INNOVAREBUSN9049 · AI in Business

Applied Analysis

Module 1 frameworks applied against real organisations, real decisions, and real failures — not restated, not illustrated, but used as analytical instruments that produce findings.

Module 1: AI Foundations — 5 cases
Module 2: How Machines Learned — 6 cases
Updated: 1 July 2026
Smart search — find cases and concepts
How these cases work — applied vs illustrative

Each case uses the week's framework as an instrument — running it against a real-world scenario to produce a finding. The finding is the point, not the framework explanation.

Applied (what these cases do)
Framework → real data → finding. The Innovare Index's 10 models applied to He/Cao/Tan produces: every current model is Stage 4, yet 4 capability levels of empty space sit above them. That's a result. Robodebt run through Wilson & Daugherty produces: zero augmentation criteria passed, $1.76B cost.
Illustrative (what these cases avoid)
Explaining the framework, then giving a generic example that merely restates it. "Netflix uses recommendation algorithms — this is an example of the filter bubble." That's description, not analysis. The framework has to produce something the source material didn't already say.
Module 1 · AI Foundations · 5 cases
What is AI, how did it actually develop, and what are the ethical and human implications?

Frameworks from He, Cao & Tan (four-stage model); ANI/AGI/ASI taxonomy; Morris et al. (DeepMind AGI levels); Netflix ethics template; Wilson & Daugherty (augmentation); Pariser (filter bubble).

Module 1 · Video Walkthrough
→
Case 1 of 5
The Corrected Timeline
The AI taxonomy was named in 1950. The thing it describes first existed around 2012. 62 years of aspirational labelling, mapped against what systems could actually do.
ANI/AGI/ASI Corrected timeline Huang AGI claim
→
Case 2 of 5
The Four-Stage Matrix
He, Cao & Tan's four stages applied to 10 real index models with live pricing. All 10 are Stage 4 — and yet 4 capability levels of empty space sit above them.
He/Cao/Tan Innovare Index 19× price gap
→
Case 3 of 5
The Ethics Template
The Netflix ethics template applied to Coles + Palantir. Prerequisite check fails before the template even runs. $15.7M ACCC settlement as evidence. Plus the Netflix → TikTok slope.
Netflix template Coles / ACCC Algorithmic slope
→
Case 4 of 5
Human-AI Collaboration
Wilson & Daugherty's augmentation model applied to Robodebt: 470,000 false notices, $1.76B settlement, zero augmentation criteria passed. Plus the private sector substitution trap.
Wilson & Daugherty Robodebt Substitution trap
→
Case 5 of 5
The Filter Bubble
Pariser's 2011 warning vs TikTok's For You Page in 2024. He named the right mechanism, the wrong harm. 13 years from book to Australia's world-first under-16 social media ban.
Pariser TikTok / Meta Australia ban 2024
Module 2 · How Machines Learned to Think · 6 cases
From perceptrons to generative AI — the full technical story

Concepts from Storey et al. (2025) foundation models; Vaswani et al. (2017) attention mechanism; learning paradigms (supervised / unsupervised / RL / RLHF / self-supervised); AlexNet / ImageNet breakthrough; NLP and transformer evolution; EU AI Act risk tiers; alignment vs augmentation.

Module 2 · Video Walkthrough
→
Case 1 of 6
The Learning Machine
AlphaFold, Spotify, and ChatGPT were all built differently. The difference comes down to where the training signal comes from — and it changes everything the model can and cannot do.
Supervised Unsupervised RLHF
→
Case 2 of 6
The Deep Revolution
Before 2012, machines needed humans to describe what features to look for. AlexNet changed that — and 700+ FDA-approved medical imaging tools followed. The error rate went from 25% to superhuman.
AI ⊃ ML ⊃ DL Deep Learning
→
Case 3 of 6
The Language Barrier
GitHub Copilot makes developers 55% faster — and it doesn't know what code does. LLMs predict what text looks like after your prompt. That's enough to pass bar exams. It's not enough to know when they're wrong.
Transformers LLMs Attention
→
Case 4 of 6
The Creative Machine
Klarna's AI handled 2.3M conversations in month one. Then quality dropped. Then they started rehiring. Foundation models generalise — until the edge cases arrive. Storey et al. call this emergence and homogenisation.
Foundation Models Emergence
→
Case 5 of 6
The Governance Gap
A court ruled Air Canada liable for its chatbot's false advice. The airline said the bot was a separate legal entity. The court disagreed. The EU AI Act, finalised the same year, had already built a framework for exactly this.
EU AI Act Liability Alignment
→
Case 6 of 6
The OpenAI Playbook
The board fired Sam Altman on a Friday. By Monday he was back. 700 employees threatened to quit. The safety-focused governance structure designed to control the world's most powerful AI lasted 96 hours against commercial reality.
Safety vs Capability Commercial AI
Key Terms & Concepts — click any term
▼ Show
Taxonomy Framework Ethics / Risk Concept Model type
Flinders University · BUSN9049 AI in Business · Module 1–2 Applied Analysis · 2026