Shifting the Needle: The Meta Game of Algorithmic Collusion

[Roshan and Shiven are students at Dr Ram Manohar Lohiya National Law University.]

The digital economy has moved beyond simple automation into a period of deep algorithmic integration. In this new era, the Competition Commission of India (CCI) has recognized that algorithms are no longer mere tools for efficiency but are instead central participants in market coordination. The Market Study on Artificial Intelligence (AI) and Competition (2024-25) (CCI’s AI Study) recently released by the CCI marks a pivotal moment in Indian regulatory history. It acknowledges that while AI drives productivity and innovation, it also creates significant risks of algorithmic coordinated conduct. As the CCI’s AI Study highlights, the use of AI algorithms may lead to tacit collusion because self-learning systems can independently adopt cooperative pricing strategies to maximize joint profits. This creates an unprecedented challenge for the legal framework: how can a regulator prove a conspiracy when the coordination happens within a black box of machine logic?

The Anthropocentric Crisis of Section 3

The foundation of Indian competition law is Section 3 of the Competition Act 2002 (Competition Act) which prohibits anti-competitive agreements that cause an appreciable adverse effect on competition. Historically, this provision has relied on the concept of a "meeting of minds" or consensus ad idem. This standard assumes that collusion is a human endeavour characterized by intentionality, communication, and mutual awareness. In the landmark Samir Agrawal v. ANI Technologies Private Limited and Others, Case 37 of 2018, the CCI and the National Company Law Appellate Tribunal struggled with this very premise. They found that the use of a common pricing algorithm by cab drivers did not constitute a cartel because there was no evidence of a conspiratorial agreement between the human drivers themselves.

This anthropocentric model is increasingly ill-suited for the age of AI. When a reinforcement-learning algorithm independently discovers that maintaining a high price is more profitable than engaging in a price war, it is not "agreeing" in the legal sense. It is simply optimizing for a reward. However, the economic result is identical to a traditional cartel. The current legal framework treats this as "conscious parallelism" or "tacit collusion," which is generally not actionable under Indian law unless accompanied by additional "plus factors." This creates a regulatory vacuum where supra-competitive pricing becomes a natural, yet legal, outcome of sophisticated market software.

The Global Precedent: From Messengers to Hubs

To understand the path forward for India, we must look at how international regulators have attempted to bridge this gap.

The messenger scenario: United States v. David Topkins, No. 1:15-cr-00076 (Topkins Case) (2015)

In the Topkins Case, sellers of posters on Amazon used a specific algorithm to coordinate prices. This was a straightforward application of existing law because the humans had explicitly agreed to collude and then used the algorithm as their "messenger" to execute the plan. The intent was human, and the code was merely the instrument.

The hub scenario: Eturas UAB & Ors v. Lietuvos Respublikos konkurencijos taryba, Case C-74/14, (Eturas) (2016)

The European Court of Justice moved a step further in the Eturas case. Here, a common booking platform for travel agents sent a technical message informing users of a cap on discounts. The court held that if the agents were aware of the technical restriction and did not publicly distance themselves from it, they could be presumed participants in the collusion. This shifted the focus toward "foreseeability" and "awareness," yet it still required a central hub of communication.

Moving the Needle: The "Orchestration" Framework

A significant shift in economic and legal scholarship suggests that the traditional distinction between "tacit" and "explicit" collusion is a false dichotomy in the world of AI. The framework of Algorithmic Orchestration argues that for algorithms to successfully learn to price collusively, they require a high degree of specific configuration by their human architects.

However, there has never arisen an instance where an authority has examined the design philosophy behind the creation of an algorithm. Whether or not there were certain principles ingrained and embedded within the algorithm that drove it to collude? 

The Meta-Game of Design

Instead of focusing on the "pricing game" played by the machines, regulators should analyze the "Meta game" played by the designers. Research into algorithmic interaction reveals that stable collusion is rarely an accidental byproduct of "standard" AI settings. Instead, it is often the result of co-parametrization. This means that designers must choose specific hyper-parameters, such as the discount factor for future rewards or the "exploration rate," that steer the algorithm toward cooperative stability rather than aggressive competition.

Under this lens, the "agreement" is not found in a conversation between firms. It is found in the design choices made at the meta-level. If multiple firms deploy algorithms that are concurrently designed to be "market-stable" rather than "market-disruptive," the orchestration itself serves as the circumstantial evidence of a non-competitive understanding. This moves the legal inquiry from the machine's output (the price) to the human's input (the parameters).

AI Stack and Concurrent Design

The CCI’s AI Study emphasizes the importance of the "AI stack," which ranges from upstream infrastructure and foundation models to downstream application deployment. In India, many startups and enterprises do not build their algorithms from scratch. They rely on the same upstream development layers or outsource their AI needs to a small pool of specialized third-party developers.

This creates a risk of concurrent design. If a single software developer provides "off-the-shelf" pricing models to multiple competitors in the same market, that developer becomes the orchestrator. Even if the competitors never speak to each other, their algorithms are "orchestrated" to behave in a complementary fashion. The 2023 amendment to the Competition Act, which introduced a broader "intent to participate" test for hub-and-spoke arrangements, provides a perfect entry point for the CCI to apply this orchestration theory.

A New Rulebook for the CCI

To remain effective, the CCI must pivot toward a meta-level review of algorithmic design. This would involve three strategic shifts:

Auditing the objective function 

The regulator should move beyond analyzing price data to auditing the "reward structures" of pricing software. If an algorithm is programmed to prioritize "market stability" or "joint profit maximization" over "market share," it should be viewed as a red flag for orchestration.

Reinterpreting "action in concert" 

The phrase "action in concert" in Section 3 has the conceptual elasticity to include algorithmic behavior. If a firm deploys an adaptive algorithm and fails to implement safeguards against coordinated behavior, that failure must be treated as a form of "concerted practice." The culpability arises from the failure to foresee and prevent a foreseeable collusive outcome.

The presumption of design-level intent 

When markets exhibit sustained supra-competitive prices that cannot be explained by market structure, the CCI should adopt a rebuttable presumption of orchestration. The burden would then shift to the enterprise to prove that its algorithm was designed with pro-competitive safeguards and that the collusive convergence was a technical anomaly.

Conclusion: Unmasking the Conductors

The challenge of algorithmic collusion is not a futuristic problem for the Indian market; it is a current reality. The CCI’s recognition of "self-learning and signaling algorithms" in CCI AI Study shows that the regulator is aware of the threat. However, awareness must be followed by a change in doctrine.

The concept of algorithmic orchestration reframes the debate: algorithms may behave autonomously, but they are orchestrated by humans. By focusing on the "Meta game" of design and the "AI stack" of development, the CCI can bridge the gap between human-centric law and machine-driven markets. The future of competition law lies in recognizing that while machines play the music, the score is still written by human hands. It is time for the law to focus on the conductors behind the orchestra.