AI On Trial As Competition Authorities Worldwide Zero In On AI Market

Discourse on the intersection of artificial intelligence (AI) and competition policy is expected to intensify this year. According to media reports, the Competition Commission of India (CCI) plans to study this intersection, representing a growing global trend, with competition regulatory bodies in the United Kingdom, the United States, and Australia also actively exploring the multifaceted challenges and opportunities AI presents. The Federal Trade Commission (FTC) in the United States plans to host a summit on AI later this month. As the discourse on AI further evolves, it becomes imperative to identify and address key themes, including market dynamics and the potential impact on innovation and consumer welfare.

Anti-competitive and pro-competitive nuances of AI markets

Foundational Models (FMs) are the building blocks of AI. They're trained on massive amounts of mostly public data (text, code, images) that can perform tasks like understanding language, generating text, or recognising objects. Large companies often have access to vast amounts of proprietary data and significant computing resources, creating a potential barrier for smaller competitors who may struggle to obtain similar datasets and infrastructure.

This advantage can result in a significant market power imbalance. However, neither data access nor computing power guarantees better or more successful products - for example, Open AI was largely built without proprietary data. Evidence also highlights that as AI technology continues to advance rapidly, the amount of required computing power and the cost of building and running AI models is falling.

FMs can be either closed-source or open-source. Generally, open-source models share their knowledge, including the code, architecture, and/or data for public use and collaboration, while closed-source models keep these blocks private and accessible only through a controlled interface. Each model can have pro and anti-competitive nuances. Often, companies keep high-performing FMs closed-source, thereby protecting the internal ‘knowledge’ of the model as a trade secret.

However, if access to FMs is restricted, it can limit competition by preventing smaller or newer players from accessing the tools necessary for developing competitive AI applications. Furthermore, without access to the code, detecting and addressing potential biases in the model can be challenging, leading to unfair outcomes for certain groups. On the other hand, closed-source models also exhibit pro-competitive behaviour. Companies investing heavily in research and development to create state-of-the-art models may drive technological advancements. When protected as proprietary technologies, these advancements can incentivise competition among companies to develop better models, facilitating innovation within the AI ecosystem.

Further, since the model’s development and operations involve only a specific company, consumers can easily attribute responsibility for any harm they face. Open-source models can democratise access to infrastructure and data for a wide range of contributors, thereby mitigating concentration with a few major players. This supports a more competitive landscape where various entities can participate and innovate. Further, the transparency of open-source models can lead to improved scrutiny and knowledge transfer.

Open-source models also compete directly with incumbents to drive down development costs. However, the open nature of the code may also expose it to misuse or exploitation, as malicious actors could identify vulnerabilities or design attacks easily. They may also be prone to misuse for social harms, including creating deepfakes that target individuals. Further, the involvement of multiple developers makes it difficult to ensure accountability, leading to an impact on the protection of consumer interests.

Impact of AI on competition in other markets

AI is empowering small entrants in various markets to compete with established market players by offering innovative service enhancements. For example, traditionally, established players in manufacturing have relied on manual inspection processes for quality control, which can be a cost-related barrier for smaller players. However, market entrants can leverage AI-powered computer vision systems on production lines to automate quality control checks. This can result in delivering higher-quality products, reducing operational costs, and optimising production processes.

However, deploying AI in specific contexts also raises concerns about potential anti-competitive effects. Currently, there is limited literature available on the use of AI to enable anti-competitive conduct. However, antitrust authorities worldwide have already dealt with cases concerning algorithmic automated collusion. This involves firms using autonomous algorithms to intentionally or unintentionally coordinate their actions, leading to anti-competitive outcomes such as price-fixing or market sharing.

In the ASUS case (AT.40465), the European Commission investigated several electronics manufacturers on suspicion of using pricing algorithms that led to similar pricing of their products across online retailers.

Meaningfully exploring the future of AI necessitates analysing the competitive landscape of FM markets. The potential disruptive impact of the technology on market dynamics in other sectors warrants attention. Regulators and policymakers will need to be mindful of both the anti-competitive and pro-competitive effects that are being exhibited. Market studies, like the one by the Competition and Markets Authority in the United Kingdom, can be an effective tool in this endeavour. However, these studies should involve multiple stakeholders, including technical experts, industry, investors, FM developers, civil society organisations, legal scholars and consumer bodies, analysing the impact of AI on competition. Countries should prioritise building the regulators’ capacity on AI’s technical nuances through workshops, open discussions and stakeholder consultations. These steps can ensure that the future of AI is characterised by healthy competition, inclusivity and fairness.

(The article is written by Saksham Malik, Senior Programme Manager and Bhoomika Agarwal, Senior Research Associate, at The Dialogue)

(The opinions presented belong solely to the authors.)