Drift
Working definition
When an agent's work slowly moves away from what was intended, because the goal was never clear or someone intervened at the wrong level.
Origin. I use drift as a working term here. Gama and colleagues provide the separate machine-learning definition distinguished below. (João Gama, Indrė Žliobaitė, Albert Bifet, Mykola Pechenizkiy, Abdelhamid Bouchachia (2014))
Not to be confused with
- Concept drift: in machine learning, when the relation between a model's inputs and its target changes over time (Gama et al., 2014); a property of data, not of a goal.
Example. A changed goal, stated clearly to the agent that owns the outcome, is not drift. Stepping in below that agent, at the level of the agents doing the work, is how drift starts.