to leave the copyright in the hands of the meta-creator.