Emergent Behavior Without Assignable Intent in Markets

1. Introduction

Emergent behavior without assignable intent describes situations in which market outcomes arise from the interaction of many independent decisions, algorithms, platforms, firms, or consumers, even though no single participant can readily be shown to have intended the resulting market outcome.

Examples include:

algorithmic price increases without an express cartel agreement;

parallel reductions in output;

automated matching producing market concentration;

platform algorithms independently converging on similar prices;

network effects producing a dominant ecosystem;

autonomous pricing systems learning that higher prices are collectively more profitable;

AI systems adapting to one another without direct human instructions.

The central competition-law question is:

Can competition law intervene when an anticompetitive market outcome emerges from individually autonomous conduct, but there is no clearly identifiable agreement or human intention to produce that outcome?

This is increasingly important because traditional competition law was developed around identifiable human conduct—agreements, decisions, communications, policies and deliberate exclusionary strategies—whereas algorithmic markets can produce outcomes through distributed and adaptive processes.

2. The Meaning of “Emergent Behaviour”

An emergent market outcome is one that is not necessarily programmed as a specific objective by any single participant but results from the interaction of multiple decisions.

A simplified example is:

Firm A algorithm
↓
observes Firm B
↓
changes price

Firm B algorithm
↓
observes Firm A
↓
changes price

Repeated thousands of times:

independent decisions → adaptation → convergence → persistent high prices

Neither company may have expressly instructed:

“Coordinate with the competitor.”

Nevertheless, the market may exhibit a coordinated outcome.

3. Intent and Competition Law

Competition law does not have one universal requirement of subjective intent.

Different doctrines use different standards.

For example:

Cartels

An agreement or concerted practice generally requires some form of coordination between undertakings.

Abuse of dominance

Certain forms of abusive conduct may be established without proving a subjective desire to eliminate competitors.

Merger control

Authorities may intervene because a transaction is likely to produce harmful competitive effects, even though the merging firms have legitimate business objectives.

Article 101 TFEU

The focus is principally on agreements, decisions and concerted practices rather than on whether executives subjectively intended a particular market outcome.

Therefore:

Absence of assignable intent does not automatically mean absence of competition-law relevance.

But it does make certain forms of liability considerably more difficult.

4. Emergence Versus Explicit Collusion

There is an important distinction.

Explicit collusion

Firm A tells Firm B:

“We will both maintain prices at ₹100.”

This creates an identifiable coordination mechanism.

Algorithmic emergence

Firm A's algorithm observes Firm B's price.

Firm B's algorithm observes Firm A's price.

Each independently learns that undercutting produces retaliation.

Eventually both maintain ₹100.

There may be:

no meeting;

no email;

no phone call;

no explicit agreement;

no individual instruction to coordinate.

The result nevertheless resembles coordinated pricing.

This creates an important attribution problem.

5. The Attribution Problem

Traditional enforcement asks:

Who did what?

Emergent systems require additional questions:

Who designed the algorithm?

Who selected the objective function?

Who supplied the data?

Who authorised deployment?

Who monitored the system?

Could the undertaking predict the outcome?

Could it intervene?

Did the undertaking benefit from the resulting behaviour?

Was the algorithm intentionally designed to facilitate coordination?

The answer may involve several actors rather than one identifiable decision-maker.

6. Case Law

1. Wood Pulp – Ahlström Osakeyhtiö v Commission

The Wood Pulp litigation is an important authority concerning parallel behaviour and concerted practices.

The European Court of Justice recognised that parallel conduct may constitute evidence of coordination, but parallel behaviour by itself does not automatically establish a concerted practice where there are plausible alternative explanations.

Relevance

This principle becomes particularly important in algorithmic markets.

If multiple pricing algorithms independently produce identical prices, the authority cannot simply assume:

identical outcome = illegal coordination.

It must examine whether the parallel conduct is sufficiently explained by autonomous market conditions or whether there is evidence of coordination.

Principle

Parallel conduct alone is not necessarily proof of concertation.

7. 2. Eturas v Lietuvos Respublikos konkurencijos taryba

Eturas is one of the most relevant cases for technologically mediated coordination.

An electronic travel-booking system implemented a mechanism affecting discounts available to participating travel agencies.

