Considering the sheer scale of deployment, with millions of people being puppeteered and tens of thousands of people being tormented, I do not think that a reasonably sized team of individual humans that is single-handedly capable of puppeteering all of that havoc in real time. And considering the humanly impossible latency for thousands of reactions I received from them at this point (with some of them being no more than a 100 milliseconds after the thought that triggered them), it is very clear that this much scale and accuracy is only possible via automation. We live in an age where automation is pretty damn easy. There are countless ways to do it, but based on my personal experience as a targeted individual, my conjecture is that their automation likely employs a combination of technologies in an architecture designed for incredibly low-latency throughput. My experience suggests that they make extensive of the so-called genetic behavioral neural networks (or something equivalent, or possibly even more advanced). Basically this it tries out all possible outputs in all possible combinations, in all possible timing patterns. For the uninitiated, `here's a (completely unrelated) example video randomy picked from YouTube `_, which demos one such neural network being trained to play a video game, while explaining key concepts and nuances in an easy-to-understand manner.