of course the mechanic of individually selecting people to follow will create a filter bubble containing a select group of people. of course the potential for homogenization bias exists in such a model. but that potential is based entirely upon the criteria for selection, which is up to each user. if you want homogenization (or don't consciously avoid it), that's what you'll get. the model is partly subverted in hubski due to the presence of features allowing one to follow tags and domains, but those features are subject to "subreddit death" and other detrimental phenomena...
Precisely, it is up to each user to determine what he or she will follow. Because of hubski's decentralized nature, it would be hard to call the site as a whole an echo chamber because it's up to each user to make it not so.but that potential is based entirely upon the criteria for selection, which is up to each user. if you want homogenization (or don't consciously avoid it), that's what you'll get.
I believe it's likely that most users will indeed construct an echo chamber. It's very easy to do by accident; most people won't intentionally follow people who share stuff they don't like. I suspect that since the feed shown to logged-out users will be biased according to the collective preferences of the established userbase, it will form a reinforcing effect as new users show up and either participate or decline to participate based upon the content of the site at first glance. There will always be space for 'alternative' views on hubski but they will get harder and harder to find.
Indeed, I often have to make a conscious effort not to construct an echo chamber, and I occasionally stumble upon a new user or tag and think "How am I not already following this?" The answer, of course, is that I was content in my own bubble and it became difficult to break out of. Perhaps the feed shown to logged-out users needs (more) randomization? Maybe some sort of algorithm to ensure it shows posts from different "cliques" of users?