The Court of Justice considered whether knowledge of the electronic mechanism and continued participation could support an inference of concerted conduct.

Importance

Eturas demonstrates that competition law can operate where coordination is mediated through a technical system rather than traditional face-to-face communication.

However, the case also illustrates the importance of evidence connecting individual undertakings to the mechanism.

Principle

Technology can constitute the medium of coordination, but liability still requires legally sufficient evidence connecting the undertaking to the coordinated practice.

8. 3. T-Mobile Netherlands v Raad van Bestuur van de Nederlandse Mededingingsautoriteit

In T-Mobile Netherlands, the Court of Justice considered the concept of a concerted practice under Article 101.

The Court emphasised that even a single meeting can constitute coordination where it is capable of reducing strategic uncertainty between competitors.

Relevance

In algorithmic markets, strategic uncertainty can similarly be reduced through:

shared algorithmic signals;

public pricing mechanisms;

automated monitoring;

common software;

standardised pricing inputs.

The case demonstrates that competition law can focus on the exchange or reduction of strategic uncertainty, rather than requiring a formal written cartel agreement.

Principle

Coordination can arise from conduct that reduces competitors' uncertainty about future market behaviour.

9. 4. AC-Treuhand v Commission

AC-Treuhand is significant because the European Court of Justice accepted that an undertaking facilitating cartel activity can potentially fall within Article 101 even though it is not itself a conventional seller of the cartelised product.

Relevance

Modern algorithmic markets can involve:

pricing-software providers;

data intermediaries;

cloud providers;

AI developers;

optimisation vendors.

If a technology intermediary knowingly designs a system to facilitate anticompetitive coordination, its role may become legally significant.

Principle

Competition-law responsibility need not always be limited to the firms directly selling the affected product.

10. 5. United States v Apple Inc.

The Apple e-books litigation is relevant to the distinction between market outcomes and identifiable coordination.

The case demonstrated that technological and contractual structures can be used to implement coordinated changes in market conditions.

Relevance

An algorithm or platform cannot be treated as an autonomous legal shield where humans have deliberately designed the system to achieve an anticompetitive outcome.

Conversely, where no such coordination exists, authorities face a more difficult attribution question.

Principle

The technological implementation of commercial conduct does not remove the underlying competition-law analysis.

11. 6. Brooke Group v Brown & Williamson Tobacco Corp.

Brooke Group is an important US Supreme Court authority concerning predatory pricing.

The case is particularly relevant to the question of competitive effects versus subjective intention.

The Court required rigorous economic analysis of pricing conduct rather than relying simply upon allegations of exclusionary intent.

Relevance

An algorithm may independently reduce prices.

The mere fact that the outcome harms competitors does not automatically establish unlawful predation.

Authorities may need to establish the legally required economic conditions.

Principle

Antitrust liability generally requires more than observing a harmful outcome; the applicable doctrinal and economic requirements must be satisfied.

12. 7. Matsushita Electric Industrial Co. v Zenith Radio Corp.

Matsushita is another important US authority concerning alleged conspiracy.

The Supreme Court emphasised the importance of distinguishing legitimate independent conduct from coordinated anticompetitive behaviour.

Relevance

This is directly relevant to emergent algorithmic behaviour.

If several algorithms independently produce similar prices, investigators must distinguish:

independent optimisation

from

concerted conduct.

Principle

An inference of conspiracy must be supported by evidence that distinguishes unlawful coordination from rational independent conduct.

13. 8. Bell Atlantic Corp. v Twombly

Twombly concerned allegations of parallel conduct and conspiracy.

The Supreme Court held that parallel behaviour alone does not necessarily establish an antitrust conspiracy.

Relevance

This becomes increasingly important where AI systems produce highly similar outputs.

Suppose ten firms use independent algorithms and all produce similar prices.

The similarity may be evidence worth investigating, but it does not automatically establish an agreement.

Principle

Parallel outcomes are not necessarily proof of an agreement.

14. Emergent Behaviour and Tacit Coordination

The most difficult category lies between:

independent competition

and

explicit collusion.

This middle category is sometimes described as:

tacit coordination;

conscious parallelism;

coordinated effects;

algorithmic collusion;

autonomous coordination.

