ChatGPT and the future of AI in Europe: What’s at stake?

Artificial Intelligence is already reshaping our world and exploding into every part of our lives. After two years of discussion, tomorrow’s committee vote in the European Parliament is a key moment for the EU’s AI Act. And it comes at a time when generative tools like ChatGPT have reignited the debate, as AI brings significant changes to our homes and workplaces.??

Given the press frenzy of recent months, it’s easy to forget that AI has been around for a long time. And it seems like we’ve been talking about regulating it for just as long. In this respect, the EU is far ahead of the rest.?

Generative AI has rocked the debate in Europe, but humans have been adapting to technological change for millennia. Think of the humble hammer. With the same tool you can build a house, but also smash a window.???

Around four years ago, the Commission’s High Level Expert Group (of which I was part) helped put in place the building blocks of today’s risk-based approach. The higher the risk, the more you need to do to prove that your AI tool is safe.?

The issue with generative AI is that it does not fit neatly into one category or another. The technology behind it is general purpose AI, a software that can be used for multiple tasks, and therefore on its own does not have particular risk attached – it is how you train it, on what data and for what use that really matters.??

This has been a key question for lawmakers as we enter the final stages of negotiations, because it has huge implications for where the responsibilities for making sure it is safe and where the regulatory burden lies.?

Tomorrow’s vote will also have huge implications for Europe’s economy.?

Today, only 3% of the world’s AI unicorns come from the EU and we see 10 times more private investment in AI in the US, and 5 times more in China. By 2030 the global AI market is expected to reach $1.5 trillion, and we need to make sure that companies in Europe are tapping into that, without getting tangled up in red tape.?

In short, we want Europeans to be the creators of AI, not just the users. Overburdening them now will simply mean they give up, move somewhere else, or get outpaced by innovators working in other countries. I am particularly concerned about the growing AI healthcare sector in Europe – people like OnCompass, our Future Unicorn Award Winner 2021, which uses AI to spot cancer in scans that doctors would not be able to see otherwise. AI use in health will be categorised as high- risk, so how can we make sure we keep these jewels from leaving our economy???

The regulations in place also need to be easy to implement for companies without huge legal departments to put in place. This means that the wording must be precise and offer legal clarity, both on the definition of what AI is, and in the list of uses considered high risk. Vague words today can have huge implications for innovation down the line. To take one example I saw from an earlier text, AI defined as having ‘varying levels of autonomy’ could also mean having no autonomy at all.??

This is where harmonised standards also play a key role. These are the rules that translate European legal texts into specific instructions when manufacturing or coding a product. If you do it like this, in a way that can be easily measured, you are sure to be in conformity with EU law. The problem is that these standards don’t yet exist. We must ensure that they are ready and that they can cover all legal requirements. At least two years of testing and trial are needed to understand how the regulation will play out in a real business environment before it becomes law.??

Going back to the hammer analogy above, the maker of a tool cannot and should not be held liable for everything that someone chooses to do with it. With AI it’s more complicated, but it makes sense that most obligations should still fall on the user of the tool. The person supplying the technology (or algorithm) must provide information about how it works, but there should be flexibility in allocating responsibilities based on the actual uses.??

Lastly, my liberal mind tells me that while protective regulations may be necessary, the biggest market failure on tech and AI that needs to be addressed by regulation is the educational system. A digitally engaging democracy with citizens having the right skills and knowledge on digital is crucial for Europe to remain competitive in the global market.??