Why can AutoML only return a single output?

For other models, it is fine to have multiple outputs. However, the AutoML will automatically compare different models to find the best one. To compare these models, it will have to look at the accuracy of the prediction, and it’s easier for now to focus on the accuracy of one output. If you had multiple outputs, then you would also need to provide much more details to the AutoML, such as which output is more important? Are they equally important? How do you define a good overall model etc.

In summary, it would be much harder to define one best model with multiple outputs, and this is why for the moment we don’t allow it. However, you can easily train multiple AutoML models, each on a different output.