[stopachka]

This is a very interesting solution: in effect _remove_ our interventions.

If done cautiously, it could be very effective.

Some of the counter to it could look like this:

a. "Addicting" can be another synonym for "something people want, despite themselves", which means that it may not be up to Facebook:

  - If you remove sugar from snickers for example, it doesn't immediately follow that they will eat less sugar: they could just as well buy kit kats instead.

The interesting question could be: how can we remove the machine learning intervention, while keeping engagement high?

[sp332]

I'd argue that engagement is the wrong metric. For example, when Microsoft made the Windows Phone OS, they tried to keep interactions with the phone as brief as possible. Palm, too, had a three-click-maximum rule for their UI design.

If we can't think of a good metric that should apply to everyone, just give users more control over what they see so they have a chance of figuring out something for themselves.

[stopachka]

This issue lies in the example: why isn’t windows phone dominant today?

We may argue that engagement is the wrong metric, but ultimately what matters is what customers will choose. If choosing between A and B, they choose the more engaging one, you’re in a tough spot.

I agree with your alternative, but with a twist: in effect you need to come up with something, that users will like _more_ then what they are doing now. More customization could be an option, but this hasn’t proven to be true in the market.