2 ways your data is teaching AI, and how to take back control.
Written by Caitlin Roxburgh
Like it or not, AI is everywhere.
So, how can you use it? It’s more than just spell-checking your emails with ChatGPT, getting a slightly more specific answer on a search engine, or even receiving a prompt for your LinkedIn post. AI is a tool that improves the more it is used; the more information it is fed, the more accurate it becomes.
However, we are just at the beginning of what AI can do for us, and things are moving very quickly.
So quickly that, with all the AI models being developed by every platform and company, we actually don’t know exactly where all the information we feed into these models ends up.
AI pulls from so many different areas that there’s no way of knowing completely if the information you provide won’t eventually appear on someone else’s screen. This is why your firm may have a rule against feeding confidential internal information into AI models. They’re aware of the potential risks of using AI as a secure tool, when it absolutely is not. This is about to become even more important to keep in mind in the wake of ChatGPT’s new announcement.
As of the 13th of May, ChatGPT is expanding capabilities for their free users, giving them access to ChatGPT-4, which pulls answers from both the model’s own learning and the up-to-date web. Think of it like a super-charged crawler.
This capability makes accurate answers far more accessible, but it still doesn’t solve the problem of where all that important data ends up; in fact, it exacerbates it. The AI’s job is to answer your question, and it takes in a lot of information in order to fulfil that request.
By being able to pull from the internet, the model processes even more data. While we have certain guidelines for this searching of the internet (only publicly available information, etc.), people tend to forget that once they feed info into a model, it can pull up that info for its own learning indefinitely. This leads to the risk of that confidential information being used to answer someone else’s question or that info being used to fill in holes from your firm’s website that were omitted for confidentiality.
This, of course, is not specific to ChatGPT, nor is the model trying to be malicious, it is just using the information available to it to answer your question. It may not realise that combining the info that was fed into its model from a user and the info it pulled from a firm’s website will result in a data breach, even if it is only answering another user’s question.
So, it’s up to humans to regulate what info is fed into this tool, which requires us to have control over what data we want to be made available. This is an important choice that Meta is trying to sidestep for their European users through their privacy policy update at the end of this month.
You may already be aware, but on the 26th of June 2024, Meta is updating its privacy policy for its European users, allowing them to use any data uploaded to any of your accounts on their associated platforms (Facebook and Instagram) to train their AI models indefinitely.
If you weren’t aware of this, I’m not surprised.
The information was sent out in an email towards the start of June, but apart from this, they have not been publicising it widely. The process of objecting to this usage has also been made fairly difficult. For starters, you need to go to each of the accounts you don’t want them to pull data from and object through each account’s privacy policy, regardless of whether these accounts are all linked to the same email address or not.
This is also not an opt-in/opt-out situation. This is simply the opportunity to raise an objection, and you need to have a good enough reason. They can still reject your opposition to your data being used in this way indefinitely.
So, why am I telling you all of this? Why do you need to know this? Am I trying to get you to don a tin foil hat and swear off AI forever?
No I’m not.
AI is one of the best tools created in the 21st century and was the dream of science fiction for decades before that. Tools like this work best when they are fully understood, and that means being aware of the risks as well as all the advantages.
It’s in this interest that I tell you that ChatGPT will now give better answers for more users, whilst also warning you about Meta’s plans for their European users’ data. Whether you are completely opposed to AI or you are currently using ChatGPT to summarise this article into one easily digestible sentence, you deserve to know where your data is going and how it is being used.
