Imagine using machine learning to create shadow profiles of users. You don't need to focus so much on the content of their messages directly, but rather you score the replies that come after a user's message.
When a user's shadow profile reveals them to be a conflict amplifier, or one who responds to other amplifiers, then the usage of the site is throttled subtly for them in a way that discourages engagement. For example you could lock the reply component for 10 seconds on page load, or don't show replies that are less than 90 seconds old, and other small nuances.
All of this is to say, I bet there is a technological solution, contrary to the author's assertions.
When a user's shadow profile reveals them to be a conflict amplifier, or one who responds to other amplifiers, then the usage of the site is throttled subtly for them in a way that discourages engagement. For example you could lock the reply component for 10 seconds on page load, or don't show replies that are less than 90 seconds old, and other small nuances.
All of this is to say, I bet there is a technological solution, contrary to the author's assertions.