Benkler defines them as "production systems that depend on individual action that is self-selected and decentralized, rather than hierarchically assigned." (Benkler 62) In other words, peer production is when multiple individuals work together on a common project.
To clarify, the word "work" doesn't just refer to generating content. Benkler outlines the basic process of peer production into three basic steps: the initial "utterance," the acceptance of it as valid and/or credible and/or useful, and the subsequent distribution (Benkler 68). For that reason, I think it is fair to include p2p file-sharing (i.e. the numerous clients that run off the gnutella network, Bittorrent, etc) as well peer-review sites such as the volunteer run Open Directory Project. This last example illustrates how collective intelligence and peer production are closely related. Accredidation is in itself a "product" - it is not a good, but it is a service.
A major component of peer production is that it is a non-proprietary system which doesn't rely on the price mechanism. This has two important implications. of The first is that if peer producers aren't motivated by the possibility of getting richer, then why do they do it? This question of motivation is a complex one with no single answer. In the case of fan communities such as the spoiler sites Henry Jenkins mentions, posters contribute their ideas in part because they enjoy the fame and prestige of fellow posters. But in a succesful peer production, individual competition must always take a backseat to the fact that its a team effort - this is what I think Benkler means when he says peer production lacks hierarchy.
This leads into the second important condition of this non-proprietary mode of production: no one person can keep the end product all to himself. Everyone can enjoy the fruits of their labor...and equally.
Benkler makes a point of distinguishing between "coordination" and "cooperation" (Benkler 63). Commons-based systems might also be non-proprietary and essentially a shared resource, but peer production implies that people are contributing. Not all common-based systems involve input and output. Someone who enjoys the benefits of something like wikipedia by reading the articles yet doesn't even haven an account is not a peer producer.
Another variable in peer production aside from motivation is the number of participants. Benkler cites the classic example of someone writing a piece of open source code, another person identifies a bug and tells others about, and then another individual fixes it. Peer production often takes the form of collective problem solving, but in the case of something like Linux this three person model was expanded to include thousands of different programmers.
Yochai Benkler, The wealth of networks : how social production transforms markets and freedom (New Haven: Yale University Press, 2006.