So, in computer science terms, consensus algorithms are about approximating distributed consensus in the face of benign and malicious threats or communication failures, which devolves to "quorum / majority" pretty quickly in a crisis.
After the crisis, some kind of compensation, reconciliation or excommunication has to happen with the portion that disagreed.
A distributed system can easily make a majority decision that is unwise, like if the votes are based off of bad information. But if the different options of the decision could be formalized and checked/proven as part of the decision making process, based off of axioms and values that participants all agree on, then perhaps the most rational decision could be selected even if it's not what the majority was initially in favor of.