This is timely, as just last night I was by a fire listening to a field of crickets and wondering if I could use a GAN to learn to interact with their fabric of noise and then convince the crickets to modulate their chirps according to the GAN's feedback. I then wondered how plausible it could be that this had already happened to us. Indeed, that kind of night. But these are the most interesting problems, and adversarial AI for insect control using their own audible telemetry has to be the most viable path for having it interact with biological systems. If it can't pass an insect Turing test, it probably can't pass a human one.
Thank you for posting this, it was truly a hidden gem.