We'd have to take pictures and feed them into a convolutional neural network to see if it could identify factors that correlate with mbti/socionics/enneagram type. Might make the practice of Visual Identification more rigorous and scientific. Unfortunately you would need a lot of training data. Likely several hundred thousand to several million profile pictures. Maybe you could use Facebook to do it with Cambridge's apply magic sauce framework to guess people's types.

Apply Magic Sauce - Prediction API -

What do you guys think?