Information Entropy Reduction in Algorithmic Ecosystems

Detailed Explanation with At Least 6 Case Laws

1. Introduction

Information entropy reduction in algorithmic ecosystems refers to the process by which a powerful digital platform, algorithm, data intermediary, or AI system reduces uncertainty about users, competitors, suppliers, markets, or future demand by collecting, combining, processing, and predicting information.

In competition law, this concept matters because information itself can become a strategic competitive asset. A dominant platform may possess substantially more information than its business users, rivals, regulators, or consumers. Advanced algorithms can transform fragmented data into highly accurate predictions concerning:

• consumer preferences;
• demand and supply;
• prices;
• switching behaviour;
• competitor strategies;
• advertising effectiveness;
• inventory;
• seller performance;
• customer willingness to pay;
• product quality;
• future market conditions.

The competitive concern arises when the reduction of uncertainty becomes so substantial that the platform acquires an information advantage capable of reinforcing market power.

A useful conceptual formulation is:

Data collection → information aggregation → prediction → uncertainty reduction → strategic advantage → possible market-power reinforcement.

Importantly, entropy reduction is not itself an antitrust violation. The legal issue arises when information advantages are used to exclude rivals, discriminate against business users, facilitate coordination, exploit consumers, or strengthen an existing dominant position.

2. Meaning of Information Entropy

In information theory, entropy broadly measures uncertainty or unpredictability.

If a market participant knows very little about future consumer behaviour, demand, prices, or competitors, uncertainty is high.

If an algorithm can accurately predict these variables, uncertainty is reduced.

For example:

Before extensive data analysis:

Platform does not know which users will purchase Product X.

After analysing millions of behavioural signals:

Platform predicts with high probability which users will buy Product X, when they will buy it, and what price they are likely to accept.

The algorithm has therefore reduced informational uncertainty.

A simplified representation is:

H(X)=−∑p(x)log⁡p(x)H(X) = -\sum p(x)\log p(x)

where H(X)H(X) represents uncertainty concerning possible outcomes.

An algorithm may reduce uncertainty through:

• historical data;
• real-time behavioural data;
• location information;
• transaction histories;
• search queries;
• clickstream data;
• device information;
• social graphs;
• seller data;
• third-party datasets;
• machine-learning inference.

3. Information Entropy Reduction in Digital Markets

Traditional businesses generally possessed imperfect information.

Digital platforms can approach a substantially different model.

Traditional market

Consumer   ↓ Purchase   ↓ Limited transaction information   ↓ Seller learns about one transaction

Algorithmic ecosystem

Consumer behaviour       ↓ Search data       ↓ Clicks       ↓ Purchases       ↓ Reviews       ↓ Location       ↓ Device information       ↓ Cross-platform signals       ↓ Machine-learning model       ↓ Prediction       ↓ Strategic decision

The result is a continuous process of uncertainty reduction.

4. Sources of Entropy Reduction

A. Data Aggregation

Platforms can combine numerous datasets.

For example:

• browsing history;
• purchasing history;
• advertising interactions;
• search behaviour;
• app usage;
• payment information.

The combination can generate information that does not exist in any individual dataset.

B. Real-Time Data

Algorithms can continuously update predictions.

A platform may know:

• current demand;
• current traffic;
• inventory shortages;
• changes in consumer interest;
• seller conversion rates.

This creates an informational advantage over slower competitors.

C. Cross-Market Data

A conglomerate platform may obtain information from several markets simultaneously.

For example:

Search  ↓ Advertising  ↓ Marketplace  ↓ Payments  ↓ Cloud  ↓ AI services

Information generated in one market can improve competitive performance in another.

This is particularly important for ecosystem competition.

D. Behavioural Prediction

Machine-learning systems can transform historical observations into predictions.

Instead of merely knowing:

"This consumer bought Product A."

the system may infer:

"This consumer is highly likely to buy Product B within the next seven days."

