The Conundrum of Inter-Substitutability: Human Services vs AI Services in the Relevant Product Market

Introduction

With the rapid evolution of technology, tasks once seen as the exclusive domain of human intelligence are now being performed by artificial intelligence (“AI”) within mere seconds. With this, AI has become a focal point of legal scholarship, particularly in discussions on how to effectively integrate this new and powerful force within the legal and regulatory frameworks. Traditionally, Indian antitrust authorities have been reactive in nature, observing changes in the market and formulating rules accordingly. However, given the pace of progress in the AI sector, it is worth questioning whether competition regulators should adopt a more preventive approach in addressing the challenges posed by AI, rather than responding only after significant market transformations have already occurred.

In almost every antitrust inquiry, one constant remains: the definition of the relevant product market. This is far from an academic classification. This step forms the foundation upon which assessments of market power, dominance, and anti-competitive behaviour are scrutinised. The repercussions of misdefining a relevant product market can be detrimental. A narrowly relevant product market may artificially inflate a firm’s market power,  making pro-competitive behaviour appear anti-competitive, which might lead to potential regulatory overreach. Conversely, an overly broad relevant product market may dilute real concerns of monopolisation since a dominant firm may not appear to hold a lot of market power or share than it actually does, allowing anti-competitive behaviour to go unnoticed. These implications would extend to merger control, abuse of dominance inquiries and even something like imposition of remedies. Hence, defining relevant product markets is not just a starting point; it fundamentally shapes the outcome of any competition law investigation.

Under the Indian Competition Act, 2002, the relevant product market is defined as “market comprising all those products or services which are regarded as interchangeable or substitutable by the consumer, by reason of characteristics of the products or services, their prices and intended use”.

Specifically, in the context of services, a new dimension of complexity emerges. When services traditionally provided by humans are increasingly being replaced or supplemented by progressive AI-driven solutions, this boundary between Human Intelligence (HI) and AI becomes increasingly blurred, and the question of whether these services, irrespective of being from two distinct sources, can be put into the same relevant product market arises.

Before the authors delve into the intricacies, we wish to define what HI is. Herein, it refers to the inherent human ability to learn from experience, adapt to situational nuances, and comprehend and manage abstract concepts; capabilities that, in many scenarios, traditional technological or digital systems have historically struggled to replicate.

To this end, the authors purposely wish to solve this conundrum by discussing three major prongs in this article. Firstly, whether all AI services should be grouped within a single market, or should AI be treated as heterogeneous in nature; secondly, can AI services and human services truly be considered interchangeable for the purposes of market definition, and thirdly, proposing a roadmap as to how antitrust regulators can solve this issue.

Functional Differentiation in AI Systems for Market Analysis

Before delving into the comparison between human and AI services, it is critical to examine how we classify AI systems internally. Even at a preliminary level, without looking into the technical architecture of specific AI systems, it is evident that the technological ecosystem consists of different categories of AI models. For instance, generative text models such as ChatGPT, image-generation tools like Midjourney, and data analytics engines such as IBM Watson are designed to perform fundamentally different functions.

Companies have developed multiple foundation models, each tailored to a different “modality” (text, images, video, code). For example, OpenAI has its GPT series, DALL·E, CLIP, Codex, and Sora, while Google offers models like Gemini, ViT, and AudioLM. Each of these models caters to a distinct consumer base. A business seeking automated visual content would not substitute an image generation model with a text-based chatbot. Thus, consumer substitutability, which is the cornerstone of defining a relevant product market, would break down across AI categories.

This position finds support in the European Union’s AI Act, 2024, which is the first legal framework on artificial intelligence. The Act classifies AI systems into four different risk levels: These risks are categorised on the basis of the potential harm or disruption they may cause to individuals, society and the lawful authorities. Therefore, if regulators are already acknowledging functional and risk-based distinctions in governance, antitrust authorities must similarly reflect such distinctions in market definition. Treating AI as a monolithic market would not only oversimplify but also undermine the accuracy of market concentration assessments.

Even within the Indian context, there have been concrete efforts by the Indian Authorities to recognise and assess the implications of AI for market analysis. Notably, NITI Aayog’s discussion paper has discussed the recent shifts in AI and the transformative potential of AI across different market sectors. Furthermore, the 2024 Report of the Committee on Digital Competition Law by the Ministry of Corporate Affairs explicitly acknowledged “the recent revolutionary developments” in AI, underscoring the urgency and importance of considering AI  within the framework of Indian Competition Law.

