Every organization must cope with ambiguity and uncertainty. At a former employer, we used to interview candidates for the skill “tolerance of ambiguity”… because we had plenty of it. Many management articles discuss ambiguity as something externally given to the organization to deal with. For example, Courtney, Kirkland, and Viguerie characterize four levels of uncertainty, ranging from “forecasting the future” with some certainty, all the way to situations where the range of outcomes can not be identified. The term “forecasting” alone indicates that the ambiguity is about some external force or state.
In contrast, at MIT in 1993, Schrader, Riggs, and Smith argued that uncertainty and ambiguity are different, and that they are chosen by how the problem solver frames the problem.
In the MIT model, they describe the decision process as having a model to evaluate outcomes. Uncertainty occurs when the model is understood, but information relating to the variables of the model is not understood. Ambiguity is where the model is unknown, and in the worst case (level 2) the variables to input to the model are also unknown. The matrix from their paper is here:
The difference is important because to resolve uncertainty, we just need to go collect some data, then apply it to our model. To resolve ambiguity, however, we have to create a model, then get the data. In the worst case we have to propose the variables to put into the model as well. Ambiguity is significantly harder.
Some environments will have inherent and extreme ambiguity as a consequence of the leadership and organization in place; according to the MIT model, they have chosen the ambiguity. My theory is that some organizations bias towards the chaos of ambiguity precisely because it defers decision making. If you are trying to lead in such an organization without either power or authority (or both), the best you might be able to do is reduce ambiguity by making or forcing decisions that enable models to be formed. Start with decisions that have relatively well understood consequences and little resistance, and use them as guideposts to build a fence around the problem. Eventually you should have a smaller number of variables so that a model can be envisioned.