Chapter 4
The Rise of AI (2010s–2020s)
Artificial intelligence promised insight without effort — patterns mined from oceans of data. But when machines learn from flawed inputs, they scale our bias and ignorance faster than we can audit it.
1. The New Gold Rush: Data Everywhere
By the 2010s, every transaction, sensor, and clickstream became a data point. Companies built data lakes, hired data scientists, and chased predictive models. Quantity eclipsed quality. The old accountant’s question — “Is this right?” — was replaced by “Is it big enough to train on?”
Theme tie: Abundance hides weakness. Volume feels like validity, but without structure and context, it’s just digital noise.
2. Machine Learning Meets Business Reality
Models learned historical behaviour — including its errors. Forecasts replicated the past under a statistical disguise. Pricing engines optimised margin on faulty cost data. Fraud detection models learned what auditors missed. AI didn’t invent new logic; it amplified old habits.
3. The Rise of the Black Box
Executives celebrated accuracy percentages without understanding features or assumptions. The logic became unexplainable even to its builders. Interpretability fell behind performance.
- Opacity: Non-linear models defied intuition.
- Data drift: Environments changed faster than retraining cycles.
- Automation bias: Humans deferred to algorithms that seemed confident.
4. Ethics and Accountability
Data was the new oil — and, like oil, it spilled. Customer privacy, algorithmic discrimination, and unaccountable automation emerged as boardroom risks. Regulation followed: GDPR, POPIA, and explainability mandates. The same institutions that once ignored data stewardship rediscovered it under legal pressure.
5. From Dashboards to Decisions
AI blurred the line between reporting and action. Recommendation engines priced stock, adjusted discounts, and allocated budgets in real time. Yet most executives could not trace why. The comfort of dashboards replaced the discipline of double-entry thinking.
Practice takeaway: The more complex the system, the simpler the audit question must be: “Show me the data that drove this decision.”
6. Tales from the Trenches
Insert your story here
Prompt: A model looked brilliant in training but failed in production. What input drift or mislabelled data caused it?
Prompt: Describe a time an AI recommendation conflicted with business intuition — which proved right, and why?
Prompt: A regulator or client asked for explainability. How did you reconstruct the logic behind an opaque output?
7. Theme Tie — Pattern Without Purpose
AI can recognise correlations faster than humans can blink, but it can’t decide which matter. Chapter 5 argues that the next era of analytics must reunite machine precision with human discernment — the craft we began with in Chapter 1.
Appendix ideas: model validation checklist, bias-testing framework, and explainability summary template.