How can the latent space of a neural network be used to generate speculative visions for the future of environmental photography?

How can we visualise near futures for the environment when the machine learning model will be far more complex and deal with more uncertainty and vagueness than anything we could program?