View a markdown version of this page

RAIUC01-BP02 Verify that AI is required to solve the problem - Responsible AI Lens
Implementation considerationsResources

RAIUC01-BP02 Verify that AI is required to solve the problem

Before committing to using AI, evaluate whether traditional software approaches or even manual processes could meet your requirements. Choose AI if it provides clear, substantial benefits over competing solutions, and not simply because it is technically possible to apply AI to the problem.

Level of risk exposed if this best practice is not established: High

Implementation considerations

  1. Consider whether you can solve the problem by writing down a clear and compact set of rules. It is generally infeasible, for example, to write down a compact set of pixel-comparison rules to decide if two arbitrary face images represent the same person. However, it is feasible to write down a brief set of rules to decide if two images are identical. If yes, you may be able to use a solution other than machine learning, and you may not need this best practice guidance.

  2. Consider whether you can access reliable sources of training, fine-tuning, and test examples. If it is difficult to access such examples, you may not have the data necessary to develop or evaluate an AI, and may need to consider either reframing the use case or alternate solutions.

  3. Consider alternative reformulations of the use case that might be solved by rule-based systems, traditional information retrieval systems, or other software solutions.

Resources

Related documents: