Intermediaries - Objects that simply replicate their input. They obey basic laws of cause and effect where the effect can easily be predicted by the cause.This is interesting to me. It basically parallels what a lot of researchers in complexity theory and systems theory have been finding - that even a seemingly simple function can produce very complex results, usually as a result of some sort of feedback. Mediators are like the dog in Bateson's example - unlike a ball, when you kick the dog you can't necessarily determine its trajectory from the force and direction of the kick. The dog may turn around and bite you, or it may run off to cower in a corner.
Mediators - Objects that transform their input or alter it in some way. The effect cannot easily be predicted by the cause. They are complex.
What's interesting in Latour, I think, is the dichotomy. There are two classes of objects, one which simply replicates the input and one which transforms it. A system composed entirely of intermediaries would not be very interesting; it wouldn't do anything unique, it wouldn't generate anything new and eventually it would run out of steam and collapse. In order for self-organization, pattern and other interesting processes to occur there must be some mediator(s) involved. Furthermore, an object can act in one case as an intermediary and in another as a mediator. For example, humans are typically mediators - unpredictable - but stick one in a large complex bureaucracy and they often become mere intermediaries - performing repetitive tasks with little or no thought, and simply doing what their told until they punch the clock. That person can become an intermediary if s/he notices an injustice in the bureaucratic system and, for example, does small things to gum up the works (interestingly, humans may be unique among animals in their ability to go back and forth between the two roles).
What Latour wants, as far as I can tell (and I haven't finished the book yet), is for scientists - especially social scientists - to treat all objects to the greatest extent possible as mediators. In some cases, this would mean erring on the side of interpreting something as a mediator when its role is ambiguous. That includes both the researcher and the product of the research. Most scientists think of their work as consisting of three discrete objects - the subject of the research, the researcher(s) themselves, and the text where the results are written up. These are taken as being merely intermediaries - not changing or influencing one another. For positivists, the subject is the most important, for critical scholars, the text takes center stage.
For Latour, however, all three are important because they are all mediators interacting in a complex relationship. The world doesn't exist "out there" for us to study "objectively" without interacting or influencing it, and it doesn't stop moving once the text has been written. Researchers must begin to see their work, not as discrete steps that have little relation to one another but as a continuing process of unfolding.
The relationship between the research and the object of study is complex and dynamic - producing novelty and unpredictable behavior. The same is true of the relationship between the text and the researcher, and between the text and the object, and between all of them and the rest of the world. The text collapses numerous historical threads - some of which the researcher isn't even aware of - into a singular event, and it generates numerous new threads, pathways for new possibilities to emerge.
This new way of seeing our research suggests a whole new ethic, where we must represent the process and all of the participants equally as mediators. We have to ask ourselves what our research does, what new threads might it generate, and can we live with those consequences. Research becomes a much more ethical issue as well as a far more interesting practice.