That predictive capability may have greater competitive value than raw data itself.

5. Why Entropy Reduction Can Create Market Power

Information advantages can generate several forms of competitive advantage.

1. Better pricing

A platform can estimate demand elasticity more accurately.

2. Better targeting

Advertising can be directed towards consumers most likely to convert.

3. Better product design

User behaviour can reveal which products or features are likely to succeed.

4. Better inventory management

Demand forecasting can reduce costs.

5. Better acquisition decisions

The platform can identify emerging competitors or technologies.

6. Better exclusion strategies

A platform may identify vulnerable rivals and alter rankings, commissions, access conditions, or interoperability accordingly.

Thus:

Information advantage→Prediction advantage→Operational advantage→Competitive advantage\text{Information advantage} \rightarrow \text{Prediction advantage} \rightarrow \text{Operational advantage} \rightarrow \text{Competitive advantage} 

6. Competition-Law Concerns

A. Self-Preferencing

A platform may use information obtained from third-party sellers to improve its own competing products.

For example:

Third-party sellers       ↓ Platform marketplace       ↓ Platform receives seller information       ↓ Platform identifies successful products       ↓ Platform launches competing product       ↓ Platform gives own product favourable treatment

This creates a potential information-to-exclusion loop.

7. Case Law

Case 1 — Google Shopping

European Commission v Google — Google Shopping

The Google Shopping case is one of the most important authorities concerning the use of a dominant platform's control over information, ranking and visibility.

Google operated a general search engine while also operating a comparison-shopping service. The European Commission found that Google systematically favoured its own comparison-shopping service in search results while demoting competing comparison-shopping services.

The legal importance for entropy reduction is that Google's control over search information and ranking mechanisms enabled it to determine how information was presented to users.

Relevance

The case illustrates:

• informational gatekeeping;
• algorithmic ranking;
• platform self-preferencing;
• visibility as a competitive resource;
• information asymmetry between platform and rivals.

It demonstrates that control over the information-distribution layer can affect competition even where the platform does not simply refuse access.

8. Case 2 — Amazon Marketplace Investigations

European Commission — Amazon Marketplace

The European Commission's investigation into Amazon concerned the use of non-public marketplace seller data.

Amazon operated both:

1. a marketplace hosting independent sellers; and
2. its own retail business competing with those sellers.

The central concern was whether Amazon used non-public information concerning marketplace sellers to compete against them.

Entropy-reduction significance

Amazon could potentially obtain information about:

• sales volumes;
• product performance;
• prices;
• demand;
• customer behaviour;
• inventory;
• seller success.

Such information can substantially reduce uncertainty concerning competitive opportunities.

The competition concern therefore becomes:

Marketplace information→Reduced uncertainty→Amazon retail advantage\text{Marketplace information} \rightarrow \text{Reduced uncertainty} \rightarrow \text{Amazon retail advantage}

The case is particularly important for understanding vertical information asymmetry.

9. Case 3 — Meta Platforms / Facebook

German Facebook Data Case

The German competition authority's proceedings concerning Facebook's collection and combination of user data are highly relevant.

The Bundeskartellamt examined the relationship between Facebook's dominant position and its extensive collection and combination of data from:

• Facebook;
• affiliated services;
• third-party websites.

The legal significance extends beyond privacy.

Data combination can increase the informational resolution of a platform's understanding of users.

Entropy-reduction dimension

Suppose individual services provide separate signals:

D1+D2+D3D_1 + D_2 + D_3

When combined, the platform can create a much richer behavioural profile:

Dcombined=f(D1,D2,D3)D_{combined} = f(D_1,D_2,D_3)

The resulting information may enable:

• improved targeting;
• improved prediction;
• stronger personalisation;
• greater advertising efficiency;
• increased user lock-in.

The case therefore illustrates the connection between data concentration, user profiling and market power.

10. Case 4 — FTC v Facebook

FTC v Facebook, Inc.

