The term “bias” is used in two very different ways, which often confuses students and young researchers. In statistics, bias refers to the systematic error of an estimator, the difference between its expected value and the true parameter. In regression modelling and machine learning, however, bias commonly means the intercept term or offset, simply a parameter that shifts predictions.

Despite this obvious misnomer, both usages continue to coexist even in the modern era of data science and AI. Textbooks, frameworks, and research papers consistently adopt these parallel meanings without any serious attempt to reconcile them. Perhaps this persists because each community, statistics on one hand, and machine learning/engineering on the other, remains internally consistent. Nevertheless, for learners moving across these domains, the terminology can be unnecessarily confusing.

I am curious to know how colleagues view this issue. Is the dual usage harmless if the context is clear, or should we as a community encourage more precise terminology? In your teaching or research, how do you address the potential confusion that arises from this dual meaning of “bias”? Any Comments?