Data literacy and artificial intelligence – biases and limitations

2. What are biases in data?

Bias in data refers to a systematic error that can lead to distorted or inaccurate analysis results, resulting in unfair decisions in education. It arises when certain groups of students, behavioural patterns, or conditions are not properly represented in the data, or when models are applied that do not account for the specifics of the learning environment.

Bias can occur at various stages of data collection and processing, and it is typically classified into several main groups.

• Sampling bias occurs when an artificial intelligence system collects data that is not representative of the actual population.
  

  Example: If an artificial intelligence system uses data only from students in technical faculties, its learning models cannot be accurately applied to students from other fields.

• Measurement bias occurs when artificial intelligence incorrectly records or interprets data.
  

  Example: If the system tracks only time spent in e-learning without distinguishing between active and passive participation, the results on student engagement may be inaccurate.

• Algorithm bias occurs when an AI is trained on biased data, reproducing existing inequalities.
  

  Example: An algorithm that recommends content may favour students with better prior performance.

• Confirmation bias refers to the tendency of an AI system (or its users) to interpret data in a way that confirms existing assumptions.
  

  Example: If an AI model learns that the speed of solving problems is an indicator of knowledge, it may confirm the false assumption that faster students are more successful, even though speed does not necessarily indicate better understanding.

Understanding the different types of bias is essential for the responsible use of AI in education. Only by identifying, analysing and mitigating them can we ensure fair, objective and reliable results.