The United States Facebook litigation concerning Facebook's conduct toward competing social-networking applications provides another important framework.

The broader competition concern involved Facebook's ability to use its ecosystem and control over platform access to maintain its position.

From an information-entropy perspective, a large social platform can accumulate extensive information regarding:

• users;
• developers;
• applications;
• engagement;
• network connections;
• consumer preferences.

The more comprehensive the dataset, the greater the potential predictive advantage.

Competition significance

The case helps illustrate the relationship between:

network effects + data accumulation + ecosystem control + competitive exclusion.

11. Case 5 — Google Android

Google Android — European Commission

The European Commission's Android case concerned Google's practices relating to mobile operating systems, search, browsers and application distribution.

The case is relevant because the Android ecosystem allows control at several interconnected layers.

Operating system      ↓ App distribution      ↓ Search      ↓ Browser      ↓ User behaviour      ↓ Data      ↓ Advertising

The ecosystem can therefore generate feedback effects.

More users create more data.

More data improve services.

Improved services attract more users.

More users generate still more data.

This creates a data feedback loop:

Users↑→Data↑→Prediction↑→Service Quality↑→Users↑Users \uparrow \rightarrow Data \uparrow \rightarrow Prediction \uparrow \rightarrow Service\ Quality \uparrow \rightarrow Users \uparrow

The competition concern is that such feedback may reinforce an incumbent's position and raise barriers for competitors.

12. Case 6 — Google Search / AdSense

Google AdSense

The European Commission's AdSense case concerned restrictions imposed by Google on the placement of competing search advertisements on third-party websites.

The case demonstrates the importance of control over information intermediation.

Google possessed information concerning:

• advertiser demand;
• publisher behaviour;
• search queries;
• advertising performance;
• consumer interaction.

Such information can improve the platform's ability to optimise advertising markets.

Entropy-reduction significance

The more accurately an intermediary understands both sides of a market, the more effectively it can optimise:

• matching;
• ranking;
• pricing;
• advertising placement;
• targeting.

This creates the possibility of information-based intermediary power.

13. Case 7 — United States v. Google

The U.S. search-advertising litigation against Google provides another important illustration of the role of information and distribution advantages in digital markets.

Search and advertising systems generate enormous quantities of data concerning:

• queries;
• clicks;
• advertisers;
• users;
• conversion;
• auction performance.

Such information can make a platform increasingly capable of predicting market behaviour.

The case is relevant to the broader proposition that control over a high-volume digital intermediary can produce cumulative informational advantages.

14. Case 8 — Intel

Intel Corp. v European Commission

Although Intel predates the current AI ecosystem, it remains important for understanding exclusionary conduct by a dominant undertaking.

The case concerned rebates and competitive foreclosure.

Its relevance to algorithmic ecosystems is methodological: competition law examines not merely whether a dominant firm possesses an advantage, but whether its conduct can foreclose equally efficient competitors.

An algorithmic information advantage can similarly become problematic where it is coupled with conduct that makes rival entry or expansion more difficult.

15. Information Advantage Versus Information Abuse

An important distinction must be maintained.

Information advantage

A platform legitimately develops better information through:

• innovation;
• investment;
• efficient data processing;
• superior technology.

This is generally not unlawful.

Information abuse

Problems may arise where information is obtained or used through:

• exclusionary conduct;
• discriminatory access;
• tying or bundling;
• self-preferencing;
• misuse of commercially sensitive information;
• exploitative data practices;
• anti-competitive data combination;
• restrictions on interoperability;
• discriminatory ranking.

Therefore:

Information Advantage≠Antitrust ViolationInformation\ Advantage \neq Antitrust\ Violation

but:

Information Advantage+Exclusionary Conduct→Potential AbuseInformation\ Advantage + Exclusionary\ Conduct \rightarrow Potential\ Abuse 

16. Algorithmic Feedback Loops

One of the most important risks is the self-reinforcing information loop.

