The Opt Out: 4 privacy concerns in the age of AI

You are more than just a data point. The exit mechanism will help you regain your privacy.

The latest wave of artificial intelligence development forces many of us to rethink key aspects of life. For example, digital artists now need to focus on protecting their work from image-generating websites, while teachers need to deal with situations where some students may outsource essay writing to ChatGPT.

But the emergence of artificial intelligence also presents important privacy risks that everyone should know – even if you are not going to figure out what this technology thinks you look like as a mermaid.

Laboring transparency

The Brookings Institution (a nonprofit agency based in Washington, D.C., conducts research to solve a wide range of national and global issues) said Jessica Brandt, policy director for AI and Emerging Technology Initiatives: “We often have little understanding of who is using our personal information, how it is used and for what purpose.”

Broadly speaking, machine learning—the process in which artificial intelligence systems become more accurate— requires a lot of data. The more data the system has, the higher its accuracy. Generative artificial intelligence platforms like ChatGPT and Google's Bard, as well as image generator Dall-E, obtain part of the training data through a technology called crawling: They scan the internet to collect useful public information.

However, sometimes due to human error or negligence, private data that should not have been disclosed, such as sensitive company files, images or even login lists, may enter the accessible part of the internet, which anyone can find with the help of Google search operators they. Once this information is crawled and added to the training dataset of AI, few people are able to delete it.

"People should be able to share freely," said Ivana Bartoletti, global chief privacy officer at Indian technology company Wipro and researcher in accessing cybersecurity and privacy enforcement at Virginia Tech's Pampurin School of Business. Photos, without worrying about how it will eventually be used to train generative AI tools, or worse – their images may be used to make deep fakes. “Crawling personal data on the internet destroys people’s abilities to it Data control. ”

Data crawling is only one of the potential sources of problems in artificial intelligence system training data. Another source is the secondary use of personal data, said Katharina Koerner, a senior privacy engineering researcher at the International Association of Privacy Professionals. This happens when you voluntarily give up some of the information for a specific purpose, but it ends up for other purposes that you do not agree to. Businesses have been accumulating customer information for years, including email addresses, delivery details, and the type of product they like, but in the past, they have not been able to do much with that data. Today, sophisticated algorithms and artificial intelligence platforms provide an easy way to process this information so that they can learn more about people’s behavior patterns. This can benefit you by providing you with only advertisements and information that you may really care about, but it may also limit the supply of products and increase prices based on your postal code. Given that some companies already have a lot of data provided by their customers, it is very tempting for businesses to do so, Korner said.

She explained: "AI makes it easy to extract valuable patterns from available data that can support future decision-making, so it's very easy for businesses to get personal data when data is not collected for this purpose. for machine learning. ”

It doesn't help for developers to selectively delete your personal information from large training datasets. Of course, it can be easy to delete specific information (such as your date of birth or social security number (do not provide personal details to a generic AI platform). But for example, implementing a complete deletion request that complies with the European General Data Protection Regulation is another matter and perhaps the most complex challenge to be addressed, Bartoletti said.

[Related: How to stop school equipment from sharing data from your family]

Selective content deletion is difficult even in traditional IT systems due to its complex microservice structure (each part works as a separate unit). But Korner said that in the context of artificial intelligence, this is even more difficult and even impossible at the moment.

That's because it's not just a matter of clicking on "ctrl F" and deleting all data with someone's name - deleting one's data requires expensive programs that retrain the entire model from scratch, she explained.

The exit mechanism will become increasingly difficult

A well-nutritious AI system can provide incredible amounts of analysis, including pattern recognition that helps users understand people's behavior. But it’s not just because of the advantages of technology – it’s because people tend to act in predictable ways. This particular aspect of human nature allows AI systems to work properly without having to know a lot of specific information about you. Because when it's enough to know someone like you, what's the point of knowing you?

Brenda Leong, a partner at BNH.AI, a law firm focusing on artificial intelligence audits and risks in Washington, D.C., said: "We have arrived at the fewest pieces of information that only require — just three to go. Five pieces of data about a person, which is easy to obtain—they will be immediately absorbed into the prediction system. “In short: it is becoming increasingly difficult or even impossible to stay away from this system nowadays.

This leaves us with little freedom because even those who have been working to protect their privacy for years will make decisions and recommendations for them. This may make them feel that all their efforts have been in vain.

Liang continued: "Even if this is a beneficial way for me, like giving me a loan that matches my income level, or an opportunity that I am really interested in, it's also in my inability to control in any way What did it for me under the circumstances."

Using big data to classify the entire population without any nuances—for outliers and exceptions—we all know that life is full of these. The devil is in the details, but also applying generalized conclusions to special circumstances, things can get really bad.

Weaponization of data

Another key challenge is how to instill fairness in algorithmic decision-making—especially when conclusions of AI models may be based on wrong, outdated or incomplete data. It is well known that artificial intelligence systems may perpetuate the biases of their human creators, sometimes with terrible consequences for the entire community.

As more companies rely on algorithms to help them fill positions or determine driver risk profiles, our data is more likely to be used against our own interests. You may one day be harmed by automated decisions, suggestions or predictions made by these systems with little to no recourse available.

[Related: Autonomous Weapons May Make Serious Mistakes in War]

This is also a problem when these predictions or labels become facts in the eyes of artificial intelligence algorithms that cannot distinguish between true and false. For modern artificial intelligence, everything is data, whether it is personal data, public data, factual data or completely fictitious data.

More integration means less security

Just as your internet presence is as powerful as your weakest password, the integration of large AI tools with other platforms also provides attackers with more prying points to try when trying to access private data. Don't be surprised if you find some of them don't meet the standards in terms of safety.

This does not even take into account all the companies and government agencies that collect your data without your knowledge. Think about surveillance cameras near your home, tracking your facial recognition software around concert venues, kids wearing GoPros running around your local park, and even people trying to get popular on TikTok.

The more people and platforms are processing your data, the greater the chance of errors. A larger space for error means that your information is more likely to be leaked to the internet, where it is easily crawled into the training dataset of the AI ​​model. As mentioned above, this is very difficult to undo.

What can you do

The bad news is that there is nothing you can do about it at the moment - neither can you solve the potential security threats derived from the AI ​​training dataset that contains your information, nor can you solve the prediction system that may prevent you from getting your ideal job. At present, our best approach is to require supervision.

The EU has passed the first draft of the Artificial Intelligence Act, which will regulate how companies and governments use the technology based on acceptable levels of risk. Meanwhile, U.S. President Joe Biden has funded the development of ethical and fair AI technologies through executive orders, but Congress has not passed any laws to protect the privacy of U.S. citizens in terms of AI platforms. The Senate has been holding hearings to learn about the technology, but it is not close to creating a federal bill.

In the process of government work, you can and should advocate privacy regulations including artificial intelligence platforms and protect users from misprocessing of their data. Have meaningful conversations with those around you about the development of AI, make sure you understand your representative’s position on federal privacy regulations and vote for those who care most about your interests.

Read more PopSci stories.

The above is the detailed content of The Opt Out: 4 privacy concerns in the age of AI. For more information, please follow other related articles on the PHP Chinese website!