That can only be decided in hindsight. By the time everybody agrees that the system is clearly generally intelligent, it will have been for ages already. It will already be far more intelligent than even very smart humans.
But I think general problem solving is a part of it. Coming up with its own ideas for possible solutions rather than what it generalized from a training set, and being able to try them out and iterate. In an environment it wasn't specifically designed for by humans.
Humans are very different from computers. In particular there are some things that computers are vastly better at (computation, memory, etc), and humans are optimized for surviving in their biological environment, not necessarily for general intelligence.
I think asking of an AGI to do what humans do is asking a submarine to swim. It's not very useful.
So I think that when we have useful computer AGI, it will be much better at it than humans.
You already see that even with say ChatGPT -- it's not expert level, but the knowledge it does have is way way wider than any human's. If we get something that's as smart as humans, it will probably still be as widely applicable.
And why even try, otherwise? We already have human intelligence.
But I think general problem solving is a part of it. Coming up with its own ideas for possible solutions rather than what it generalized from a training set, and being able to try them out and iterate. In an environment it wasn't specifically designed for by humans.
(not claiming most humans can do that)