The European Union (EU) faces a crucial decision as lawmakers aim to finalize the groundbreaking Artificial Intelligence (AI) Act by the end of this year. The EU has the opportunity to set a global example for AI regulation, but it must strike a balance between promoting innovation and implementing effective rules. Building an ecosystem that fosters AI development without stifling it is key.
Some experts suggest that differentiating regulations based on the size of AI companies could be a viable solution. By imposing more stringent requirements on larger AI companies and less burdensome rules on smaller ones, the EU can create a more tailored approach. This approach would apply primarily to “foundation models,” highly capable AI models trained on vast amounts of data that can generate content like text, video, images, audio, and code.
However, the EU must consider the implications of incentivizing the development of less sophisticated AI models by imposing stricter regulations on the most advanced and inherently safer foundation models. To cultivate a thriving AI ecosystem, policymakers should define sensible rules applicable to all AI developers, considering specific risks rather than the size of the company.
Contrary to the belief that the EU lags behind other regions in AI development, there is already a vibrant AI ecosystem in Europe. Companies like Mistral AI, Aleph Alpha, Hugging Face, Stability AI, and Synthesia are among those driving innovation. The EU should think boldly and think like its own companies if it aims to lead the way in AI, not just in terms of regulation.
Outside the EU, the competition in the generative AI market is fierce, with numerous advanced models available. OpenAI’s GPT-3.5 and GPT-4 are widely known, but other models like Anthropic’s Claude, Google’s PaLM2, and Meta’s Llama2 also make significant contributions. It is essential to base regulations on factual understanding rather than assumptions. Bruegel’s Christophe Carugati emphasizes that the market for foundation models is competitive, with multiple providers and varying degrees of openness.
Regulatory intervention in such a dynamic and competitive market at an early stage is neither appropriate nor desirable. As technology evolves, market conditions change rapidly. Barriers to entry, such as the cost of computing resources, can decrease significantly with advancements in graphics processing units (GPUs). Smaller models can even outperform significantly larger ones in certain tasks.
Criteria for model differentiation, such as computing resources and capability, are not constant and will evolve over time. Other factors, like the investment in a model, may not be relevant in assessing systemic risks. It is crucial to consider these dynamics and adapt regulations accordingly to enable a thriving AI ecosystem while mitigating potential risks.
FAQ
1. What are foundation models in AI?
Foundation models are highly capable AI models that are trained on vast amounts of data and can generate various types of content, including text, video, images, audio, and code.
2. Why is it important for the EU to strike a balance in AI regulation?
Balancing AI regulation is crucial to foster innovation while ensuring effective rules are in place to mitigate risks associated with AI development.
3. Can different regulations for large and small AI companies be a viable solution?
Differentiating regulations based on company size could be a potential solution, imposing more stringent requirements on larger companies and less burdensome rules on smaller ones.
4. What are the risks of incentivizing the development of less sophisticated AI models?
Incentivizing less sophisticated AI models through stricter regulations on advanced models may hinder progress and limit the potential advancements in AI technology.
5. How can the EU maintain a thriving AI ecosystem?
To foster a thriving AI ecosystem, the EU should think ambitiously, support its home-grown companies, and adopt sensible rules applicable to all AI developers based on concrete risks rather than the size of the company.