AI Epistemic Scaffolding Systems and Knowledge Dependency

Detailed Explanation with At Least 6 Case Laws

1. Introduction

AI epistemic scaffolding systems are AI systems that do more than provide isolated answers. They structure how users find, evaluate, organize, remember, and apply knowledge. Examples include AI search engines, generative-AI assistants, AI research platforms, automated legal-research systems, medical decision-support systems, educational AI tutors, enterprise knowledge copilots, and AI systems that summarize or rank information.

The competition-law concern arises when users, businesses, researchers, professionals, or public institutions become dependent upon one AI ecosystem for access to knowledge, data, sources, rankings, recommendations, verification, and decision-making tools.

The central competition question is therefore not simply:

“Does an AI company possess a large amount of information?”

It is:

Can control over the infrastructure through which knowledge is accessed and interpreted be converted into durable market power, exclusion of rivals, or control over downstream markets?

AI epistemic dependency can arise at several layers:

1. Data layer – control over training and reference datasets.
2. Retrieval layer – control over search, indexing and information retrieval.
3. Model layer – control over foundation models.
4. Inference layer – control over how information is synthesized.
5. Interface layer – control over the user's principal knowledge interface.
6. Verification layer – control over citations, provenance and factual validation.
7. Workflow layer – integration into professional or institutional decision-making.
8. Feedback layer – collection of user interactions that further improve the dominant system.

The resulting concern is sometimes described as knowledge dependency: competitors or users may technically have access to information but nevertheless become economically dependent upon the dominant AI system's method of organizing and presenting it.

2. Meaning of AI Epistemic Scaffolding

A. Meaning of “epistemic”

“Epistemic” concerns knowledge—how knowledge is obtained, validated, organized and relied upon.

Traditional search engines primarily provide access to information.

An AI epistemic system increasingly performs additional functions:

Question → Retrieval → Selection → Interpretation → Synthesis → Recommendation → Decision

Thus, the system can become a cognitive intermediary between the user and the underlying information environment.

B. Meaning of “scaffolding”

Scaffolding refers to technological structures that help users perform cognitive tasks.

For example, an AI legal platform may:

• locate authorities;
• determine which authorities appear relevant;
• summarize judgments;
• identify competing interpretations;
• generate arguments;
• draft legal propositions;
• recommend further authorities;
• assess the strength of an argument.

The user consequently does not merely consume information. The AI system helps determine what information the user sees and how that information is understood.

3. Competition-Law Theory of Knowledge Dependency

Knowledge dependency becomes a competition concern where a dominant AI intermediary controls an essential or strategically important knowledge pathway.

A simplified chain is:

Underlying information → AI access mechanism → AI interpretation → user decision → downstream market

If rivals cannot effectively reproduce the intermediary's access to data, distribution, feedback, reputation, or ecosystem integration, the AI platform may acquire a structural advantage.

Potential theories include:

• refusal to supply;
• discriminatory access;
• self-preferencing;
• tying and bundling;
• exclusive dealing;
• interoperability restrictions;
• data foreclosure;
• leveraging;
• exploitative conduct;
• exclusionary conduct;
• merger-related entrenchment;
• algorithmic discrimination;
• manipulation of ranking or recommendations.

4. Knowledge Dependency as a Network Effect

AI systems can exhibit unusually powerful feedback loops:

More users

↓

More queries and behavioural data

↓

Better personalization and model adaptation

↓

Greater usefulness

↓

More users

This can be reinforced by:

More developers → more integrations → more workflows → more switching costs → greater dependency.

The competitive problem is particularly serious where the AI platform becomes the default epistemic gateway.

For example, if lawyers, doctors, engineers, students and businesses increasingly obtain their initial knowledge from one AI system, rival providers may face difficulty obtaining user attention even when their underlying information is comparable.

5. Relevant Competition-Law Case Laws

Because “AI epistemic scaffolding” is a relatively new concept, there are not yet many reported decisions applying that exact terminology. The following cases provide direct or closely analogous competition-law principles concerning access to information, infrastructure, interoperability, digital ecosystems, essential inputs and leveraging.

