It is very difficult to answer your question. In fact there is a theorem ("No free lunch theorem") which says that no algorithm can be better than the other in all ways.
However, ANN and GP has its own advantages and limitations. The main advantage of GP over ANN is that GP is more transparent, meaning that ANN functional relations are of a black-box nature, and knowledge extraction is next to impossible.
On the other hand, GP suffers from bloat, often ending up in too large expressions, making any physical meaning from the input-output expressions difficult.
Personally speaking, I love GP and try to keep the relations (depth of tree) simple (small), so that short equations explain much of the input-output relationships. The bottom line is: if you want just good prediction, ANN is sufficient; but if you want to know the underlying relations in the data, GP would be a better choice.
In my opinion, the potential of the ANN is significantly higher than that of GP. This is especially true of recurrent neural networks that operate in real time. As for the ANN as a black box, it is possible to argue. Those ANN, I'm working on, logically transparent. Advantages of the ANN as compared with the GP in the fact that they allow to establish communication between the objective events and use them to predict events. Using GP, we rely on strict rules proposed by the programmer. These rules are always limited.
Just FYI: There is a similar discussion going on here on RG.... Visit https://www.researchgate.net/post/Which_is_more_powerful_Genetic_Programming_or_Artificial_Neural_Networks
It is an interesting question. From my own experience, the ANN are excellent as pattern recognition tools but have limitations when it comes to forecasting. Especially for data sets spread over wide range for the forecast parameters.