Algorithms can make this problem more severe because they can:

observe markets continuously;

react immediately;

remember previous interactions;

predict competitor responses;

punish deviations;

optimise long-term profits.

This can make coordination more stable even without direct communication.

15. The Difference Between Tacit Coordination and Concerted Practice

This distinction is essential.

Tacit coordination

Competitors independently recognise that aggressive competition may be undesirable and adapt their conduct accordingly.

Concerted practice

There is some form of coordination or communication that substitutes practical cooperation for competitive independence.

The former may be economically harmful but difficult to attack under traditional cartel provisions.

The latter may fall squarely within competition law.

16. Algorithmic Pricing Example

Consider three firms:

Firm A raises to ₹110.

Firm B's algorithm predicts that Firm A will maintain the higher price and raises to ₹110.

Firm C follows.

A then observes B and C and maintains ₹110.

After repeated iterations:

₹100 → ₹105 → ₹110 → ₹115

No executive has instructed:

“Form a cartel.”

Yet the market may move toward a stable supra-competitive equilibrium.

This is the essence of the emergent-behaviour problem.

17. Why Intent Is Difficult to Assign

AI systems may involve multiple layers:

Board

↓

Management

↓

Software engineers

↓

Model designer

↓

Training data

↓

Machine-learning model

↓

Pricing algorithm

↓

Market interaction

The resulting behaviour may not correspond directly to any single human decision.

An executive may have intended only:

“Maximise long-term profit.”

The algorithm might discover:

“Maintaining higher prices produces better long-term returns.”

Who intended the resulting coordination?

This is the intent attribution problem.

18. Objective Function Versus Emergent Outcome

A crucial distinction is between:

Designed behaviour

The system is explicitly programmed to:

maintain competitor prices.

This creates a strong attribution pathway.

Learned behaviour

The system is trained to:

maximise profit.

It independently discovers that matching competitors is optimal.

The second case is harder.

The company may argue:

“We never instructed the algorithm to coordinate.”

The authority may respond:

“You designed, deployed, monitored and benefited from the system.”

The legal significance of that distinction will increasingly depend on the applicable competition-law doctrine.

19. Responsibility Without Subjective Intent

Competition law may sometimes impose responsibility based upon:

objective conduct;

foreseeable effects;

knowledge;

participation;

control;

implementation;

economic effects.

Therefore, subjective intention is not always the decisive issue.

For example, under dominance doctrines, the question may be whether conduct objectively produces exclusionary effects rather than whether an executive privately desired competitor elimination.

Similarly, merger control is fundamentally prospective.

The authority does not need to prove that the merging companies subjectively intend to harm consumers.

20. Emergent Effects in Platform Markets

Platforms create particularly powerful emergence mechanisms.

Consider:

more users

↓

more data

↓

better recommendations

↓

more engagement

↓

more users

This feedback loop can produce dominance without a single decision to eliminate competitors.

Similarly:

more sellers

↓

more consumers

↓

more sellers

creates network effects.

The resulting market concentration may be an emergent structural phenomenon rather than the product of a single exclusionary act.

21. Does Emergence Itself Violate Competition Law?

Generally, not automatically.

A market may become concentrated because of:

innovation;

economies of scale;

network effects;

consumer preferences;

superior efficiency;

learning effects.

Competition law ordinarily does not punish success merely because dominance emerges.

The critical distinction is:

Lawful emergence

Superior product → adoption → network effects → dominance.

Potentially unlawful emergence

Exclusionary conduct → artificial network effects → foreclosure → dominance.

Therefore:

The existence of an emergent outcome is not itself sufficient to establish an infringement.

22. Evidentiary Problems

Emergent systems create difficult evidence questions.

Traditional evidence includes:

emails;

meetings;

contracts;

telephone records.

AI systems require additional evidence:

source code;

model architecture;

training data;

system logs;

parameter changes;

prompts;

reward functions;

deployment instructions;

API interactions.

Competition authorities may increasingly need technical expertise to understand whether an outcome was:

programmed;

learned;

accidental;

foreseeable;

predictable;

deliberately maintained.

23. The Role of Foreseeability

Foreseeability could become especially important.

Suppose a company knows that its pricing algorithm:

monitors competitors;

rapidly responds to their prices;

penalises aggressive deviations;

produces persistent high prices.

