Linked – ChatGPT in trouble with the EU (again)

This story seems inevitable to me. As the tech world pushes us increasingly into asking an AI chatbot for information instead of looking it up ourselves, and the AI model has incorrect information, where do you get it corrected?

When OpenAI was asked to correct or remove this misinformation, they said it was ‘technically impossible’ and also failed, when asked, to disclose information about how the data was processed, instead offering to simply filter or block data on certain prompts, related to the complainant.

Under GDPR law, people in the EU have rights about their personal information, including a right to have incorrect data, corrected. So, OpenAI’s refusal to remove or rectify the complainant’s information seems to be a direct breach of GDPR law.

If you understand how OpenAI built the language models for ChatGPT, it’s not hard to understand why it can’t just correct it. They didn’t create the incorrect information; they indexed it from millions of combined sources. (I know that’s not technically accurate to say they indexed it, but it’s close enough.)

The interesting thing about this is that the law was on the books. If you’re building an encyclopedia or online database of information, you have to provide a way for people to get their information corrected. OpenAI built a tool that couldn’t do that, completely ignoring that law. They more likely assumed it didn’t apply to them because what they did was more important than silly laws about public data.

As much as I enjoy learning about AI and thinking about the possibilities, there is this nagging doubt about it. That doubt comes from the fact that the people building these things do not account for obvious downsides. Ignoring this very basic GDPR requirement is one example.

Want another? How about a deep fake audio leaked onto social media of a principal making antisemitic remarks? Sure, months later, they’ve identified it as fake and arrested the high school athletic director. Still, that principal had to go into hiding, and even now, who knows what kind of unhinged person who hasn’t seen the update might do to them or their family?

Audio deep fakes were not only an obvious use of AI with audio but they are encouraged and advertised. Isn’t it fun to have Tom Hanks “say” something to your friends or audience during a presentation? This kind of misuse was going to happen. Eventually, someone is going to get killed because of it, much as someone will die after having AI-faked nudes or porn videos leaked.

Or how about some more fake audio of a BBC presenter used to sign an endorsement deal that she was never part of?

Again, technologists have unleashed this tool of the world and provided access to anyone who wants it for whatever purpose. Now, governments and other agencies are trying to play catchup. In the meantime, all of us are at risk of scammers, harassers, haters, and fake news outlets because of these tools. The companies that make them are so heavily invested in making sure they eventually turn a profit that the rest of these concerns are nothing. OpenAI, Google, Meta, Microsoft, etc., have poured trillions of dollars into AI. So what if a few million of us have to get harmed financially or physically? That’s nothing compared to the money they lose if they limit what AI can do.

So, we’ll see if Europe punished OpenAI for this disregard of its laws. We’ll see if the US can avoid major political upheaval due to fake political news during this election cycle. We’ll see if the world figures out how dangerous deep fake audio, photos, and video are to everyone. Or if they push all those concerns away to try to help these companies avoid massive losses from developing AI tools.

On the other hand, there is one place that is making a ton of money already from this AI hype:

The tiny island nation of Anguilla.

Revenue for dot ai domains grew from $7.4 million in 2021 to around $30 million in 2023. This year, Cate says that figure could be north of $50 million dollars — more than 40 percent of the country’s annual revenue. Most of that is from registration fees, which now cost $140 for a two-year contract. But some of it comes from auctions.

Good for you, Anguilla.

https://www.aitoolreport.com/articles/chatgpt-in-trouble-with-the-eu-again

Follow these topics: Tech