The ocean plays an important role in regulating climate and the carbon cycle by absorbing and releasing carbon through the air-sea interface. In order to better understand these dynamics, we need to accurately quantify the amount of carbon exchanged between the ocean and atmosphere reservoirs, known as our air-sea carbon flux. Since the data can't be retrieved by satellites, it is challenging to get a global scale monthly product, so interpolation techniques such as neural networks are used. While these techniques have proven to provide robust observation-based estimates, uncertainties can be high, especially in regions where few observations are available. We calculate the uncertainty and bias created while using a two-step neural network machine learning method, the SOM-FFN. We find the sources of flux uncertainty vary regionally, with subtropical uncertainty dominated by choice of wind product but polar uncertainty influenced most by the coefficient chosen for the air-sea gas exchange transfer. Areas with fewer observations correlate with higher uncertainty and bias. This analysis provides important motivation for maintaining and increasing global ocean carbon observations, and is an important step toward closing the carbon budget through accurate quantification of the fluxes at the air-sea interface.