Search

Begin New Search Topic Science and Technology Author Bertin Martens

1. How DeepSeek has changed artificial intelligence and what it means for Europe

Author:
Bertin Martens
Publication Date:
03-2025
Content Type:
Policy Brief
Institution:
Bruegel
Abstract:
By mid-2024, artificial intelligence large language models (LLMs) were running into diminishing returns to scale in training data and computational capacity. LLM training began to shift away from costly pre-training to cheaper fine-tuning and allowing LLMs to ‘reason’ for longer before replying to questions. Fine-tuning uses chain-of-thought (CoT) training data that includes questions and the logical steps to reach correct answers. This increases the efficiency of learning for smaller AI models, such as DeepSeek. CoT data can be extracted from large ‘teacher’ LLMs to train small ‘student’ models. These changes shift the cost structure of AI models from high pre-training costs to lower fine-tuning costs for model developers and more inference costs for users. While smaller models are cheaper to use, a positive AI demand effect is likely to exceed the negative price effect. Price competition between models will increase, resulting in tighter margins for AI firms. Specialised models can still fetch premium prices. Cheaper LLMs are an opportunity for European Union companies to catch up in building smaller AI models and applications on top of LLMs. Increased demand for AI services will require more investment in computing infrastructure, including in the EU. Investing in large LLMs and the corresponding hyperscale infrastructure is riskier, especially as price competition between models increases. Knowledge extraction between AI models puts pressure on model developers to protect their investments against free-riding by others. It also creates a dilemma for policymakers: should they favour free-riding to promote competition and innovation, or should they clamp down and reinforce monopoly rents to stimulate investment in AI models? Past policy will not be an appropriate response in a world that offers vastly expanded opportunities for knowledge pooling and innovation at lower cost.
Topic:
Science and Technology, European Union, Economic Policy, Microeconomics, Innovation, Artificial Intelligence, Large Language Models (LLMs), and DeepSeek
Political Geography:
Europe

2. Economic arguments in favour of reducing copyright protection for generative AI inputs and outputs

Author:
Bertin Martens
Publication Date:
04-2024
Content Type:
Working Paper
Institution:
Bruegel
Abstract:
Generative artificial intelligence (GenAI) models have stirred considerable controversy about copyright protection for AI training inputs and model outputs. The European Union’s AI Act will require model developers to be transparent about their use of training inputs such as text, images and music. The EU Copyright Directive allows free text and data mining of these media inputs unless copyright holders have opted out and want license payments. The right to opt-out amounts to economically inefficient overprotection of copyright. Free use of media content for GenAI training does not affect media sales to consumers. Optouts only strengthen the bargaining position of copyright holders, who decide depending on their private interests. That generates windfall profits without any increase in consumer surplus or social welfare. The licensing of training inputs reduces the quantity of data and the quality of GenAI models, creates transaction costs and reduces competition between GenAI firms. This slows down GenAI-induced innovation in media products and production processes, and productivity gains in all service sectors that apply GenAI. Ultimately, it slows down economic growth compared to what it could be with competitive and high-quality GenAI. Bargaining over license pricing is arbitrary as there is no objective revenue benchmark to start from. Defenders of the moral right to remuneration argue that any arbitrary remuneration is better than no remuneration. But this private moral right comes at the expense of social welfare. The ongoing bargaining and court cases between media producers and GenAI developers risk entrenching this market failure in jurisprudence. Early regulatory intervention and elimination of opt-outs for GenAI training, or weakening them in AI Act implementation guidelines, would solve this. There is no need for copyright on GenAI outputs either. GenAI reduces the marginal cost of machine-production of media outputs to close to zero, on par with the marginal cost of reproduction. That eliminates incentives for piracy. Moreover, composite human-machine outputs benefit from a de-facto extension of copyright on the human component.
Topic:
Science and Technology, Intellectual Property/Copyright, European Union, Regulation, Economic Growth, and Artificial Intelligence
Political Geography:
Europe

3. Why artificial intelligence is creating fundamental challenges for competition policy

Author:
Bertin Martens
Publication Date:
07-2024
Content Type:
Working Paper
Institution:
Bruegel
Abstract:
The rapidly evolving market for artificial-intelligence services is apparently thriving and very competitive, with a growing number of AI start-ups and ever larger and more capable AI models. Investors are pouring large amounts of money into start-ups and into a few big tech firms that have the large financial resources and computing capacity that constitute key inputs to the production and running of AI-driven services. However, exponentially growing AI model training costs create a market entry barrier for AI start-ups, forcing them to cooperate with big tech firms to get access to computing infrastructure and end users. At the same time, their AI models compete with those of the big tech firms. Competition authorities investigate these co-opetition agreements, looking for market-distorting clauses. Beyond computing costs, other competition choke points in the AI supply chain include access to dedicated AI processor chips and training data. Strict enforcement of copyright on data may further tighten an already scarce supply of affordable training data. Policymakers have few, if any, effective tools to deal with these potential competition bottlenecks in AI industries. Co-opetition agreements are necessary to enable AI start-ups to access hyperscale and costly computing infrastructure. At the same time, the agreements give big tech privileged access to the latest AI models. Competition authorities can try to minimise market-distorting provisions in these agreements but there are technical and economic limits to the extent of fragmentation in the AI value chain that can be meaningfully imposed on big tech firms. Licensing fees for copyright-protected data further tighten the already scarce supply of affordable data. It is hard to implement a fair market-clearing mechanism for dedicated AI processors. If model training costs continue to grow exponentially, as appears to be the case for the foreseeable future, the entire competition policy setting for AI industries may need a revision, with collaboration, including between big tech firms, dominating competition to keep the pace of AI innovation going.
Topic:
Science and Technology, Governance, European Union, Microeconomics, and Artificial Intelligence
Political Geography:
Europe

