It would take some kind of paradigm shift. Software programming hasn't changed much since the advent of compilers and the C standard. You specify a very limited range of inputs and specify a limited range of outputs and there you have a software module. You create a bunch of software modules to implement an overall design supplied by a project engineer. All of the intelligence is with the engineers. None of it is with the computer. At it lowest level the computer is just a machine flipping the binary switches it was programmed to flip. It has no choice about what switches to flip, it must flip the switches it was programmed to switch. Even within the context of neural networks and fuzzy logic, this is true.
How would somebody go about programming a computer to think of a novel solution for global warming? It would be problematic in the extreme. A computer can never DISCOVER knowledge. That knowledge MUST be programmed into it. It of course can acquire data but that is not knowledge. Knowledge is passed from the programmer to computer via software and then used by the computer to deal with the data.
They are doing that very thing with AI now, or at least trying to. Giving a set of parameters to the AI and seeing if it can come up with a solution. I cant find it now, but there was a test where the AI was told of the parameters in a very simple system, given examples of mechanical motion, and told to come up with a design to move an object and it did. Very simple, but still getting closer.
