If energy can be recognized in the machine learning sense then its imagination can be directed with physical laws as the target. That is. Energy ?is directed imagination.
If energy is just directed imagination then it would be possible to reverse events that seems irreversible.
Since you can use a Generative Adversarial Network, GAN to differentiate between fake/real images and photos. It would be possible to tell if a generated state transition or movement is real or fake. Just two images. Do they look real and do the belong to each other in time.
Then I don’t see a problem with reversing an event in time. Some kind of reverse entropy change.