The emergence of life and pattern formation in physical systems are examples of self-organization. Despite the second law of thermodynamics, self-organization can increase complexity locally.
The possibility of a complete formal theory of physics exists wherein running a program would reproduce any physical phenomena in the universe; however, computational irreducibility makes it challenging to pick away and determine a fundamental theory.
Scientist Stephen Wolfram explains the principle of computational equivalence, which reveals that when a computing process seems too sophisticated to be simple, it's actually reached a level of sophistication equivalent to everything else. This principle will likely be considered an obvious fact in the future of computing.
The structure of the universe and the laws of physics are amenable to our understanding and have been useful so far despite not understanding everything. The existential risks of AI should not necessarily be our main concern compared to the risks that arise from human nature.
The levels of emergence of novelty in sentient creatures and the attempt to create the same in machine learning are both exciting and daunting as these levels of complexity cannot be fully understood or predicted. However, acknowledging the existence of these levels and having an understanding of what they might entail is crucial for the process.