It is well-attested that people are heavily biased when it comes to evaluating arguments and evidence. They tend to evaluate evidence and arguments that are in line with their beliefs more favorably, and tend to dismiss it when it isn't in line with their beliefs. For instance, Taber and Lodge (2006) found that people consistently rate arguments in favor of their views on gun control and affirmative action more strongly than arguments that are incongruent with their views on these matters. They also had a condition where people could freely pick and choose information to look at, and found that most participant actively sought out sympathetic, nonthreatening sources (e.g., those pro-gun control were less likely to read the anti-gun control sources that were presented to them).
Such attitudes can frequently lead to belief polarization. When we focus on just those pieces of information that confirm what we already believe, we get further and further strengthened in our earlier convictions. That's a bad state of affairs. Or isn't it? The argumentative theory of reasoning, put forward by Mercier and Sperber suggests that confirmation bias and other biases aren't bugs but design features. They are bugs if we consider reasoning to be a solitary process of a detached, Cartesian mind. Once we acknowledge that reasoning has a social function and origin, it makes sense to stick to one's guns and try to persuade the other.
Like an invisible hand, the joint effects of biases will lead to better overall beliefs in individual reasoners who engage in social reasoning: "in group settings, reasoning biases can become a positive force and contribute to a kind of division of cognitive labor" (p. 73). Several studies support this view. For instance, some studies indicate that, contrary to earlier views, people who are right are more likely to convince others in argumentative contexts than people who think they are right. In these studies, people are given a puzzle with a non-obvious solution. It turns out that those who find the right answer do a better job at convincing the others, because the arguments they can bring to the table are better. But is there any reason to assume that this finding generalizes to debates in science, politics, religion and other things we care about? It's doubtful.