Case 1: Bronner v Mediaprint

Principle

The European Court of Justice considered when refusal by a dominant undertaking to provide access to an infrastructure may constitute an abuse of dominance.

The Court established a demanding framework for treating an infrastructure as indispensable.

Relevance to AI

An AI knowledge infrastructure could become analogous to an indispensable facility where:

• users cannot reasonably obtain equivalent functionality elsewhere;
• replication is technically or economically impracticable;
• access is necessary to compete downstream;
• refusal substantially eliminates effective competition.

For example, suppose a dominant AI platform controls an indispensable knowledge-indexing infrastructure and refuses reasonable access to rival AI applications.

The Bronner framework becomes relevant.

Competition concern

The mere fact that an AI system is popular would not automatically make it an essential facility.

The crucial question would be whether effective competition is practically impossible without access.

Case 2: IMS Health GmbH & Co. OHG v NDC Health

Principle

The European Court of Justice examined refusal to license a protected information structure where access could potentially be indispensable for competing downstream.

The case is particularly relevant because the infrastructure involved the organization of information rather than a conventional physical facility.

Relevance to AI epistemic systems

This provides an important analogy for AI knowledge infrastructures.

Consider a dominant AI provider controlling:

• a proprietary knowledge graph;
• unique datasets;
• specialized classification systems;
• highly valuable information architecture;
• industry-specific mappings.

If competitors cannot effectively reproduce those inputs, refusal to provide access could potentially raise competition concerns.

Important limitation

The existence of valuable data alone does not create an automatic obligation to share it.

Competition law must distinguish between:

valuable information

and

indispensable competitive infrastructure.

Case 3: Magill TV Guide / Radio Telefis Éireann and Independent Television Publications

Principle

The European Court developed the exceptional circumstances under which refusal to license intellectual-property material can constitute an abuse of dominance.

The case involved control over information necessary for a downstream publication market.

AI relevance

The analogy to AI epistemic scaffolding is especially strong.

An AI platform may control:

• structured information;
• databases;
• proprietary classifications;
• access to information sources;
• information aggregation systems.

If that control prevents the emergence of downstream knowledge services, information foreclosure may become a competition issue.

Example

Suppose an AI platform controls a uniquely comprehensive database of professional information and simultaneously operates the dominant downstream AI research assistant.

It could potentially have an incentive to:

1. restrict access to the database;
2. give its own assistant privileged access;
3. deny equivalent access to competing assistants.

That could raise both input foreclosure and leveraging concerns.

Case 4: Microsoft Corp. v Commission

Principle

The European Commission and EU courts examined Microsoft's conduct concerning interoperability information and the relationship between a dominant operating-system platform and downstream products.

A central competition concern was that control over an important technological interface could disadvantage competitors in adjacent markets.

Relevance to AI epistemic scaffolding

AI ecosystems increasingly depend upon:

• APIs;
• model interfaces;
• retrieval systems;
• tool protocols;
• identity systems;
• data connectors;
• application ecosystems.

Control over those interfaces can create epistemic interoperability barriers.

For example, if an AI platform allows its own applications to access:

• richer search results;
• proprietary knowledge graphs;
• user history;
• enterprise databases;
• reasoning tools;

while imposing materially inferior access conditions on rival AI systems, competition authorities could examine whether interoperability restrictions reinforce dominance.

Broader principle

Control of an interface can become control of competition in downstream markets.

Case 5: Google Search (Shopping)

Principle

The European Commission and EU courts examined Google's treatment of its comparison-shopping service within general search results.

The case concerned the competitive significance of ranking and visibility in a digital ecosystem.

AI epistemic relevance

AI assistants increasingly perform the equivalent of ranking.

Instead of presenting ten blue links, an AI system may provide:

“Here are the three most relevant options.”

That creates a new form of competitive gatekeeping.

The AI system can potentially determine:

• which business is mentioned;
• which source is cited;
• which product is recommended;
• which legal authority is emphasized;
• which medical information is surfaced;
• which competitor is omitted.

Thus:

Search ranking → AI recommendation ranking

can create a comparable competition concern.

