Consensus and Consent
Since we launched the aiX Faculty Fellowship Program last fall, I have kept coming back to a question: why do some changes around AI in education move quickly while others stall?
The way I have come to put it is this: getting consensus to change is hard, but getting permission to try is good enough for reform.
According to the Wikipedia, the distinction I have in mind comes from consent-based decision-making. Its roots go back to a school. In 1926, the Quaker educator Kees Boeke co-founded the Werkplaats in Bilthoven, in the Netherlands, where teachers and children governed the school together through regular meetings. One of his pupils, Gerard Endenburg, turned Boeke’s approach into a formal method, which he described in Sociocracy: The Organization of Decision-Making (Dutch original 1981; English edition, Eburon, 1998). The book’s subtitle states the core idea: “’no objection’ as the principle of sociocracy.”
Sociocracy For All puts the difference in two sentences: “Consensus means a decision is made when everyone agrees. Consent means a decision is made when no one objects.” Under consensus, anyone can block, including over a preference. Under consent, the question is narrower: does anyone have a reasoned objection that trying this would cause harm? A preference is not an objection. Sociocracy 3.0, a guide by Bernhard Bockelbrink, James Priest, and Liliana David, names the problem with unanimity directly: “supremacy lies with whoever decides to block.” Consent-based groups often sum up their test as “good enough for now, safe enough to try.” In S3’s own wording, proposals are accepted “when they are considered good enough for now and safe enough to try until the next review” (S3, Consent Decision-Making).
The idea is not confined to sociocracy. The Internet Engineering Task Force, which has run on “rough consensus and running code” for decades, observes in RFC 7282 that “requiring full consensus allows a single intransigent person who simply keeps saying ‘No!’ to stop the process cold.” It suggests that the useful question is not “Is everyone OK with choice A?” but “Can anyone not live with choice A?” Jeff Bezos made a related point in his 2015 shareholder letter about “one-way door” decisions that cannot easily be undone and “two-way door” decisions that can. He warned that large organizations tend “to use the heavy-weight Type 1 decision-making process on most decisions, including many Type 2 decisions.”
At this time, faculty are far from agreeing about AI but there is an agreement on the urgency of taking some actions. Pilots and fellowship cohorts, like aiX, run closer to consent. A department chair supporting an aiX application isn’t asked to endorse a view of AI in the curriculum. They are asked to let one project go forward for a term, with an evaluation and a final report at the end. The decision is bounded, it can be reversed, and it is reviewed. That is a much easier thing to say yes to, and I think it is a large part of why these efforts move: they only need permission to try.
As a statistician, I also like how S3 frames this. Its Principle of Empiricism asks groups to “treat decisions as experiments” and to define “specific thresholds for success and failure of the experiment” in advance, including “what you would accept as evidence that your hypothesis is false.” That is close to how we would want an educational pilot to work: a stated hypothesis, a plan for evaluation, and a date to look at the results.
Others are running into the same tension. MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training released its report in August. It notes that “the administrative process for making even small changes to curriculum requirements is time-consuming and burdensome,” and it proposes an AI Pilot Fund that instructors can apply to. The report urges faculty governance “to review its processes this fall to ensure the presence of efficient pathways for curricular experiments that enable more rapid exploration, paired, of course, with thorough evaluation of the results.”
Consent is not a free pass. Practitioners argue about who decides whether an objection is “reasoned,” and deciding which objections count is a form of power. RFC 7282 also offers a caution: “An engineering solution that has no objections, but also has no base of support and is met with complete apathy, is not a solution that has any useful sort of consensus.” Permission to try gets a pilot started. It does not get the pilot adopted. For that, the pilot has to produce evidence, and someone has to care about the benefits when it does.
Consensus still has its place. Some decisions should be hard to make. But with a technology that changes this fast, much of the learning we need can only come from asking the smaller question first: is this safe enough to try? And then, just as importantly: what did we learn when we tried it?
