It’s been hard to escape AI since OpenAI launched ChatGPT out into the world. Indeed, the recent CES event nicely captured the intense hype around the technology, with scarcely a vendor at the event not including AI in their pitch in some way, even when not at all warranted.
Even aside from this questionable “innovation”, much has been made about the importance of data for the current generation of AI. Indeed, the recent UK government paper on supporting AI made much of creating space for data centers in economically deprived areas.
GDPR and AI innovation
One might imagine that the EU’s General Data Protection Regulation (GDPR) might hinder the kind of access AI companies need to train their models. A paper from the European Parliamentary Research Service was bullish on this a few years ago, suggesting that GDPR was perfectly compatible with AI, as its principles can be applied in a sufficiently flexible manner to ensure AI applications are developed.
“The GDPR is compatible with AI, as its principles can be interpreted in such a way that they do not substantially hinder the application of AI to personal data,” the authors explain. “However, the GDPR’s prescriptions are often vague and open-ended, requiring controllers to balance competing interests, which can create uncertainties and compliance costs.”
A recent paper from Oxford University’s Carl Frey is less optimistic, however. The paper looks at how GDPR has influenced the development of AI around the world by looking at patent applications in 57 countries and 76 different industries between 2010 and 2021.
Shifting innovation
A number of key trends emerge. For instance, the study finds that innovation in AI has moved towards innovations that help to do more with less when it comes to data. For instance, deep learning methods are designed to be less data-intensive, but there have also been innovations in areas like synthetic data generation.
“The EU’s emphasis on privacy protection—although reducing AI patenting overall—has fostered innovation in data-saving technologies,” the researchers explain. This regulatory-driven specialization suggests that policy choices not only affect the volume of innovation but fundamentally shape its direction.”
This is almost certainly a good thing, as it’s impossible for computing clusters to get ever bigger to process vast data sets, both because of the energy costs and the costs involved in synchronizing the data. The hunt for less resource-intensive ways to train AI models is therefore inevitably a good thing.
Regional differences
It’s perhaps not surprising, therefore, that China dominates the more data-intensive form of AI innovation as no such restrictions apply. Indeed, various state-led and public-sector initiatives provide an almost limitless supply of data to work with.
“China’s AI patenting activity has surged dramatically, particularly in the public sector, with universities and government institutions contributing 86% and 54% of global AI patents in these categories, respectively, far outpacing the 3% and 4% shares of their U.S. counterparts,” the researchers explain.
It’s perhaps also noticeable that while GDPR is a European regulation, it’s actually the United States that leads the way in less data-intensive forms of AI. Given the nature of GDPR, it’s perhaps not surprising that Europe also leans towards this form of AI innovation, but it continues to lag behind America, perhaps due to the longstanding lack of tech giants in the EU.
Regulatory impact
That’s not to say that GDPR has had no impact, of course. Indeed, the study shows that complying with it has inevitable costs, especially for data-intensive innovations that rely on more data.
“Applicants exposed to the GDPR increased their data-saving AI patents while reducing data-intensive ones by approximately 1% in absolute terms.,” the researchers explain.”This effect is particularly pronounced among EU-based applicants, who increased data-saving patents by 1.6 percentage points while reducing data-intensive patents by 1.5 percentage points after 2018.”
Equally important is the impact of GDPR on market concentration. Previous studies have shown that the introduction of the regulation resulted in a concentration of market power in the social media sector, and the Oxford paper shows a similar trend in AI. The researchers found that established firms grew ever more dominant, which could have a stifling impact on both innovation and competition by crowding out smaller firms.
The paper highlights how GDPR has a much wider impact than purely in terms of privacy, as it also impacts both the volume of innovation and the types of technologies being developed. As regulators get to grips with generative AI and large language models, these findings are worth keeping in mind so that unintended consequences don’t emerge.