Key issue

The important question is not merely whether the AI output is inaccurate.

Competition law may instead ask whether a dominant platform systematically uses control over the knowledge interface to advantage its own services or disadvantage competing services.

Case 6: Google Android

Principle

The European Commission's Android decision addressed contractual and ecosystem practices involving Google's dominant mobile operating-system position and related services.

The case illustrates how dominance in one layer of a technological ecosystem can be used to strengthen positions in adjacent markets.

AI relevance

AI ecosystems are similarly vertically integrated.

A single company may simultaneously operate:

• cloud infrastructure;
• chips;
• foundation models;
• app stores;
• search;
• browsers;
• productivity software;
• advertising;
• AI assistants.

This creates opportunities for cross-layer leveraging.

For example:

Cloud dominance

↓

preferential AI-compute conditions

↓

Foundation-model advantage

↓

preferential assistant distribution

↓

Knowledge-interface dominance

↓

Downstream ecosystem dependency

The Android reasoning therefore provides a useful analytical framework for studying multi-layer AI ecosystems.

Case 7: United States v. Microsoft Corp.

Principle

The US Microsoft litigation examined the use of dominance in one technological market to protect and extend market power in an adjacent market.

A major concern involved the strategic use of control over the operating-system ecosystem.

AI relevance

AI epistemic scaffolding can similarly become a distribution bottleneck.

An AI assistant embedded into:

• an operating system;
• browser;
• search engine;
• office suite;
• smartphone;
• cloud platform;

may receive substantial distribution advantages unavailable to independent competitors.

This creates a potential distinction between:

competition to build the best AI

and

competition to obtain access to users' attention and workflows.

A technically superior rival may still struggle if the dominant ecosystem controls the principal interface through which users interact with knowledge.

Case 8: Google Search (AdSense)

Principle

The Google AdSense proceedings examined contractual restrictions affecting competition in online advertising and Google's position within an interconnected digital ecosystem.

AI relevance

AI-generated answers may become increasingly connected with:

• advertising;
• product recommendations;
• commercial search;
• shopping;
• travel;
• financial services;
• employment;
• professional services.

If the same company controls both:

epistemic intermediation

and

commercial monetization,

it may have incentives to manipulate the knowledge layer to reinforce commercial positions.

For example, an AI system might systematically privilege its own:

• shopping service;
• advertising marketplace;
• cloud service;
• financial product;
• professional platform.

This creates a potential knowledge-to-commerce leveraging theory.

6. Forms of AI Knowledge Dependency

A. Data Dependency

Rivals may depend upon access to:

• proprietary datasets;
• user-generated information;
• search indexes;
• transaction data;
• expert annotations;
• feedback data.

A dominant AI firm could potentially create competitive advantages by controlling unique datasets.

B. Retrieval Dependency

An AI model may depend upon a dominant retrieval infrastructure.

For example:

AI model → dominant search index → answer

If competitors cannot obtain comparable retrieval results, they may suffer a structural disadvantage even if their underlying models are competitive.

C. Model Dependency

Downstream firms may become dependent on a small number of foundation models.

This could occur where AI applications rely upon one provider for:

• inference;
• fine-tuning;
• embeddings;
• moderation;
• reasoning;
• multimodal capabilities.

Switching may become difficult because applications become architecturally optimized for one model.

D. Interface Dependency

Users may gradually stop interacting directly with multiple information providers.

Instead:

User → AI assistant → information ecosystem

The assistant becomes the gatekeeper of attention.

This can make inclusion in the AI's answer economically important.

7. Epistemic Ranking as a Competition Problem

Traditional search engines rank websites.

AI systems increasingly rank claims, sources, products and interpretations.

This creates a potentially more powerful form of gatekeeping.

Traditional model

100 sources → ranked results → user chooses

AI model

100 sources → AI synthesis → 3 conclusions → user relies on synthesis

The second structure gives the intermediary greater influence over information visibility.

Potential competition issues include:

• self-preferencing;
• source demotion;
• competitor exclusion;
• selective citation;
• discriminatory retrieval;
• preferential training;
• preferential recommendation;
• suppression of rival information services.

