Rough Consensus And Group Decision Making In The IETF

I forget what prompted it, but recently I found myself reading through RFC 7282, “On Consensus and Humming in the IETF” (P. Resnick, 2014). It describes the ideas behind the IETF’s tradition of “humming”, and how it differs from voting or “majority rule”. It’s an interesting window into one way of thinking about decision-making in a group.

In IETF discussions, humming is used as a way to “get a sense of the room”. Meeting attendees are asked to “hum” to indicate if they are for or against a proposal, so that the person chairing the meeting can determine if the group is close to a consensus. The goal of the group is to reach a “rough consensus”, defined as a consensus that is achieved “when all issues are addressed, but not necessarily accommodated”.

There is a subtle and important difference between “humming” and “voting”. The RFC talks about how, in the IETF’s experience, voting leads to minority views being ignored, the wrong type of compromises being made and, ultimately, worse technical outcomes. The goal of the rough consensus is to arrive to a lack of strong disagreement, which is different from reaching agreement, or from choosing the majority view.

This system that makes a lot of sense to me in the context of technical discussions. Strong engineering cultures I’ve worked in tended to this style of decision making, even if they didn’t use these precise terms (or humming).

It seems to me that the tricky part of such a system is having skilled “consensus callers” (the people whose role it is to move the group to a decision). The consensus caller needs to be able to listen well and hear all objections, and then exercise good judgement when deciding if the objections are raising a truly problematic point, and if it has been adequately addressed. The RFC states that “simple capitulation on an issue is not coming to consensus”, but doesn’t really discuss potential pitfalls or how to address them, which was disappointing. Any such semi-formal system is only ever going to be as good as the participants’ understanding of power and privilege, bias and group dynamics.