4. Catch-up with the US or prosper below the tech frontier? An EU artificial intelligence strategy

Author:
Bertin Martens
Publication Date:
10-2024
Content Type:
Policy Brief
Institution:
Bruegel
Abstract:
European Union policymakers want to close the artificial intelligence innovation gap with the United States, as a way to accelerate lagging productivity growth. The EU focus is on expanding an existing supercomputer network with more AI hardware and computing infrastructure, with taxpayer support. However, this computing infrastructure is not adapted to AI modelling. The cost of catching up with leading big tech AI computing centres is already prohibitive for EU budgets, and is set to become even more so. The hardware focus overlooks missing EU markets for complementary services that are required to set up a successful AI business: large-scale business outlets for frontier generative AI models to generate sufficient revenues to cover huge fixed model training costs, hyperscale cloud-computing infrastructure and private equity financing for AI start-ups. In the absence of (or with insufficient) complementary services markets in the EU, start-ups are forced to collaborate with US big tech firms. Injecting taxpayer subsidies to make up for these missing markets may further distort EU markets. Regulatory compliance costs, including uncertainty about the implementation arrangements for the EU Artificial Intelligence Act, add to market problems. The EU should address a wider range of market failures in its policy initiatives. It should strive to increase productivity growth below the AI technology frontier, by facilitating investment and applications of AI-driven services produced by derived and specialised generative AI models, or AI-applications that build on top of existing generative AI models. Building these below-frontier AI applications requires far less computing capacity and less heavy investment costs. Promoting the uptake of AI application services across a wide range of industries can substantially stimulate productivity growth. That requires a razor-sharp focus on pro-innovation guidelines, standards and implementation provisions for the EU AI Act, shortening the Act’s regulatory uncertainty horizon as much as possible, and facilitating collaborations between EU AI startups and big tech companies. Widening and deepening the EU private equity and venture capital market would also be very helpful.
Topic:
Markets, Science and Technology, European Union, Investment, and Artificial Intelligence
Political Geography:
Europe and United States of America

5. What should be done about Google’s quasi-monopoly in search? Mandatory data sharing versus AI-driven technological competition

Author:
Bertin Martens
Publication Date:
07-2023
Content Type:
Working Paper
Institution:
Bruegel
Abstract:
The first part of this paper focuses on competition between search engines that match user queries with webpages. User welfare, as measured by click-through rates on top-ranked pages, increases when network effects attract more users and generate economies of scale in data aggregation. However, network effects trigger welfare concerns when a search engine reaches a dominant market position. The EU Digital Markets Act (DMA) imposes asymmetric data sharing obligations on very large search engines to facilitate competition from smaller competitors. We conclude from the available empirical literature on search-engine efficiency that asymmetric data sharing may increase competition but may also reduce scale and user welfare, depending on the slope of the search-data learning curve. We propose policy recommendations to reduce tension between competition and welfare, including (a) symmetric data sharing between all search engines irrespective of size, and (b) facilitate user real-time search history and profile-data portability to competing search engines. The second part of the paper focuses on the impact of recent generative AI models, such as Large Language Models (LLMs), chatbots and answer engines, on competition in search markets. LLMs are pre-trained on very large text datasets, prior to usage. They do not depend on user-driven network effects. That avoids winner-takes-all markets. However, high fixed algorithmic learning costs and input markets bottlenecks (webpage indexes, copyright-protected data and hyperscale cloud infrastructure) make entry more difficult. LLMs produce semantic responses (rather than web pages) in response to a query. That reduces cognitive processing costs for users but may also increase ex-post uncertainty about the quality of the output. User responses to this trade-off will determine the degree of substitution or complementarity between search and chatbots. We conclude that, under certain conditions, a competitive chatbot markets could crowd out a monopolistic search engine market and may make DMA-style regulatory intervention in search engines redundant.
Topic:
Science and Technology, Governance, European Union, Digital Economy, Innovation, Artificial Intelligence, and Competition
Political Geography:
Europe

6. Towards efficient information sharing in network markets

Author:
Georgios Petropoulous, Bertin Martens, Marshall Van Alstyne, and Geoffrey Parker
Publication Date:
11-2021
Content Type:
Working Paper
Institution:
Bruegel
Abstract:
Digital platforms facilitate interactions between consumers and merchants that allow collection of profiling information, which drives innovation and welfare. Private incentives, however, lead to information asymmetries resulting in market failures both on-platform, among merchants, and off-platform, among competing platforms. This paper develops two product-differentiation models to study private and social incentives to share information within and between platforms. We show that there is scope for ex-ante regulation of mandatory data sharing that improves social welfare better than competing interventions, such as barring entry, break-up, forced divestiture or limiting recommendation steering. These alternate proposals do not make efficient use of information. We argue that the location of data access matters and develop a regulatory framework that introduces a new data right for platform users, the in-situ data right, which is associated with positive welfare gains. By construction, this right enables effective sharing of information together with its context, without reducing the value created by network effects. It also enables regulatory oversight but limits data privacy leakages. We discuss crucial elements of its implementation in order to achieve innovation-friendly and competitive digital markets.
Topic:
Markets, Science and Technology, Data, and Information
Political Geography:
Global Focus