Exercise 1: Where the Rules Come From, and Why ML Didn't Simply Replace Expert Systems — Possible Solution ==================================================================== WHERE MYCIN'S OWN RULES CAME FROM, PER THIS CHAPTER ------------------------------ Per this chapter's own compare-table, expert systems' rules come from "a human expert, written by hand" — stored as "explicit, readable IF/THEN rules." MYCIN's antibiotic-recommendation rules existed because a human medical expert sat down and wrote out, one rule at a time, "if this symptom and this lab result, then recommend this antibiotic" — the knowledge originated entirely inside a person's head and was manually transcribed into code. WHERE A MACHINE LEARNING MODEL'S OWN RULES COME FROM ------------------------------ Per this chapter's own compare-table, machine learning rules are "discovered from data by an algorithm" and stored as "learned numeric weights/coefficients." Per the chapter's own body text, "an algorithm is shown many real examples... and it discovers a pattern connecting them on its own, without anyone ever writing that rule down explicitly." No human ever writes "if mileage is high, then price is low" as an explicit statement — the algorithm infers that relationship purely from seeing many real (mileage, price) pairs, and the result is a set of numbers, not a sentence a person could have written themselves. WHY THIS IS THE FUNDAMENTAL DIFFERENCE ------------------------------ The distinction isn't about which system ultimately makes better or worse decisions — it's about the SOURCE of the decision logic. One approach requires a human to already possess and manually articulate the knowledge; the other requires only a sufficient quantity of labeled examples, with the pattern-extraction work done automatically by the algorithm rather than by a person's own explicit reasoning. WHY ML SOLVED A BOTTLENECK RATHER THAN REPLACING EXPERT SYSTEMS ------------------------------ Per this chapter's own tip-box, "expert systems didn't fail because hand-coded rules are a bad idea — historyai2-6 covers real, honest reasons for their 1980s decline (brittleness, the knowledge-acquisition bottleneck of interviewing experts one rule at a time). Machine learning solves a specific version of that bottleneck — getting rules from data instead of from a slow, manual interview process — not every problem hand-coded rules are good at." The specific problem ML solves is the SPEED and SCALE of getting a rule set built — it replaces a slow process of manually interviewing an expert with a faster process of feeding an algorithm labeled examples. This is a genuinely different claim from "ML is simply superior to expert systems in general" — it's a claim about which specific historical bottleneck each approach does or doesn't suffer from, which is exactly why the tip-box frames this as solving one specific problem, not a wholesale replacement. WHY THIS WORKS AS AN ANSWER ------------------------------ It identifies the precise origin of each system's own rules using the chapter's own compare-table and body text, and explains — using the chapter's own tip-box — why the relationship between the two approaches is "solves a specific bottleneck," not "made the other one obsolete."