ML or Not?
You now know what machine learning is, how it works, and where it fails. This lesson is a practice lesson — a chance to sharpen your ability to look at a real system and correctly identify whether it uses machine learning or fixed rules. This skill matters in the real world: knowing which type of system you are dealing with tells you what it can adapt to, where it might fail, and whether human oversight is especially important.
The Distinguishing Questions
When you look at a system and want to decide if it uses machine learning, ask these three questions. Question 1: Did the system improve by seeing data? A fixed-rule system does not improve with use. A calculator will always add 2+2=4 on day one and day one thousand. A spam filter that improves as it sees more emails is using machine learning. Question 2: Were the decision rules written by a human or discovered from data? If you can read the full decision logic in a programmer's file — if-then-else statements, lookup tables, formulas — it is rule-based. If the rules are numerical parameters adjusted during training on data, it is ML. Question 3: Does the system handle variation and ambiguity gracefully? Fixed-rule systems tend to be brittle: they handle exactly the cases their rules cover and fail on everything else. ML systems tend to be more robust to variation because they generalize from patterns, not from enumerated cases.
If a programmer could theoretically enumerate every rule the system uses in a reasonable-length document, it is probably rule-based. If the system's decisions are encoded in millions of numerical parameters learned from data, it is machine learning.
Let's apply these questions to a set of real systems. A thermostat that turns on heat when temperature drops below 68°F. Rules: if temp < 68 then heat = on. No learning from data. Rule-based. A streaming service that suggests the next show based on your watch history. The suggestions change as you watch more. The decision logic is numerical parameters trained on millions of viewing histories. Machine learning. A calculator. No data, no learning, no variation. Pure rule execution. Rule-based. A phone keyboard that learns your nickname for your best friend and auto-suggests it. Adapts to personal usage patterns from data. Machine learning. A traffic light that switches on a fixed 90-second timer. No data, no learning. Rule-based. A traffic system that adjusts signal timing based on real-time traffic density learned from historical patterns. Adapts from data. Machine learning.
Some systems are hybrids — they use rules for some parts and ML for others. A navigation app uses geometric rules for map rendering but ML for traffic prediction. When in doubt, ask which specific component you are analyzing.
Match each term to its definition.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
A parental control app blocks websites listed on a manually curated blocklist. A technician updates the list each month. Is this ML or rule-based?
A photo editing app automatically adjusts brightness and contrast based on what millions of professional photographers did to similar photos. Is this ML or rule-based?
Which question is MOST useful for identifying whether a system uses machine learning?
ML Detective
- Step 1: Pick three digital systems or features you use regularly — apps, devices, online services.
- Step 2: For each, apply the three distinguishing questions: Does it improve with data? Were its rules discovered from data? Does it handle variation gracefully?
- Step 3: Classify each as rule-based, machine learning, or hybrid.
- Step 4: For each ML system you identified, name one piece of training data that the system likely learned from.
- Step 5: Share your most surprising finding with a classmate and see if they agree with your classification.