8. Knowledge Lock-In

Knowledge dependency may also create switching costs.

Users can become dependent upon:

• personalized prompts;
• historical conversations;
• proprietary workflows;
• accumulated context;
• customized agents;
• stored enterprise knowledge;
• proprietary plugins;
• model-specific APIs.

A firm may therefore face significant costs when switching to another AI provider.

Result

Switching cost → reduced multi-homing → stronger network effects → greater entry barriers

This resembles traditional platform lock-in but operates at the knowledge and cognitive-workflow level.

9. AI and Self-Preferencing

Self-preferencing could occur where a dominant AI intermediary gives preferential treatment to its affiliated services.

For example:

The competition concern becomes stronger where the AI system is the primary gateway to consumer or professional knowledge.

10. Knowledge Asymmetry

AI systems can create substantial information asymmetry between the platform and its users.

The platform may know:

• which sources were considered;
• which sources were rejected;
• how rankings were produced;
• what users searched for;
• which recommendations converted;
• which competitors were frequently omitted.

Users may see only the final answer.

This creates a potential epistemic opacity problem.

From a competition perspective, opacity may make it difficult for rivals and users to determine whether disadvantage results from:

• legitimate quality differences;
• algorithmic optimization;
• discriminatory treatment;
• self-preferencing;
• contractual restrictions;
• deliberate exclusion.

11. Tacit Coordination Through AI Knowledge Systems

Knowledge infrastructure can also facilitate coordination.

AI systems may simultaneously process:

• competitor prices;
• demand data;
• market forecasts;
• inventory;
• consumer behaviour.

If competing firms rely upon common AI systems or common algorithmic providers, there can be risks of:

• parallel pricing;
• synchronized recommendations;
• common forecasting;
• standardized terms;
• reduced strategic uncertainty.

The competition concern is particularly significant when firms use a common intermediary that observes large quantities of competitively sensitive information.

12. AI Epistemic Dependency and Essential Facilities

The essential-facilities doctrine should be applied cautiously.

Not every successful AI system is an essential facility.

A stronger case would require evidence that:

1. the resource is genuinely indispensable;
2. competitors cannot reasonably reproduce it;
3. access is necessary for viable competition;
4. denial has substantial exclusionary effects;
5. sharing does not undermine legitimate investment incentives;
6. appropriate access conditions can realistically be defined.

This is why Bronner and IMS Health are particularly important analogies.

13. Leveraging Across AI Markets

AI ecosystems make leveraging especially important.

A firm could possess market power in:

Cloud computing

and leverage it into:

AI compute → foundation models → AI assistants → enterprise software → knowledge services.

Similarly:

Search

could potentially be leveraged into:

AI retrieval → AI answers → commercial recommendations.

Or:

Operating system

could potentially be leveraged into:

default AI assistant → user data → model improvement → downstream AI services.

The competition-law inquiry should therefore examine ecosystem architecture, rather than looking at each AI product in isolation.

14. Merger and Acquisition Concerns

AI epistemic dependency is also relevant to merger review.

A transaction may combine:

• a major AI model;
• a dominant search engine;
• a large data provider;
• a professional information database;
• a cloud provider;
• an AI distribution platform.

Even if the parties operate in apparently different markets, the merger could increase control over the knowledge supply chain.

Authorities may therefore examine:

Horizontal effects

Whether two competing AI providers are combined.

Vertical effects

Whether a model provider acquires a critical data or distribution supplier.

Conglomerate effects

Whether complementary services can be bundled.

Data effects

Whether combining datasets strengthens entry barriers.

Ecosystem effects

Whether the combined entity can control several sequential AI layers.

15. Remedies

Potential competition remedies could include:

1. Interoperability

Require technically meaningful interoperability between competing AI systems.

2. Data portability

Allow users to transfer relevant data and conversational history.

3. API access

Provide non-discriminatory access under defined circumstances.

4. Non-discrimination

Prevent discriminatory treatment of rival AI providers.

5. Ranking transparency

Require meaningful explanations of commercially significant ranking mechanisms.

