Active Learning (AL) operates on the concept that machine learning algorithms can achieve greater accuracy with fewer training labels if they can choose the data points from which they learn. Typically, human annotators provide the labels, but their decisions are often influenced by heuristics and biases. Research in social psychology indicates that humans tend to make decisions based on a limited set of attributes, sometimes relying on a single attribute (referred to as Take the Best). This paper presents a closed-form expression that provides the probability of a data point being mislabeled by the human heuristic. Expanding on this, we also introduce a new drop-out mechanism that impacts the human labeller’s attribute selection, thereby nearly doubling the effectiveness of Active Learning.