This is a long overdue post about why Google+ and Facebook (and Twitter, and everyone else) get the "Friend" model wrong. I've mentioned before that I've thought a lot about this problem (and with good reason; many of the tools I'm best known for focus on communication in social networks). And I've come to the conclusion that the friend/follow model is fundamentally broken.
Here's the basic reason why: friend lists, circles, follows, etc. all have to be manually curated. This doesn't come anywhere close to representing relationships in real life, which are continuously fluctuating and notoriously difficult to categorize. In fact there are a huge number of unimaginative startups today trying to draw conclusions based on the assumption that all relationships are basically the same, with predictably inconsistent results.
Some notion of "friends" that reflected real-life relationships would be not be a Boolean, on/off state but rather a number representing the intensity of your connection with a person. In this model, connections are formed by interactions like direct communication, commenting on the same content, etc. The Relationship Number would rise and fall depending on the number and importance of mutual connections you have as well as additional metadata like location, workplace, and education. The reality is that we have the data available to build the algorithms to do this. Why we haven't is beyond me. (Yes, Facebook has Smart Lists. Those are based on profile attributes, not really on interaction.) It would take some tweaking to discover the right signals for what kinds of interaction really indicate connections, but nothing superhuman.
At this point it's worth asking the question: what is the value of "friend" relationships on websites, and is it really even the right goal for these virtual relationships to imitate real-world relationships?
My take is that virtual relationships essentially make it easier for you to see content and activity generated by people related to you. With this in mind, doesn't it make sense that the strength of your connection with someone would be the best indicator of whether you want to see content they produced? (The pattern is not quite as you might expect; it turns out people are less interested in reading what their very closest friends create, because they already know what their close friends have to say. But strength of connection to the author is still a very strong indicator of the value of content.) Yes, the implication here is that there's no way I want to read every single tweet in my Twitter stream, and some of it needs to be filtered out to be manageable. This is one reasonable way to do that.
These relationships in our new model, by the way, can have different weights in each direction. For example, I can read every Paul Graham article in existence and he may still not know I exist. That probably means I want to see anything new PG writes in my activity stream, but he should have no reason to expect my posts to show up in his stream.
It turns out that this model also allows surfacing content that you may never have known about if you had to curate your friend lists manually. You may have a friend-of-a-friend who you never would have known about on Twitter but has really interesting things to say, and a model that doesn't limit your activity stream to people you explicitly add would make it possible for this person's opinion to make its way to your news feed. Hooray for diverse viewpoints! So Mr. Graham may end up seeing my writing after all.
Of course, manual intervention would be necessary at some level. Your network may have a strong attraction to someone you don't particularly care about. There are other problems too -- people with very large or very small networks, for example, require a little more thought. Still, the current model is not the right one for the job, and we have the tools to fix it.