Ever since ChatGPT first made it to the market, governmental bodies and lawmakers have been wondering how to put reins on AI technology. After more than a year of development in the field, the EU stands as the first entity to bring a strict set of rules governing AI technology. These rules all lie under the EU AI Act. What is this act and what does it mean for AI technology going forward? Here’s what you need to know about this act.
At the time of writing this article, the EU AI Act was fresh out of the printer, so there’s still a lot of information yet to be uncovered. As such, this article will be updated as more information comes out. Be sure to check back every now and then. If you live in Europe, then these laws and regulations could have an effect on how you use AI technology in the future.
What is the EU AI Act?
While the EU AI Act will soon govern the use of generative AI in Europe, it actually came about before the generative AI craze. The Act was first conceived back in 2021. This act will take the different AI systems and divide them into groups based on their potential risk to humanity. For example, an AI that’s designed for mundane tasks like generating get-well-soon cards would be considered a Low risk. If that AI encounters an error, the potential effects are minimal. However, an AI that’s designed to be used by major governmental entities would be considered a High risk. If something goes wrong with the latter AI, it could pose a major threat to millions of people.
The act will also ban certain types of AIs if they pose a “clear risk to fundamental rights”. For example, an AI designed to collect and process biometric information would be banned. Each risk level will have a different set of rules. As you can imagine, the higher the risk level, the stricter the rules.
Companies developing and training these systems will have to comply with specific transparency obligations to the government. This is similar to one requirement set by US President Joe Biden’s executive order. The order requires companies making large AI models to report significant findings to the government. The EU could have a similar approach. This way, the government knows what’s going on with these models and whether they pose an imminent threat to the people of Europe.
What types of AI will be affected?
As of March 11th, 2024, this legislation is still being finalized, so we don’t know all of the information just yet. However, from what we do know, these laws will govern the use of large general AI systems. This will include AI models from companies like Google, xAI, Anthropic, OpenAI, Meta, ETC. Along with the models, there are also provisions pertaining to applications and services using these models such as Gemini, ChatGPT, MidJourney, and so on.
AI systems classified as “High risk” would be used in critical infrastructure, education, healthcare, law enforcement, border management, or elections, according to BBC. These are the areas that will see the tightest regulations.
So if you’re in Europe, most of the AI tools you use will have some sort of regulation keeping it in line.
What are the consequences of breaking the laws?
This sort of information might not be set in stone, but companies could face some serious fines if they violate the laws. Based on the level of the offense, larger companies could face fines between €7.5 million and €35 million. If the company can’t swing millions of euros, then the company could pay between 1.5% and 7% of its global turnover.
Will certain companies get special treatment?
As it stands, it looks like this bill will treat each company and AI model equally. It doesn’t look like there will be any exceptions to this rule.
Do these laws apply to copyright?
Copyright is one of the most relevant issues brought up when speaking about copyright. Several lawsuits have been tossed around alleging that using data scraped from copyrighted works constitutes copyright infringement. As of the time of writing this article, The New York Times is suing OpenAI and Microsoft over the use of its copyrighted works. That’s just one example.
At this time, the regulations don’t seem to tackle the subject of copyright. As pointed out by BBC, this act still has several steps to go before becoming official law. So, we’re certain that the subject of copyright will be touched upon at a future date.
When will these laws take place?
As of March 11th, 2024, these laws are still waiting to be finalized. So, companies are not bound by them just yet. These laws, if they are put into action, will take effect in stages. When they’re officially finalized, they will officially enter legislation starting in May this year- June at the latest. However, even though the law will be official at that point, companies will not be affected until early 2025. So, AI companies will have ample time to make preparations for the regulations.
When the laws go into effect, it will be up to countries to enforce them. They will have six months to ban AI systems that need to be banned. Also, they’ll have 12 months to start regulating general-purpose AI tools like chatbots. Additionally, they’ll have a full 36 months to regulate systems designated as High-risk, according to The Verge.
So, at the very latest, we’ll start to see some major impacts in the AI landscape in the next couple of years. The only question remains whether AI technology would have advanced too far before that point.
How will this affect AI around the world?
That remains to be seen. Right now, these will only affect AI systems used in Europe. Hopefully, other companies will follow in the EU’s footsteps so that we’ll have some sort of unity over AI regulation. However, it seems that these will only affect countries in the EU.
How will this affect my AI usage?
At this point, it doesn’t seem like there will be any immediate effect on your AI usage. You’ll still be able to log on to ChatGPT or Gemini and use it as regular. As the government starts to buckle down on AI companies, ostensibly, there could be some change, but we don’t expect there will be any sort of change that you’ll feel.
2024-03-14 15:09:54