The way I understand their explanation is that users tend to view the act of rating as a stronger endorsement than it actually is. In other words, users think their ratings are more predictive of what they would enjoy in the future than the implicit signals that can be derived from actual behaviour. Netflix concluded that it's better to boost the weak(er) signal by encouraging users to rate more. They will continue to use it on top of the stronger implicit signals to provide personalized ratings, which they will display as a percentage match as long as it's greater than 50%.
So their recommendations may actually improve.