Consider an online marketplace.

Stage 1

More sellers join the platform.

Stage 2

The platform obtains more transaction data.

Stage 3

The algorithm becomes better at predicting demand.

Stage 4

The platform improves ranking and recommendations.

Stage 5

Consumers obtain more relevant results.

Stage 6

More consumers use the platform.

Stage 7

More sellers become dependent upon it.

Stage 8

Even more information is generated.

Thus:

Scale→Data→Prediction→Quality→ScaleScale \rightarrow Data \rightarrow Prediction \rightarrow Quality \rightarrow Scale

This can create a data-driven competitive moat.

17. Entropy Reduction and Network Effects

Information advantages frequently interact with network effects.

A simplified model is:

N↑⇒D↑N \uparrow \Rightarrow D \uparrow

where:

• NN = users;
• DD = data.

Then:

D↑⇒Prediction Accuracy↑D \uparrow \Rightarrow Prediction\ Accuracy \uparrow

and:

Prediction Accuracy↑⇒Service Quality↑Prediction\ Accuracy \uparrow \Rightarrow Service\ Quality \uparrow

which produces:

Service Quality↑⇒N↑Service\ Quality \uparrow \Rightarrow N \uparrow

This creates a potentially powerful data-network-effect cycle.

18. Entropy Reduction and AI Systems

AI dramatically intensifies this problem.

Traditional software might identify:

"Customer purchased X."

An AI system may predict:

"Customer is likely to purchase Y next week."

It may also infer:

• price sensitivity;
• churn probability;
• political or cultural preferences;
• willingness to switch;
• likely response to advertising;
• susceptibility to particular recommendations.

Therefore, the competitive value of information increasingly lies not in the data itself, but in the inferences derived from data.

This produces three layers:

Layer 1 — Raw data

What happened?

Layer 2 — Information

What patterns exist?

Layer 3 — Prediction

What is likely to happen next?

The third layer may provide the strongest competitive advantage.

19. Entropy Reduction and Consumer Lock-In

A platform with superior information can personalise its services more effectively.

Personalisation can create switching costs.

For example:

More usage   ↓ More personal data   ↓ Better personalisation   ↓ Higher user satisfaction   ↓ Greater dependence   ↓ Higher switching cost   ↓ More usage

This can produce data-enabled lock-in.

Competition authorities therefore increasingly need to consider whether a platform's informational advantage makes users or business customers less capable of moving to alternative ecosystems.

20. Entropy Reduction and Algorithmic Pricing

Algorithmic pricing provides another major application.

Suppose competing firms use pricing algorithms that observe:

• competitors' prices;
• inventory;
• demand;
• consumer responses.

Each algorithm can reduce uncertainty about competitor behaviour.

The danger is that highly transparent algorithmic environments may make coordination easier.

The concern can be represented as:

Market Uncertainty↓→Price Prediction↑→Strategic Interdependence↑Market\ Uncertainty \downarrow \rightarrow Price\ Prediction \uparrow \rightarrow Strategic\ Interdependence \uparrow

If algorithms facilitate coordination, competition authorities may need to determine whether the conduct involves:

• explicit agreement;
• tacit coordination;
• algorithmic facilitation;
• unilateral conscious adaptation;
• hub-and-spoke coordination.

21. Entropy Reduction and Collusion

Information systems can function as coordination infrastructure.

For example, a platform might provide competing businesses with:

• standardised pricing information;
• real-time demand forecasts;
• common benchmarks;
• algorithmic recommendations.

This can reduce uncertainty about competitor behaviour.

The critical legal distinction is between:

Legitimate market transparency

Consumers and firms receive useful information.

Anti-competitive transparency

Information exchange materially facilitates coordination among competitors.

Thus, regulators must examine who receives the information, what information is shared, how frequently it is updated, and how algorithms use it.

22. Entropy Reduction and Information Asymmetry

Algorithmic ecosystems frequently create asymmetric information.

