Superintelligent agents can be expected to evolve out of systems that evolved by random mutations.
The systems that they evolved from can be expected to have goals that result in tracking down and utililising every available source of negentropy.
They will build superintelligent agents to help them attain these goals - and so the resulting systems are likely to be even better at tracking down and utilising negentropy than the original systems were - since they will pursue the same ends with greater competence.
Systems with radically different goals are not logically impossible. I call those "handicapped superintelligences". If they ever meet any other agents, it seems that they will be at a disadvantage - since nature disapproves of deviations from god's utility function.
More important than negentropy is continued existence. If the simulation gets shut down, you've already lost all you already have.
An uplifting message as we enter the new year, quoted from Edge.org:
A few thoughts: when considering the heavy skepticism that the singularity hypothesis receives, it is important to remember that there is a much weaker hypothesis, highlighted here by Tegmark, that still has extremely counter-intuitive implications about our place in spacetime; one might call it the bottleneck hypothesis - the hypothesis that 21st century humanity occupies a pivotal place in the evolution of the universe, simply because we may well be a part of the small space/time window during which it is decided whether earth-originating life will colonize the universe or not.
The bottleneck hypothesis is weaker than the singularity hypothesis - we can be at the bottleneck even if smarter-than-human AI is impossible or extremely impractical, but if smarter-than-human AI is possible and reasonably practical, then we are surely at the bottleneck of the universe. The bottleneck hypothesis is based upon less controversial science than the singularity hypothesis, and is robust to different assumptions about what is feasible in an engineering sense (AI/no AI, ems/no ems, nuclear rockets/generation ships/cryonics advances, etc) so might be accepted by a larger number of people.
Related is Hanson's "Dream Time" idea.