The company may have difficulty claiming that the market outcome was entirely unforeseeable.

This creates a continuum:

unforeseeable emergence → foreseeable emergence → monitored emergence → deliberately maintained emergence.

The further the conduct moves toward the right side, the stronger the potential attribution argument becomes.

24. Compliance and Governance Solutions

Companies using autonomous systems should establish:

Algorithmic competition audits

Test whether systems facilitate coordination.

Human oversight

Allow humans to intervene.

Parameter restrictions

Prevent systems from using competitor-specific information unnecessarily.

Independent validation

Review models for competition risks.

Documentation

Maintain records explaining:

objectives;

data;

constraints;

model updates.

Kill switches

Allow potentially harmful systems to be suspended rapidly.

25. Regulatory Approaches

Competition authorities could consider:

A. Behavioural monitoring

Identify persistent unexplained convergence.

B. Algorithmic audits

Examine whether pricing systems learn coordinated behaviour.

C. Information-exchange controls

Restrict competitively sensitive data flows.

D. Presumptions based on design

Where an algorithm is deliberately designed to facilitate coordination, authorities may have stronger evidence of responsibility.

E. Outcome-based investigation

Persistent supra-competitive outcomes could trigger investigation without automatically constituting liability.

26. A Proposed Attribution Framework

A useful analytical framework is:

Stage 1 — Identify the outcome

Is there:

price convergence?

reduced output?

market foreclosure?

exclusion?

discriminatory access?

Stage 2 — Identify the mechanism

Did the outcome arise through:

human decisions?

algorithmic interaction?

platform architecture?

network effects?

Stage 3 — Examine design

Was the system deliberately designed to produce the behaviour?

Stage 4 — Examine knowledge

Did the undertaking know or reasonably understand what was occurring?

Stage 5 — Examine control

Could the undertaking modify or stop the behaviour?

Stage 6 — Examine participation

Did the undertaking continue operating the system after discovering the effects?

Stage 7 — Apply the relevant competition-law doctrine

Only then should authorities determine whether the conduct constitutes:

agreement;

concerted practice;

abuse of dominance;

exclusionary conduct;

merger-related harm;

or no infringement.

27. The Central Legal Tension

Two competing principles must be maintained.

Principle 1: No automatic liability for independent conduct

Businesses should not be punished merely because independent algorithms happen to produce similar outcomes.

Principle 2: Technology cannot become a liability shield

Businesses should not avoid competition law simply by delegating decisions to algorithms.

The proper approach lies between these extremes.

28. Key Lessons from the Case Law

The authorities discussed above collectively establish several important propositions:

Parallel conduct does not automatically prove collusion — Wood Pulp, Matsushita, Twombly.

Technological mechanisms can facilitate concerted practices — Eturas.

Reducing strategic uncertainty can be legally significant — T-Mobile Netherlands.

Technology intermediaries can have competition-law significance — AC-Treuhand.

Harmful pricing outcomes require the applicable economic and legal conditions to be established — Brooke Group.

Technological implementation does not immunise coordinated commercial conduct — United States v Apple.

Emergent market concentration is not automatically unlawful; the underlying mechanism and competitive effects remain critical.

29. Conclusion

Emergent behaviour without assignable intent represents one of the most difficult problems for modern competition law.

Markets increasingly operate through systems in which:

human objective → algorithm → interaction → adaptation → collective outcome

The final outcome may not correspond neatly to the intention of any single actor.

Competition law therefore faces a fundamental challenge: it must distinguish between lawful independent adaptation and algorithmically facilitated coordination or exclusion.

The case law—from Wood Pulp, Eturas, T-Mobile Netherlands, AC-Treuhand, Matsushita, Twombly, Brooke Group, and Apple—provides the foundations, but modern autonomous systems create circumstances that traditional doctrine did not fully anticipate.

The most defensible approach is neither to impose liability merely because an undesirable market outcome emerged nor to treat algorithmic autonomy as a complete defence.

Instead, regulators should examine the entire causal chain:

design → deployment → information → algorithmic interaction → emergent behaviour → knowledge → control → competitive effect.

The future of competition law will increasingly depend on whether that chain can provide a legally sufficient basis for attributing market behaviour to an undertaking even when no single human decision-maker can be identified as having intended the final outcome.