AI Technical AptitudeAI Methods9-12.AITA.AIM.4

Rules, Data, or Both

9–12 CS / engineering50 minutes90 minutes (lab extension)

Standard quoted exactly

Justify whether a problem is best solved using procedural instructions, rule-based logic, data-driven methods, or a combination of these approaches.

Example from the standards. Logistic regression for a binary task on an edge device, a decision tree for 20 questions, and a neural network for image classification.

Student-friendly learning targets

I can tell procedural instructions, rule-based logic, and data-driven methods apart with Idaho examples.
I can justify a pick, including a combination, not only a single family.
I can map the standards example (logistic on a small device, tree for 20 questions, neural net for images) into plain jobs.

Essential questions

When is a formula better than a learned model?
When do IF/THEN rules beat both formulas and neural nets?
Why do real systems so often combine the three?

Objectives

Define the four options: procedural, rule-based, data-driven, combination.
Justify logistic regression, a decision tree, and a neural net for the three jobs in the standards example.
Sort eight Idaho problems into the four bins with reasons.
Write one combination design (rules for safety, data for ranking).

Key vocabulary

Procedural instructions: A fixed recipe of steps or a formula: growing-degree days, a checksum, 'add these columns then divide.' No learned boundary.
Rule-based logic: IF/THEN (and AND/OR/NOT) written by people: red-flag if wind and dry; XOR interlock. Related to DA.6's XOR rule.
Data-driven method: A function shaped by examples: logistic regression, trees, neural nets. The data writes a lot of the behavior (AIM.3).
Logistic regression: A linear classifier that outputs a probability for a yes/no task. Cheap enough for a small device; cannot draw XOR's bent boundary.
Combination: A system that uses more than one approach: a formula for GDD, a model for yield risk, a rule that refuses to spray in wind.

Teacher background

AIM.4 is the four-way justification; DA.6 was 'pick a model' with XOR vs. LLM as the headline. Do not rerun the XOR lab — mention it as the rule-based win. Teach the standards example plainly: (1) logistic regression for a binary task on an edge device — yes/no, probability, small compute, linear; (2) a decision tree for 20 questions — readable branches a person can follow; (3) a neural network for image classification — pixels are too many for a hand rule. Procedural is the missing first leg: GDD = f(temps), no learning. Combinations are how Idaho Power, irrigation districts, and dispatch actually ship. Spreadsheet GDD calculation is the procedural demo. Offline card sort is the core.

Materials and prep

Materials

Four bin posters: Procedural / Rules / Data-driven / Combination.
Eight problem cards: GDD formula, XOR pump interlock, red-flag AND, frost logistic on a weather station, 20-questions crop tree, smoke-image net, LLM FAQ, 'rules refuse spray + model ranks hour.'
Standards-example triad cards: logistic+edge, tree+20Q, net+images.
Tiny GDD worksheet (procedural): five days of max/min temp, base 50 F.

Before class

Compute the GDD key. Keep DA.6 XOR graph in reserve as a 60-second reminder, not a re-teach.
Make combination the hardest bin so students do not dump everything there without a reason.

Instructional sequence

Five days of heat, no model

5 min

Compute one GDD by the formula as a class. Ask: did any labeled yield data write this? No — procedural.
Name the other three bins on the wall.

Four approaches, three famous picks

10 min

Define procedural (steps/formula), rule-based (IF/THEN), data-driven (examples write f), combination (name both parts).
Unpack the standards example: logistic on a tiny frost logger (binary, edge, cheap); decision tree as 20 questions (interpretable); neural net for a smoke photo (pixels).
Show a combination: procedural GDD feature → logistic risk → rule 'never auto-spray.'

Place three cards together

12 min

Place GDD, red-flag AND, and smoke-image as a class with a because-sentence each.
Place the combination spray card. Insist both parts are named.
If a pair says 'AI' as a bin, send it back — family or approach required.

Sort the rest and justify

10 min

Pairs place remaining cards, including logistic-on-edge and 20-questions tree.
They write a because-sentence for each and one rejected approach.
Offline: the posters are paper on desks.

Real-world examples

Procedural: Bureau of Reclamation reservoir accounting is arithmetic and policy, not a neural net.
Rules: NWS red-flag criteria; canal XOR-style interlocks from DA.6.
Data-driven logistic: a battery-powered orchard frost alarm that only stores a few weights.
Combination: wildfire dispatch uses a smoke model (net) plus rules that a human must confirm before paging crews (AIM.9 will score those pages).

Hands-on activity

Gallery of because-sentences

8 min

Each pair posts one card they struggled with.
Class challenge: steal a card into combination only if they can name both pieces.
Photograph the bins.

Discussion questions

Why put logistic regression on an edge device instead of a neural net that might score higher on Wi-Fi?
When does a 20-questions tree become unmaintainable compared with a learned model?
Is an LLM procedural, rule-based, data-driven, or a combo with a hidden prompt rule?
What approach should open a spillway?

Differentiation

Support

Two-bin start (rules vs. data) then add procedural and combination.
Because-stems already printed.

Challenge

Design a three-layer combination for irrigation: procedural soil-water formula, data-driven ET model, rules for fish-flow legal minimums.
Argue against the standards example: when would you not use a net for images?

Multilingual learners

Bin names with glosses; cards use icons (formula, IF, scatter, plus-sign).
Oral because-sentences.

IEP / 504

Fewer cards (5); movement to posters is the memory aid.
No timed sorting.

Assessment

Formative

GDD warm-up as a procedural identifier.
Because-sentences during guided placement.

Summative

Completed sort with a because and a reject for four cards, including one combination.
Must correctly place the standards triad or explain a reasoned disagreement.

Success criteria

Student defines all four approaches.
Student justifies logistic / tree / net in ordinary language.
Student does not treat 'combination' as a dump bin.

Responsible use, ethics, and privacy

Responsible use

Justify before you call an API. A neural net on a spillway without a rule layer is a misuse even if it is 'AI.'

Ethics

Combinations can hide who is accountable. Students must name which layer may not be overridden.

Privacy

Problem cards stay public (weather, crops, images of landscapes). No student-location tracking as an 'edge device' demo.

Reflection

Which approach do you overuse, and which do you forget?
Why might a grower demand the tree over the net?
What combination would you trust at 2 a.m. during a frost?

Homework

Pick one home-safe public problem (sports winner from box scores, frost, river flood). Write four short pitches: procedural, rules, data-driven, combination. Star the one you justify and say why the others lose. Paper.

Closing

Formulas, IFs, learned functions, or a stack. The standards example is three data-driven flavors; Idaho work often needs the fourth bin — both. Next we sit with those three families and their jobs.

Extensions and cross-curricular links

Go further

90-minute block: write a one-page design for a county frost system using all four approaches at least once.
Python extension: time a logistic vs. a tiny net on a 100-row table to make the 'edge device' point — optional.
Connect to DA.6 without redoing XOR plotting; one reminder graph is enough.
Printed card sort for a sub plan.

Mathematics: Procedural GDD is a piecewise formula; logistic is a linear function plus a curve to a probability.
Engineering design: Constraints (power, interpretability, law) drive approach choice as much as accuracy.
Government: Legal minimum streamflow is a rule no model is allowed to vote off.