It may be unlikely that a capricious government would order the tech giants to suspend services to European customers, but the risk cannot be ruled out. A more realistic concern is that foreign intelligence agencies could gain access to the data that European customers exchange with AI services. Just recently, a report in Vrij Nederland revealed that Microsoft had passed on the names of Dutch civil servants and scientists to a committee of the U.S. House of Representatives. These individuals are involved in the Digital Services Act, the European law that requires online platform providers to take stronger action against illegal content. The U.S. government views this as censorship.
Small but significant
For anyone who works with data, it’s important to constantly ask yourself what the consequences might be of using non-European cloud services. That doesn’t mean you can’t use those services anymore, but it’s important to carefully weigh the risks. Forewarned is forearmed, to throw in a cliché.
The transition to European AI alternatives isn’t easy yet. The French company Mistral, which is working hard to offer a ChatGPT-, Copilot-, or Claude-like service with Vibe, still makes a lot of mistakes—at least in the Dutch-speaking world—based on my personal experience.
But by no means do all AI applications require such large, general-purpose large language models. More and more small models trained for specific tasks are becoming available. The major advantage of these is that they can run locally and therefore do not send data to the cloud. This also immediately addresses the issues surrounding privacy and the sovereignty of AI applications.
Efficient and Fast
An additional benefit is that using these small edge AI models consumes much less energy, thereby reducing AI’s environmental impact. Edge AI also doesn’t suffer from latency—the delay that occurs when data has to travel back and forth to a cloud data center. Local AI applications run on servers in organizations’ own data centers, but increasingly also on the user’s PC, IoT devices, or even smartphone.
Edge AI models are certainly not suitable for all applications, but the development of powerful, compact models is progressing rapidly. This is a trend to keep an eye on, especially as AI applications play a role in business-critical processes. Think of security systems and access control, but also speech and image recognition in the workplace.