The EU AI Act: Wait, Does this Apply to Us?!
The Act impacts all businesses using AI that affects EU citizens. This means companies worldwide need to invest in understanding the Act, adapting their business practices to ensure compliance and maintain their ability to operate competitively in the EU market.
Aug 8, 2024
·
5
min read
Intro
Here we go! The EU AI Act is now here. Whether you're a tech giant developing cutting-edge AI or a small business using AI-powered tools, this Act will affect you. Find all this daunting? Lumiera has broken down the basics of the rules that now regulate AI across the EU and also looks at how this may impact your organisation in concrete ways. Keep reading to learn more!
In this week's newsletter
What we’re talking about: Since the 1st of August, there is now a set of rules on AI that everyone has to follow: The European Union's Artificial Intelligence Act (EU AI Act, or more simply the Act).
How it’s relevant: The Act impacts all businesses using AI that affects EU citizens. This means companies worldwide need to invest in understanding the Act, adapting their business practices to ensure compliance and maintain their ability to operate competitively in the EU market.
Why it matters: By understanding the Act's intentions and requirements, companies can not only ensure compliance but also position themselves at the forefront of AI innovation in their respective sectors. This will impact business strategy, set different standards for the market, and have a direct impact on products and other processes.
Big tech news of the week
🖥️ Self-Compressing Neural Networks: This breakthrough AI technique shrinks neural networks by 97%, cutting training time and boosting speed without sacrificing accuracy.
🌍 The market had a rocky start to the week. Many investors have been concerned about the outsized influence on markets of just a small handful of tech stocks. Nvidia plummeted on Monday and rebounded on Tuesday.
⚖️ John Schulman, one of the co-founders of OpenAI, has left the company for rival AI startup Anthropic. In addition, OpenAI president and co-founder Greg Brockman is taking an extended leave and Peter Deng, product manager has also decided to leave the company.
📱AI assistant eases NHS pressure in cataract care. “Dora,” an automated voice system developed in Oxford, phones patients to ask questions, interprets their answers and identifies patients who need to be seen by a clinician.
The EU AI Act is Here
The Act is a set of rules that impacts AI systems and solutions that affect EU citizens. The EU’s goal for the Act is to strike a balance between fostering innovation and ensuring trustworthy AI development.
There are varying perspectives on this. Some view the Act as a potential roadblock to innovation, comparing it to setting up too many traffic lights on a highway. Others question whether regulation can truly foster trust in AI systems, and others are trying to figure out at which end of the Act to start when considering what it means for their business.
There are no easy answers to these considerations. A starting point can be to see it as one (quite big!) piece of a larger puzzle that is the interplay between AI technology, organisations and society. Other factors, such as ongoing technological advancements, AI literacy, public education, and how businesses adapt to the new AI landscape, all play crucial roles.
TLDR: EU AI Act
Only have one minute to spare? The table below outlines the main elements of the Act:
Its wide coverage: what and who it is relevant for
Who is in charge of making sure it is followed
The four levels of categorisation of AI systems
What the consequences can be if not following the rule
What This Means for Organisations
Compliance
All businesses using or developing AI systems for the EU market now have a set of rules to follow. For example, a healthcare startup using AI for diagnosis will ask: "Have we reviewed our AI diagnostic tool against the EU AI Act's requirements for high-risk AI systems?"
Risk assessment
A good place to start is to evaluate your AI systems to determine which risk category they fall under. For a financial services company, this means asking: "Does our AI-powered credit scoring system qualify as a high-risk AI system under the Act?"
Documentation and transparency
There are rules around maintaining detailed records of your AI systems and being prepared to explain how they work. For an e-commerce platform, a relevant question is: "Can we provide clear documentation on how our AI-driven product recommendation system makes its suggestions?"
Human oversight
Ensuring human supervision for AI systems that have a large impact on people’s lives, such as high-risk AI systems, is a core part of the Act. For a recruitment agency this could mean looking at: "Do we have qualified staff overseeing decisions made by our AI-powered candidate screening tool?"
Data governance
Implementing strong data management practices, especially for training data is also a core part of the Act. A relevant question for a social media company is: "Are we ensuring that the data used to train our content moderation AI is diverse and free from discriminatory biases?"
Regular monitoring
Continuously assessing your AI systems for potential risks and biases will look different for organisations depending on the sector. For an autonomous vehicle manufacturer, a relevant question could be: "Have we established a system to regularly test and monitor our AI navigation systems for safety and reliability?"
Want expert guidance as you adapt your organisation to a market governed by the EU AI Act? Contact us here for a personalised consultation with our expert team.
Five Things to Consider When Reading Legal Texts
If you’ve reached this point and want to dive deeper into the Act, we have put together a few factors to guide you along the way. Lawyer talk is not Everyone Else talk! Legal texts require reading with a specific set of glasses and methods. This can be challenging, even for experienced professionals.
These five points will give you a good push along the way:
Context is key. The EU AI Act is part of the landscape of existing laws on data protection, digital markets, product safety, fundamental rights, and sector-specific regulations. Think of it like joining a TV series mid-season. You need to know what happened in previous episodes (like GDPR) to fully understand the current plot (the AI Act).
Definitions matter: Pay close attention to how terms are defined in the Act, and how some definitions are found in cases from the Supreme Court (the Court of Justice of the European Union). It's like learning the rules of a new board game. The term "AI system" in the Act is like defining what counts as a "move" in chess - it sets the foundation for everything else.
Cross-references are common: Be prepared to jump between different sections in the Act. It's similar to a cookbook where a recipe might say "see page 50 for sauce instructions". You'll often need to flip back and forth to get the full picture.
Exceptions and qualifications: Look out for phrases like "except where" or "subject to".
Interpretative guidance: The Act's recitals provide valuable context and explanations. It's like reading the introduction to a textbook. The recitals explain why certain rules exist, much like how a textbook intro explains why you're learning a subject.
Find the full text of the EU AI Act here.
Get free weekly insights straight to your inbox
No spam, unsubscribe at any time