INNOVAREModule 10 · Managing & Exploring AI Capabilities and Adoption

Case 6: The Consensus Machine

The other side of the whole module: augment vs replace done rigorously, the quiet voices behind the pop clichés, and why running your strategy through the same AI as everyone else converges you to the mean.

August 2026 · Case 6 of 6
As you read — hold this question

If every firm runs its decisions through the same models, where does competitive advantage come from — and is “consensus” a strategy at all?

0.71 → 0.49
Give five leading AI models the same 10,272 prompts and measure 304 structural choices: humans scatter (rarity 0.71); the AIs cluster tightly (0.49). The machines converge on the statistical centre.

Every module closes on the other side. Module 10's is the sharpest: the tools built to “support decision-making” are consensus engines. This case fortifies the future-of-AI slide with the quiet primary voices the vendor sources skip.

The course's optimistic close
The textbook versionQuiz: augmentation + culture/ops/workforce

Augmentation, upskilling, and the culture shift

The deck's future slides argue augmentation, not replacement: AI automates routine tasks so people do creative, strategic work; the change runs across Culture → Operations → Workforce skills (with reskilling and role-based training). The Microsoft Work Trend Index is cited — the real figures are 70% more productive, 68% better quality, 57% more creative (the deck's “90% / 84%” is not what Microsoft published).

Carry this

The exam wants augmentation + Culture/Operations/Workforce. This case adds the rigorous counter-argument.

The quiet voices
The rest of the story

The primary sources behind the clichés

The pop “automation paradox” has a 1983 origin: Bainbridge, “Ironies of Automation” — automate the easy parts and you hand humans the hardest part after their skills have lapsed. On augment-vs-replace, reach past the clichés to Zuboff (1988) (automate vs informate), Brynjolfsson (2022) (the “Turing Trap” of building AI to replace), and Autor (2024) (AI could extend expertise). On decisions, Klein (1998) defends expert intuition against “prefer data over hunches”; on oversight, Elish (2019) warns the human-in-the-loop can become a “moral crumple zone,” blamed for the system's failures.

Why these
Citing Bainbridge, Zuboff, Klein and Elish — not the same three commentators everyone quotes — is the point: the antidote to the consensus is to read the quiet originals.
Consensus is not a strategy
The other side

The homogenisation evidence — and the answer that stays yours

The spine, with receipts: Doshi & Hauser (2024, Science Advances) — AI makes individual work more creative but the collective less diverse; Kleinberg & Raghavan (2021, PNAS) — “algorithmic monoculture” can lower welfare even when the shared model is more accurate; Shumailov et al. (2024, Nature) — “model collapse,” where the rare tails vanish first; and StoryScope (2026 preprint) — AI outputs cluster in a shared structural region while humans spread wide.

Applied to Teece: if everyone senses with the same models, everyone senses the same thing. AI can show you where the consensus is; where your answer needs to be different is still yours to make.

The other side

Use the frameworks to pass the quiz — but the move that wins is the one outside the cluster. Consensus is not a strategy.

Quick recall — without looking back

Test yourself on this case

Question 1 of 3

Correct the deck's Microsoft Work Trend Index figures.

Not “90% saved time / 84% more creative.” The real figures: 70% more productive, 68% better quality, 57% more creative — and they're Microsoft's own first-party research on its own product.
Question 2 of 3

Give the primary origin for two of the “other side” ideas the pop versions borrow.

Automation paradox → Bainbridge (1983); augment-vs-replace → Zuboff (1988) / Brynjolfsson (2022) / Autor (2024); human-in-the-loop critique → Elish (2019); intuition → Klein (1998) (any two).
Question 3 of 3

State the “consensus machine” argument and one study that supports it.

AI tools find the statistical centre, so firms using the same models converge and lose distinctiveness. Support: Doshi & Hauser (2024) (collective diversity falls), Kleinberg & Raghavan (2021) (monoculture), or Shumailov (2024) (model collapse).

Module 10 Videos

Module 10 · Long Form

Sources

Module content
BUSN9049 Module 10 — future of AI, augmentation, culture/operations/workforce (deck + videos). Flinders University, 2026.
Quiet voices
Bainbridge (1983); Zuboff (1988); Klein (1998); Elish (2019); Adner & Helfat (2003); Brynjolfsson, “The Turing Trap” (2022); Autor, NBER (2024).
Consensus evidence
Doshi & Hauser, Science Advances (2024); Kleinberg & Raghavan, PNAS (2021); Shumailov et al., Nature (2024); Russell et al., “StoryScope,” arXiv preprint (2026).