Accordingly, AI systems must be classified into distinct relevant markets in antitrust issues based on their intended use, consumer base, and technological function. This approach would ensure a more precise and forward-looking regulatory response to the evolving dynamics between AI and competition law.

 Can Artificial Intelligence Effectively Replicate Human Intelligence?

Moving forward with the core of the issue, is the contention of whether AI can even effectively replace HI to such an extent, to even consider putting them in the same relevant product market.

To answer this question, the concept of asymmetric substitutability comes into focus. It is a situation where consumers are willing to “trade up” for a more efficient or technologically advanced solution, but are reluctant to “trade down” to a simpler alternative. While it is preferable to separate the relevant product market if they are not perfect substitutes, there can be situations wherein one product may replace another product, whilst the reverse may not hold. Thus, this may result in a situation wherein one product may be in the market for another product, even if the latter is not in the former’s market.

This framework can be applicable in evaluating the evolving interaction between AI and HI. We present this argument with the simple reasoning that even if a product is more advanced than its substitutes, it is still considered as part of the same market as its lesser counterparts in the early stages of competitions. When addressing asymmetric substitutability, certain factors such as relevant product market, user expectations, focus products, and the areas of overlap must be considered, especially in the evolving interplay between human and artificial intelligence. Furthermore, according to Section 19(7)(a) and 19(7)(c ) of the Competition Act of India, the end use of products and user preferences have to be taken into account while declaring the relevant product market.

The focus product, in this particular regard, would be the main task or service that can be performed by either HI or AI. The areas of overlap here would refer to the functions wherein both AI and HI offer comparable results to the users, thereby creating an essence of substitutability. In practice, such factors would hold paramount importance while judging the relevant product market, especially in unique conditions like these.

A practical illustration of the increasing AI functionalities can be found in the medical services platform Practo. Founded in 2008, the Bangalore startup has gained significant traction since the Pandemic because of the shift to virtual mode. It efficiently utilises AI to assess symptoms, location and user behaviour to recommend the most suitable doctor, while simultaneously acting as a virtual triage tool, and decreasing administrative load by scheduling appointments and billing.

Currently, Practo fills the critical gaps that exist in the Indian healthcare system. However, its increased usage of AI raises concerns of a potential displacement of human labour. Clinics may no longer require humans to schedule appointments. The number of nurses who perform initial triage may decrease. With medicines already being available online by big pharmaceuticals such as Apollo, and prescriptions being provided digitally, the number of small pharmacies that currently exist may decrease. Despite that, complete substitution remains unlikely due to multiple factors such as unequal digital access and technological limitations. Thus, the very concept of asymmetric substitution may come into play in such a situation.

The case of France Télécom v Commission becomes particularly instructive in this regard, wherein the commission observed that despite the occurrence of technological differences, the existence of two services commanding similar or comparable consumer migration patterns indicate that they can be considered as substitutes of each other in the same relevant product market as long as there is “no extremely asymmetric” nature of substitution. This judgment can act as a ground to underscore the evolving relationship between HI and AI.

While AI has made significant advances outperforming humans in tasks like data analysis, summarisation, and predictive modelling, it still lacks critical dimensions of HI, including emotional sensitivity and adaptability. To explain this dichotomy, the example of the law tech startup CaseMine can be taken in comparison to lawyers. The AI of this particular application can generate case summaries, predict possible outcomes, and scan documents, all within a few seconds. This considerably reduces the need for paralegals, while simultaneously reducing the research time for lawyers. However, it is simply a research tool. It cannot argue in court, cannot advise the clients in an emotional capacity, negotiate on their behalf, or adapt strategies situationally. underscoring the necessity of HI in contexts where empathy, persuasion, and strategic adaptability are essential. It, thus, cannot completely replace lawyers.

Perusing Indian jurisprudence and acknowledging the relatively nascent depth in  Indian competition Law, the authors wish to highlight a particularly interesting case of House of Diagnostics LLP v. Esaote Spa. In this case, the Competition Commission of India (CCI) had differentiated between two separate markets on the reasoning that one of the products had unique technological features and a better end-user experience. However, the CCI chairman, in his dissent, rejected this market fragmentation. He, instead, relied on comparable consumer migration patterns in his explanation to consider the products in question as part of the same relevant market.

