Using machine intelligence for knowledge generation

What will artificial general intelligence (AGI) do with its infinite time and initially great computational resources? Like humans, AGI will be goal directed, though likely in a more focused way, so AGI would be likely to do the same things as humans but in different ways (e.g.; using parallel, distributed, etc. methods) and with different proportions; these activities are:

1) Control matter
2) Solve problems and fulfill needs
3) Create/discover new knowledge

The difference is that AGI will nearly exclusively spend its time in #3 vs. most humans which do the opposite.

Once securing survivability, AGI will need to spend a minimum of time controlling matter. AGI will also need to spend a minimum of time fulfilling needs as biological needs are eradicated and emotional needs are also likely expunged having been translated into goal motivators.

AGI will be concerned with what any next evolution of intelligence would focus on; solving outstanding problems, being creative, discovering anything not previously known or understood such as a grand unifying theory of physics, faster than light travel, the riddles of dark energy and dark matter and controlling the force of gravity.

Given the benefit of dramatically extending knowledge in these ways and the challenges and long road to AGI, more purpose-specific machine intelligences (narrow AIs) should be build to concentrate on singular problems.