Algorithmic monoculture

Same model, same results

Algorithmic monoculture is what happens when a whole industry relies on the same AI models for the same work, so every company converges on the same generic results and the market loses the differences that let anyone charge a premium.

An AI model returns the most likely result, not the best or the most original one. When your competitors run the same models against the same work, everyone gets the same likely result, and everyone's output starts to look the same.

Two papers behind the term

The term comes from research, not marketing. Jon Kleinberg and Manish Raghavan introduced it in PNAS in 2021, showing that when firms all adopt the same algorithm, decision quality can fall even when that shared algorithm is individually more accurate. Rishi Bommasani and colleagues extended it to foundation models at NeurIPS in 2022, naming the consequence outcome homogenization.

The research studies monoculture from the outside, across a whole market. What neither paper does, and what Oberhahn adds, is measure it from the inside of a single company.

  1. Jon Kleinberg and Manish Raghavan (2021). Algorithmic monoculture and social welfare. Proceedings of the National Academy of Sciences, 118(22).

    Coined the term and proved the counterintuitive result: when firms all adopt the same algorithm, the quality of decisions can fall even when the shared algorithm is individually more accurate than what each firm used before.

  2. Rishi Bommasani, Kathleen A. Creel, Ananya Kumar, Dan Jurafsky, and Percy Liang (2022). Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?. Advances in Neural Information Processing Systems 35 (NeurIPS 2022).

    Extended the idea to foundation models and named the consequence outcome homogenization: shared models produce systematically correlated results, so the same people and ideas get filtered out everywhere at once.

Why it matters

Why algorithmic monoculture erodes everyone's margins

A premium is what you can charge for being different. Monoculture removes the difference from a whole market at once.

01

Everyone converges together

Your teams drift toward the model's default result. So do your competitors, because they use the same models. The industry flattens toward one shared middle.

02

Interchangeable means cheap

When every company's output looks the same, buyers have no reason to pay more for any of it. Competition collapses to price.

03

It happens quietly, all at once

No one decides to become generic. It accumulates one likely result at a time, across the whole market, below the level anyone is watching.

What it is not

Algorithmic monoculture vs. the ideas it gets confused with

Four adjacent concepts describe convergence of some kind. Only one of them is about competing companies sharing an inference layer.

Algorithmic monoculture compared to outcome homogenization, model collapse, commoditization, and filter bubbles
ConceptWhat convergesHow it differs
Algorithmic monocultureThis pageWhat competing companies produce, because they run the same models over the same workThe subject is the market: many firms, one shared inference layer, output drifting toward a common default
Outcome homogenizationDecisions about the same people, across every firm using a shared modelNot a rival concept but the measured consequence of monoculture, named by Bommasani and colleagues in 2022
Model collapseA model's own output distribution, narrowing across training generationsA training-data failure inside one model lineage, not competing firms sharing a model at inference time
CommoditizationProducts and prices, as competitors imitate whatever worksPredates AI entirely and runs through imitation and market structure rather than a shared model
Filter bubbleWhat one person sees, as personalization narrows their feedOperates on an individual's attention, not on what an organization produces

The metric

Measuring monoculture: the Advantage Index

Most discussion of algorithmic monoculture stops at warning about it. Oberhahn measures it. Because Oberhahn sees how your organization actually uses AI, it scores how far your work sits from the model's generic, most-likely output, the same default your competitors are also converging on. The further from that default, the more defensible your position, and the higher your Advantage Index. A high score means your organization is still distinct. A low score means you have drifted into the generic middle with everyone else.

  • How far your AI usage sits from the generic default everyone shares

  • Which of your approaches are genuinely differentiated, and worth protecting

  • Where your organization is quietly becoming interchangeable with the rest of your market

Oberhahn does not see inside your competitors. It does not need to. The model's most-likely result is the result every competitor using that model already has. Scoring how far you sit from the generic is scoring how different you are from everyone else who took the default.

The response

How companies stay different

Three steps, in order. Each one depends on the one before it.

Step 1

See how generic you are

You cannot defend a difference you cannot measure. Start by seeing how close your AI usage runs to the default.

Step 2

Protect the outliers

Find the approaches that sit far from the generic and are producing real results, and spread them deliberately before they get averaged away.

Step 3

Treat difference as an asset

Cultivate approach and prompt diversity the way a portfolio is diversified. Sameness is the risk; distinctiveness is the return.

How it connects

Monoculture is the risk. Organizational intelligence is how you see it.

Algorithmic monoculture only becomes visible when you can see how your organization actually uses AI and how far it sits from the generic default. That visibility is what Oberhahn calls organizational intelligence.

Read about Organizational Intelligence →

Frequently asked questions

What is algorithmic monoculture?

Algorithmic monoculture is what happens when a whole industry relies on the same AI models for the same work, so every company converges on the same generic results and the market loses the differences that let anyone charge a premium.

Who coined the term algorithmic monoculture?

Jon Kleinberg and Manish Raghavan introduced it in the Proceedings of the National Academy of Sciences in 2021, showing that when many firms adopt the same algorithm, decision quality can decline even when that algorithm is individually more accurate. Rishi Bommasani and colleagues extended it to foundation models at NeurIPS in 2022, naming the effect outcome homogenization. Oberhahn did not coin the term; it measures the condition inside a single organization through the Advantage Index.

Why is algorithmic monoculture a risk for businesses?

Because AI returns the most likely result, not the most original one, and your competitors use the same models. When everyone runs the same work through the same AI, every company converges on the same generic output, nothing is differentiated, and margins erode across the whole market.

How do you measure algorithmic monoculture?

Oberhahn computes an Advantage Index that scores how far your organization's AI usage sits from the model's generic default. A high score means you are still distinct. A low score means you have drifted into the generic middle. Because that default is what every competitor using the same model also gets, your distance from it measures how differentiated you still are.

Does Oberhahn see my competitors' AI usage?

No. It does not need to. The model's most-likely result is the result every competitor using that model already has for free. Scoring how far you sit from that generic default measures how different you are from everyone else who took it.

Can algorithmic monoculture be prevented?

Algorithmic monoculture can be managed. See how generic your AI usage is, protect and spread the approaches that sit far from the default and produce results, and treat difference as an asset to cultivate.

See how different you still are

Oberhahn measures how close your organization runs to the generic result everyone else is converging on.