An algorithm is a defined, step-by-step procedure for turning an input into an output — nothing more. the United States’ NIST AI Risk Management Framework treats it as a distinct thing from an AI model, describing AI Development work as “the creation, selection…training, and/or testing of models or algorithms” — two related but separate artifacts, not one word for the other.
Canada’s Directive on Automated Decision-Making is the clearest Canadian source on this distinction, because it has to regulate both kinds at once. Its Appendix A defines an “automated decision system” as “Any technology that either assists or replaces the judgment of human decision makers,” then states that “These systems draw from fields like statistics, linguistics and computer science, and use techniques such as rules-based systems, regression, predictive analytics, machine learning, deep learning, and neural networks.” A rules-based system is an algorithm in the plain sense — a fixed procedure a person wrote; machine learning, deep learning and neural networks are listed as separate, more advanced techniques on the same list. The same Appendix goes on to define a “proprietary” system as one covering “systems, algorithms or software owned by an entity, such as a company or government” — treating an algorithm as intellectual property a vendor can own and license, separately from whatever data or model runs on top of it.
The Federal Court draws the same line for a different purpose. the Federal Court’s Notice on the use of AI in court proceedings requires a Declaration when a document was prepared with AI, but is explicit that the Declaration requirement “does not apply to AI that lacks the creative ability to generate new content” and, by name, “does not apply to AI that only follows pre-set instructions, including programs such as system automation, voice recognition, or document editing.” A fixed algorithm following pre-set instructions is, on the Court’s own wording, exactly the kind of tool its AI rules were not written for.
Two intake tools can sit in the same automated decision system and both count as “using an algorithm” under the Directive’s own list, while working nothing alike. One flags a file for review whenever a form field is blank — a rules-based algorithm, fully specified in advance, that behaves identically forever unless someone edits the code. The other is trained on thousands of past files to learn which patterns actually correlate with review outcomes — a machine-learning algorithm whose behaviour comes from data rather than from a rule a person can point to. Both are algorithms; only one learns.
See also AI model, machine learning and artificial intelligence.
This is one term in a plain-English glossary on how AI actually works and where it fits in a Canadian business.
