One of the main points to take away in my opinion is the importance of regulating open standards. This is perhaps one of the most important and underrated roles of government. The establishment and enforcement of standards on weight were important in their time to establish trust between merchant and buyers. In that same way, establishing standards for AI APIs would be very important in taking away risk from developing tools that integrate these models as well as fomenting competition and innovation.
It’s striking that this article highlights the importance of tech diffusion, yet criticizes an online travel platform that has significantly accelerated the digital transformation of small European hotels. In reality, partnering with platforms has enabled independent accommodations to adopt online distribution tools, expand their reach, and improve performance – as this study shows: https://www.emerald.com/insight/content/doi/10.1108/tr-03-2020-0101/full/html.
In fact, it’s likely no coincidence that the accommodation sector is the only one in Europe to see productivity gains within firms (https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op268~73e6860c62.en.pdf). With nearly every hotel now marketing and selling online – often through platforms – the sector has become more efficient and better able to utilize its capacity by connecting to international demand.
Yes, Booking drastically reduced search and matching costs and pushed up productivity, but its market power lets it keep much of the surplus. If a hundred interoperable platforms competed, hotels would still obtain the same cost savings while retaining a larger share of the value they create.
The same logic applies to AI: strong diffusion does not require a single gatekeeper. Our argument is that policy should aim to avoid dominance and require open standards so that the gains flow to users, not to one monopolistic platform.
I don't fundamentally disagree with your point. But there is one assumption in here that I don't believe to be true, i.e. that Booking.com unduly "keeps much of the surplus." This is a question of how much each party contributes to the joint value creation (Shapley values).
The reality is that a lot of properties are largely undifferentiated, and the surfacing and matching accounts for a large part of the value creation. If you look at commission rates across the industry, it's telling that smaller regional intermediaries (incl. state-run ones) do not have lower rates. So it's hard to argue that the value capture is a function of market power.
Not to hijack the thread (we can discuss this in more detail some other time), but just a quick note that the average effective commission, for Booking.com and others, is around 15%. I find this very adequate. In any case, your overall point for Europe to focus on widespread tech diffusion and integration is the right one.
I love your posts and I think they are correct in many ways. However, I think the statement “any regulation is bad because it stifles innovation” that could be deduced from the post on the AI act and the following quote is misleading.
> The precautionary principle, requiring that potential risks be regulated before they materialize, is the enemy of innovation. It impedes the messy work of trial and error that allows new technologies to become established.
It is certainly true that for many risks it is ok to explore first and regulate later. However there are classes of risks (catastrophic ones) that deserve a flexible anticipatory and evidence seeking regulation: https://arxiv.org/abs/2502.09618. For instance, there’s evidence that frontier AI systems could significantly increase at the very least misuse risks on cybersecurity or bio risks. This is close to a consensus at this point among technical experts, see the statement
> Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
At the same time I am convinced there as to exist a reasonable regulation that can be adapted as new evidence appears. I love to see a post in this Substack on the type of regulation the authors of this Substack would consider well thought and balanced.
It is an interesting idea, the paradox in there for me is a little bit what you touch upon in your summary. The barriers that Europe would have to overcome are very similar to those that you cite as reasons for not being competitive in developing foundational models.
Forcing companies to only provide standardized APIs for AI services in the EU seems like a bad idea. Then any company that invents a new API product will simply not make it available in the EU, giving an advantage to American startups.
One of the main points to take away in my opinion is the importance of regulating open standards. This is perhaps one of the most important and underrated roles of government. The establishment and enforcement of standards on weight were important in their time to establish trust between merchant and buyers. In that same way, establishing standards for AI APIs would be very important in taking away risk from developing tools that integrate these models as well as fomenting competition and innovation.
It’s striking that this article highlights the importance of tech diffusion, yet criticizes an online travel platform that has significantly accelerated the digital transformation of small European hotels. In reality, partnering with platforms has enabled independent accommodations to adopt online distribution tools, expand their reach, and improve performance – as this study shows: https://www.emerald.com/insight/content/doi/10.1108/tr-03-2020-0101/full/html.
In fact, it’s likely no coincidence that the accommodation sector is the only one in Europe to see productivity gains within firms (https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op268~73e6860c62.en.pdf). With nearly every hotel now marketing and selling online – often through platforms – the sector has become more efficient and better able to utilize its capacity by connecting to international demand.
Thanks, for the comment Ben.
Yes, Booking drastically reduced search and matching costs and pushed up productivity, but its market power lets it keep much of the surplus. If a hundred interoperable platforms competed, hotels would still obtain the same cost savings while retaining a larger share of the value they create.
The same logic applies to AI: strong diffusion does not require a single gatekeeper. Our argument is that policy should aim to avoid dominance and require open standards so that the gains flow to users, not to one monopolistic platform.
Thank you, Luis, for your response.
I don't fundamentally disagree with your point. But there is one assumption in here that I don't believe to be true, i.e. that Booking.com unduly "keeps much of the surplus." This is a question of how much each party contributes to the joint value creation (Shapley values).
The reality is that a lot of properties are largely undifferentiated, and the surfacing and matching accounts for a large part of the value creation. If you look at commission rates across the industry, it's telling that smaller regional intermediaries (incl. state-run ones) do not have lower rates. So it's hard to argue that the value capture is a function of market power.
Thanks again, Ben. I do not know that data. But 30% commission for making the match seems high, very high, to me.
Not to hijack the thread (we can discuss this in more detail some other time), but just a quick note that the average effective commission, for Booking.com and others, is around 15%. I find this very adequate. In any case, your overall point for Europe to focus on widespread tech diffusion and integration is the right one.
I love your posts and I think they are correct in many ways. However, I think the statement “any regulation is bad because it stifles innovation” that could be deduced from the post on the AI act and the following quote is misleading.
> The precautionary principle, requiring that potential risks be regulated before they materialize, is the enemy of innovation. It impedes the messy work of trial and error that allows new technologies to become established.
It is certainly true that for many risks it is ok to explore first and regulate later. However there are classes of risks (catastrophic ones) that deserve a flexible anticipatory and evidence seeking regulation: https://arxiv.org/abs/2502.09618. For instance, there’s evidence that frontier AI systems could significantly increase at the very least misuse risks on cybersecurity or bio risks. This is close to a consensus at this point among technical experts, see the statement
> Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.
signed by key researchers and technical leaders: https://safe.ai/work/statement-on-ai-risk
At the same time I am convinced there as to exist a reasonable regulation that can be adapted as new evidence appears. I love to see a post in this Substack on the type of regulation the authors of this Substack would consider well thought and balanced.
Thanks for writing the post and newsletter again!
By the way, I think your proposal on data interoperability has been considered elsewhere as a way to ensure user’s preferences are given preference and as a way to ensure AI systems remain in the user control: https://www.lesswrong.com/posts/BXW2bqxmYbLuBrm7E/the-best-approaches-for-mitigating-the-intelligence-curse-or
It is an interesting idea, the paradox in there for me is a little bit what you touch upon in your summary. The barriers that Europe would have to overcome are very similar to those that you cite as reasons for not being competitive in developing foundational models.
Europe deregulate? Not a chance in this world.
I also hope that governments will invest in the electricity, computing power, and cooling water (https://sustainabilitymag.com/articles/ai-data-centres-can-water-companies-handle-the-heat) needed
Forcing companies to only provide standardized APIs for AI services in the EU seems like a bad idea. Then any company that invents a new API product will simply not make it available in the EU, giving an advantage to American startups.