Computer-Generated Chess Problem 03285

Now, this is a 'KBNNPP vs knpp' mate in 4 chess problem generated autonomously by a computer program, Chesthetica, using the Digital Synaptic Neural Substrate (DSNS) computational creativity approach. It doesn't use endgame tablebases, neural networks or any kind of machine learning found in traditional AI. The position below contains 10 pieces which means it simply could not have been derived even from an existing endgame tablebase which is presently limited to 7 pieces.

4K3/5pkp/5N2/4N3/3n2PP/B7/8/8 w - - 0 1
White to Play and Mate in 4
Chesthetica v12.29 (Selangor, Malaysia)
Generated on 14 Apr 2021 at 10:50:27 AMSolvability Estimate = Difficult
Some of the earliest chess problems by humans are over 10 centuries old but original ones by computer are very recent. What was the machine 'thinking' when it came up with this? Did you find this one interesting or have something else to say? Leave a comment below! Feel free to copy the position into a chess engine and discover even more variations of the solution. Anyway, if standard chess isn't your thing, you might instead like SSCC.
Solution