Press Release 14/2023
November 6, 2023

The AdC warns of competition risks in the Generative Artificial Intelligence sector and will not hesitate to intervene to guarantee the promotion and application of competition law in Portugal for the benefit of consumers whenever the identified risks materialize. This warning follows the analysis that the AdC has conducted on Competition and Generative Artificial Intelligence, which has resulted in the release of the Issues Paper.

Generative Artificial Intelligence

Generative Artificial Intelligence (AI) enables the production of new content, such as text or video, displaying capabilities typically considered exclusive to human beings. This Artificial Intelligence has already been used by millions of users since various services were made available to the public (e.g., ChatGPT, Stable Diffusion, or Copilot) and has the potential to bring numerous opportunities for innovation. For example, according to a Euroconsumers survey conducted in four countries, including Portugal, 75% of respondents aged between 18 and 34, and 29% of the 55 to 74 age group, said they had already tried ChatGPT.
Generative AI could have a significant and widespread impact on the economy. It is possible that it could contribute up to 7% of the world's GDP in ten years and be present in most businesses, such as customer service, education, scientific research, drug discovery, and entertainment.
Generative AI relies on a set of necessary requirements. Generative AI models require large volumes of data, computing power, and expertise. Therefore, a company that aims to develop and train a Generative AI base model will need access to these resources. If a company controls any of these resources, it may have the ability and incentive to exploit market bottlenecks, distorting competition.
Access to base models is also crucial for the development and implementation of Generative AI services available to consumers. The type of base model, its size and speed, and the pace of access to the base model are determining characteristics in the ability of Generative AI service providers to compete with each other.
An example of an extension of Generative AI functionalities available to end consumers is plugins. These can be likened to AI apps. For example, a plugin allows a user to ask the AI for an itinerary for a trip, specifying the date and place, and the AI responds with the itinerary, flights, and a link to the travel agency for making the reservation. ChatGPT, for example, already offers more than 1,000 plugins, with the first ones developed by Expedia, FiscalNote, Instacart, KAYAK, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier.

Competition and Generative AI

Generative AI is an emblematic example of the competition challenges already identified in the digital sector. There are risks of competitor exclusion strategies in cloud computing, hardware, and generative AI model markets. It is important to ensure that companies' behavior does not unduly distort competition, jeopardizing the full innovation potential of Generative AI. On the contrary, an environment of innovation, contestability, and competition allows companies, including startups, to enter the market, offering new services to the benefit of consumers and the economy.
Considering the experience accumulated in the digital sector by competition authorities and the risks of competition, it is possible to identify some of the elements determining the degree of competition in the Generative AI sector.

The focus of competition defense in the context of Generative Artificial Intelligence will necessarily have to be on the following cornerstones:

• Access to data
• Access to cloud computing or specialized hardware
• Access to Generative AI base models

The AdC, within the scope of its powers and in the context of international cooperation, will remain vigilant regarding developments in artificial intelligence and will not hesitate to intervene to ensure the promotion and enforcement of competition law in Portugal, for the benefit of consumers, whenever the identified risks materialize."