Platform

May know:

• consumer demand;
• seller performance;
• competitor activity;
• conversion rates;
• price elasticity.

Seller

May know only:

• its own sales;
• publicly visible rankings;
• limited platform statistics.

This creates:

InformationPlatform≫InformationSellerInformation_{Platform} \gg Information_{Seller}

The platform therefore becomes capable of making strategic decisions that sellers cannot replicate.

This can strengthen platform dependency.

23. Entropy Reduction and Essential-Input Theory

In some circumstances, information may become an important competitive input.

The traditional essential-facilities analysis generally focuses on whether access to an input is indispensable.

For algorithmic ecosystems, a similar question may arise:

Can a rival realistically compete without access to a particular dataset, interoperability channel, API, or information resource?

Relevant considerations include:

1. indispensability;
2. replicability;
3. cost of duplication;
4. scale;
5. historical accumulation;
6. privacy constraints;
7. interoperability;
8. network effects.

Not every valuable dataset constitutes an essential facility.

24. Information Entropy and Market Definition

Traditional market definition often focuses on:

• prices;
• substitutability;
• demand;
• supply.

Digital markets require consideration of quality and information.

A service may have a zero monetary price but still generate enormous quantities of behavioural information.

Consequently:

Zero price does not mean zero economic value.

The relevant competitive resource may be:

• attention;
• behavioural data;
• inference;
• prediction;
• interoperability;
• algorithmic visibility.

25. Information Entropy and Dynamic Competition

Entropy reduction can also affect innovation competition.

A dominant platform may observe emerging trends earlier than rivals.

For example:

Search trends      + Consumer behaviour      + Marketplace sales      + Advertising data      ↓ Emerging-market prediction      ↓ Early strategic response

The incumbent can then:

• acquire an emerging competitor;
• copy a successful product;
• adjust pricing;
• launch a competing service;
• modify ranking;
• increase investment.

This can reduce the competitive opportunity available to smaller innovators.

26. Information Advantage and Killer Acquisitions

The informational dimension may be particularly important in nascent markets.

A dominant platform may use ecosystem data to identify:

• rapidly growing startups;
• emerging technologies;
• highly successful products;
• changing consumer preferences.

Information therefore helps the incumbent identify acquisition targets before rivals recognise their strategic importance.

This connects entropy reduction with nascent-competition theories.

27. Regulatory Information Asymmetry

There is an additional institutional problem:

Platforms may know considerably more about their algorithms than regulators do.

A regulator may observe only outputs.

The platform may possess:

• training data;
• model architecture;
• ranking parameters;
• experimentation results;
• internal metrics;
• A/B testing;
• behavioural predictions.

This creates:

Platform Knowledge≫Regulatory KnowledgePlatform\ Knowledge \gg Regulatory\ Knowledge

Consequently, enforcement can suffer from regulatory information asymmetry.

28. The "Black Box" Problem

AI systems may make decisions through complex models that are difficult for regulators to reconstruct.

This can create uncertainty concerning:

• why a seller was demoted;
• why an advertisement was rejected;
• why a price changed;
• why a consumer received a particular recommendation;
• why a rival's product became less visible.

The platform may therefore have an informational advantage not merely over competitors, but also over the enforcement authority.

29. Evidentiary Challenges

Competition authorities may need to obtain:

• algorithmic logs;
• model documentation;
• internal communications;
• training datasets;
• ranking criteria;
• experiment records;
• pricing histories;
• API records;
• recommendation outputs.

Evidence becomes especially difficult where algorithmic decision-making is:

• automated;
• continuously changing;
• probabilistic;
• distributed across multiple models.

30. Remedies

Possible competition-law responses include:

A. Data-access remedies

Controlled access to certain datasets.

B. Data separation

Preventing a dominant platform from combining datasets across markets.

C. Firewalls

Restricting internal use of competitively sensitive information.

D. Non-discrimination

Preventing discriminatory use of platform-generated information.