6. Separation

In exceptional circumstances, structural separation between infrastructure and downstream services may be considered.

7. Data-use restrictions

Limit combining data collected in one market with another where that combination creates exclusionary effects.

8. Switching mechanisms

Enable users to transfer workflows, context and knowledge repositories.

16. Analytical Framework for Courts and Competition Authorities

A useful framework is:

Step 1 — Define the relevant market

Possible markets include:

• AI foundation models;
• AI assistants;
• AI search;
• enterprise AI;
• AI knowledge services;
• professional information services;
• AI infrastructure.

Step 2 — Identify the epistemic bottleneck

Determine whether the bottleneck is:

data → retrieval → model → interface → workflow → distribution.

Step 3 — Measure dependency

Consider:

• switching costs;
• multi-homing;
• interoperability;
• alternative suppliers;
• data portability;
• technical compatibility.

Step 4 — Determine market power

Examine:

• market shares;
• entry barriers;
• data advantages;
• compute advantages;
• distribution;
• network effects;
• ecosystem integration.

Step 5 — Identify conduct

Potential conduct includes:

• refusal to supply;
• tying;
• bundling;
• exclusive dealing;
• self-preferencing;
• discriminatory access;
• interoperability restrictions;
• data foreclosure.

Step 6 — Establish competitive effects

Ask whether the conduct:

• excludes rivals;
• raises their costs;
• prevents entry;
• reduces innovation;
• increases switching costs;
• reduces consumer choice;
• entrenches dominance.

Step 7 — Consider objective justification

Possible justifications include:

• privacy;
• security;
• cybersecurity;
• intellectual-property protection;
• quality control;
• reliability;
• prevention of model abuse;
• legitimate technical constraints.

17. Distinguishing Legitimate AI Integration from Anticompetitive Dependency

Not every form of dependency is unlawful.

A successful AI provider may legitimately benefit from:

• superior technology;
• better training;
• higher-quality data;
• better infrastructure;
• lower costs;
• innovation;
• stronger security;
• better user experience.

The competition concern emerges where market power is used to create or protect dependency through exclusionary conduct rather than through competition on the merits.

This distinction is crucial.

18. Key Doctrinal Connections

19. Overall Legal Significance

The most important development is that AI competition may increasingly concern control over the pathway to knowledge, rather than simply control over a conventional product.

Earlier digital markets frequently involved control over:

search → distribution → advertising

AI can introduce a deeper layer:

search → synthesis → interpretation → recommendation → decision

Consequently, a dominant AI platform could potentially become not merely a distributor of information but an epistemic intermediary.

The relevant competition-law questions therefore include:

1. Who controls the underlying knowledge inputs?
2. Who controls access to them?
3. Who determines which information is retrieved?
4. Who determines how competing information is ranked?
5. Who controls the AI interface?
6. Can users easily switch?
7. Can rival AI systems obtain comparable inputs?
8. Can competitors interoperate?
9. Does the platform privilege its own downstream services?
10. Does control over knowledge become leverage into other markets?

20. Conclusion

AI epistemic scaffolding systems create a new dimension of competition-law dependency because they can mediate not only access to information but also the organization, interpretation and prioritization of information.

The most relevant precedents—Bronner, IMS Health, Magill, Microsoft, Google Shopping, Google Android and the US Microsoft litigation—provide different pieces of the analytical framework.

They collectively demonstrate several established principles:

• control over an important infrastructure can create competitive bottlenecks;
• information can constitute a strategically important competitive input;
• interoperability can be crucial to downstream competition;
• ranking and visibility can affect market access;
• dominance in one technological layer can potentially be leveraged into another;
• ecosystem integration can increase entry barriers;
• refusal to provide access is not automatically abusive and requires careful doctrinal analysis.

The distinctive feature of AI is that the bottleneck may ultimately become control over the user's route from information to understanding and from understanding to economic decision-making. That makes AI epistemic scaffolding particularly relevant to future analysis of dominance, essential facilities, interoperability, self-preferencing, data foreclosure, leveraging, platform dependency and AI ecosystem regulation.