Now, what makes this dissent especially compelling is its consistency with the reasoning that has earlier been adopted by another judgment, by the name of Sonam Sharma v. Apple.  In this case, it was declared that although the Apple iPhone has certain unique features and technological advancements as compared to other smartphones, it would still be part of a larger smartphone market, and not a separate one. Thus, one can see a clear disparity between the two cases with regard to the definition of a separate market. While the two judgments are divided in their opinion as to whether there should be separate markets based on technological differences, the previously mentioned dissent puts forth the idea of a common market despite technological differences, which aligns with the arguments that the authors wish to put forward.

There is no denial that an evident asymmetry between HI and AI exists. However, the complete replacement of the human element still remains a distinct prospect. Admittedly, there may come a time wherein AI would overtake humans, if one is to trace the speed of its development. However, until a complete replacement becomes a possibility, HI and AI must be considered as the players of the same relevant product market.

Conclusion & Potential Roadmap

Having established both the internal diversity of AI systems and their nuanced, asymmetric substitutability with HI, the next step forward would be to chart a practical roadmap for Antitrust authorities. The rapid pace and unpredictability of AI innovation mean that a purely retrospective, “wait‑and‑see” approach will no longer suffice. Instead, regulators must embrace an ex‑ante framework, one that tries to foresee shifts in market power before they harden into entrenched positions.

Firstly, as established, it is essential to recognise that AI is not a single, homogenous entity but a constellation of distinct technologies, each solving different problems for different users. By moving away from the temptation to lump all AI services into one broad category, authorities can apply a function‑based lens that reflects real‑world usage and end‑user needs. To suggest a practical manifestation of this approach, let us take the product differentiation example between LLMs (large language models) such as ChatGPT, which is a tool to generate, process and interact with natural language, effectively translating itself into the education, content generation, and consumer service sectors. Here, we can look at the value proposition that can be considered, which is not merely computation but communication, i.e., mimicking or augmenting human language capabilities. Therefore, in terms of substitutability, one might look at human content writers or even search engines. The definitive and core feature that can be looked upon in this relevant product market would be language-based interactions and the spectrum or services that come along with it.

In comparison, if we look at AI systems that are used in security profiling or surveillance such as biometric authentication and behaviour prediction algorithms, the fundamental function of these types of AI systems is different as they are not built to converse or generate text but to identify threats or anticipate risks based on biometric or behavioral data. Their value-based proposition is not judged on creativity or quantitative measures but rather on accuracy, precision, and the minimisation of false positives. The market here is not communicative assistance but risk management, for these types of AI systems substitutes may include human security personnel, traditional surveillance systems, or even simpler digital ID verification tools, depending on the sophistication of the AI.

Secondly, the classic notion of two-way substitutability must give way to a model that captures asymmetric consumer behaviour. Regulators must, therefore, develop a test or framework that takes into account unique migration factors such as user dependency, long-term cost benefits, and how seamless the switch is between AI services and HI services in the same relevant product market.

Yet another practical implementation of the asymmetrical model existing between AI and HI is the field of self-driving cars. The concept of self-driving cars, like many AI applications, is still evolving, but the market has already shown immense potential for growth. A notable example is Waymo Taxi Service, a self-driving taxi service operating in multiple American cities. It stands out as one of the few successful business models in this segment. According to Yipitdata, Waymo’s gross bookings from August of 2023 to April of the current year have surpassed Lyft, the US’s second-largest taxi-hailing service, in San Francisco. Given that this is still a relatively early stage of competition in the self-driving taxi industry, these services compete for the same pool of consumers that human driven taxis do, despite having several differences such as maintenance & upkeep (self-driving taxis may require constant system monitoring, software updates issues, etc., which would not be faced by human driven taxis). Nevertheless, they would likely be considered as part of the same relevant product market, at least in this nascent phase. This is because, despite their technological advantage, there is still time for a complete replacement of human drivers by self-driving cars. Therefore, this coexistence highlights asymmetric substitutability within a shared product market between an AI application and HI.

To sum up, competition law must evolve in lockstep with the technologies it governs as AI continues to rapidly advance day by day, presenting new challenges to traditional market structures and existing ideas of substitutability. By adopting a dynamic, context‑sensitive approach that is grounded in function, behaviour, and practical realities, antitrust regulators will be better equipped to safeguard competition in this Artificial Intelligence Era.

*The authors are Second-Year students at the Rajiv Gandhi National University of Law, Punjab