E. Transparency obligations

Requiring disclosure concerning ranking or data-use practices.

F. Interoperability

Allowing competitors to connect with ecosystem infrastructure.

G. Data portability

Reducing switching costs.

H. Structural remedies

In exceptional circumstances, separating conflicting platform functions.

31. A Competition-Law Test for Information Entropy Reduction

A useful analytical framework is:

Step 1 — Identify the information

What information does the platform possess?

Step 2 — Identify the source

Was it generated by:

• the platform;
• consumers;
• business users;
• competitors;
• third parties?

Step 3 — Measure informational advantage

How much does the information reduce uncertainty compared with rivals?

Step 4 — Examine the algorithm

How is the information transformed into predictions?

Step 5 — Identify competitive use

Is it used for:

• pricing;
• ranking;
• self-preferencing;
• advertising;
• exclusion;
• acquisition;
• coordination?

Step 6 — Determine foreclosure

Are competitors prevented or disadvantaged from competing?

Step 7 — Consider efficiencies

Does the conduct produce:

• innovation;
• better matching;
• lower costs;
• improved quality;
• consumer benefits?

Step 8 — Consider proportionality

Would a less restrictive mechanism achieve the same efficiencies?

32. Six Core Competition-Law Questions

For algorithmic ecosystems, authorities should ask:

1. Who possesses the information?
2. Who generated the information?
3. Can competitors replicate it?
4. How much uncertainty does it eliminate?
5. How is the resulting prediction used?
6. Does the information advantage reinforce market power?

These questions convert an abstract information-theory concept into a competition-law analysis.

33. Key Case-Law Principles

34. Distinction Between Data, Information and Entropy Reduction

It is useful to distinguish three concepts.

Data

Raw observations.

"User clicked Product A."

Information

Processed meaning.

"Users interested in Product A frequently examine Product B."

Entropy reduction

Improved predictability.

"A particular user has an 85% probability of purchasing Product B within seven days."

Therefore:

Data→Information→PredictionData \rightarrow Information \rightarrow Prediction

The final stage may be the most important for modern competition law.

35. Broader Legal Significance

Information entropy reduction suggests that competition law should increasingly move beyond asking:

"Who has the most data?"

and ask:

"Who can convert information into superior predictive power, and how does that predictive power affect competitive conditions?"

Two firms may possess similar quantities of data but have radically different informational capabilities because one possesses:

• superior computing infrastructure;
• better algorithms;
• greater historical data;
• better model training;
• more users;
• more cross-market data;
• better access to real-time signals.

Thus:

Competitive Information Power≠Data QuantityCompetitive\ Information\ Power \neq Data\ Quantity

Instead:

Competitive Information Power=Data+Scale+Algorithms+Compute+Inference+Ecosystem AccessCompetitive\ Information\ Power = Data + Scale + Algorithms + Compute + Inference + Ecosystem\ Access 

36. Conclusion

Information entropy reduction is becoming an important conceptual lens for understanding competition in algorithmic ecosystems. Digital platforms do not merely collect information; they continuously transform information into predictions that reduce uncertainty about consumers, competitors and markets.

The resulting advantage can create powerful feedback loops:

Data→Prediction→Better Decisions→More Users→More DataData \rightarrow Prediction \rightarrow Better\ Decisions \rightarrow More\ Users \rightarrow More\ Data

Cases involving Google, Amazon, Facebook and other digital platforms demonstrate different aspects of this phenomenon, particularly self-preferencing, data combination, marketplace information, ecosystem control and algorithmic intermediation.

The central competition-law principle is therefore:

The possession of superior information is ordinarily a legitimate competitive advantage; the concern arises when a dominant undertaking converts that informational advantage into exclusion, discrimination, coordination, exploitation, or durable foreclosure of rivals.

In the emerging AI economy, the decisive competitive resource may consequently be not merely data, but the ability to reduce uncertainty faster and more accurately than competitors.