Algorithms can be used to produce other algorithms in machine learning, but more precisely, a training algorithm creates a trained model that can then perform a task. The training algorithm adjusts the model's internal parameters using data, rather than normally writing a completely new program.
This topic is part of 3.2 Algorithms and 3.6 Artificial intelligence, shared by SL and HL.
In machine learning, developers first choose a model structure and a training algorithm. The model processes training data, compares its output with the expected result, and calculates the error. The training algorithm then changes the model's parameters to reduce that error.
For example, consider a system trained to classify emails as spam or not spam:
| Stage | What happens |
|---|---|
| Input | The system receives labelled examples of spam and non-spam emails. |
| Prediction | The model predicts a label for each email. |
| Error measurement | The prediction is compared with the correct label using a loss function. |
| Adjustment | The training algorithm changes parameter values to reduce future errors. |
| Output | The trained model applies its learned decision process to new emails. |
The resulting model functions algorithmically because it takes an input, processes it according to learned parameters, and produces an output. However, its decision rules may not have been explicitly written by a programmer. They emerged from repeated parameter adjustments during training.
A common misconception is that the machine learning system independently "creates its own intelligence." It does not. Humans select the training data, model structure, objective, and evaluation method. These choices can introduce algorithmic bias, especially when the training data is incomplete or unrepresentative.
In an IB Digital Society exam, an explain question requires the stages and their causal links, not just a definition. State that a training algorithm adjusts a model's parameters using data, then connect this process to transparency, accountability, bias, and the impacts and implications for people and communities.
