In het kort:
De Autoriteit Persoonsgegevens waarschuwt voor de risico’s van de snelle ontwikkeling en inzet van algoritmes en kunstmatige intelligentie (AI).
AI kan leiden tot de verspreiding van desinformatie, privacyovertredingen en discriminatie, met name als het wordt getraind met niet-representatieve data.
Ondanks deze risico’s biedt AI kansen voor efficiëntie en welzijn, bijvoorbeeld in de gezondheidszorg, waar het kan bijdragen aan bijvoorbeeld snellere diagnoses.
According to the Dutch Data Protection Authority, a privacy watchdog, the increasing use of artificial intelligence and technological innovations is happening faster than we as a society can keep up with. As a result, we were late in recognizing the risks. But also in addressing them, for example through regulation and enforcement.
The benefits scandal shows the risks that algorithms pose to our privacy. Tens of thousands of people have faced problems because they were flagged as scammers by algorithms. But artificial intelligence goes further, according to experts.
Since the advent of AI-powered chatbot ChatGPT, almost every technology company is using AI. But artificial intelligence is nothing more than a set of algorithms, says educator Rajiv Girouard of the Rotterdam University of Applied Sciences. “Once an algorithm can draw conclusions or generate something new, we call it artificial intelligence.”
According to privacy expert Rijo Zenger of Bits of Freedom, algorithms and artificial intelligence should be viewed as a black box. “You put certain data in a box, but you have no idea what is done with the data in that box. Then a result comes out or new data is published, but you have no idea how it happened.”
Ontvang een melding bij nieuwe ontwikkelingen. Stay informed with notifications
Risk of invasion of privacy, discrimination and misinformation
You can unleash algorithms and artificial intelligence on all the data you can find on the Internet. For example, on the location data of millions of Dutch people available online. You can then ask the AI to analyze people who may be involved in criminal matters. This is a huge violation of your privacy.
AI risks depend on the data the AI is trained on, explains privacy lawyer Celine van Wesberg of the More Advocate Project. For example, with non-representative training data, discrimination and generation of incorrect information (misinformation) lies.
A good example of non-representative data can be seen in healthcare, for example. “Traditionally, there has been a lot of medical knowledge about men,” Van Wesberg explains. “If you train an AI model exclusively on this, the result may not be reliable for children or women, for example.”
AI is often trained using non-representative data
In addition, according to Van Wesberg, people in our country in many cases must give permission before an AI model can be trained on their privacy-sensitive health data. There is a risk that certain groups of people will give consent while other groups of people will not.
As a result, certain groups of people may be overrepresented in the training data and other groups may be underrepresented. The result is that the data with which the AI is trained is not representative.
Zenger also emphasizes that “the conclusions an AI model draws and the information the system generates depends largely on the data you put into it.” “We assume that computer decisions are neutral, but because the data is often unrepresentative, the resulting information and conclusions can be incorrect and even discriminatory.”
Stewardship of artificial intelligence is high on the agenda
Because of all these risks, oversight of algorithms and artificial intelligence is high on the agenda of the Dutch Data Protection Authority this year. “In our oversight, we prioritize complaints, data leaks, and investigations involving algorithms and artificial intelligence.”
“We have also been the moderator of the coordinated algorithm in the Netherlands since January 2023,” the privacy watchdog continues. “We do this to protect public values and fundamental rights when using algorithms and artificial intelligence.”
The Dutch Data Protection Authority has concluded that Dutch citizens have low trust in algorithms and artificial intelligence. But this is not entirely true. “Technology can also contribute to sustainable prosperity and well-being.”
Artificial intelligence offers many opportunities
Privacy lawyer Van Weisberg also asserts that “AI doesn’t just have a dark side.” “In healthcare, for example, we really need to embrace it. There are already examples of AI being able to identify cancer better than the human eye.”
But it is certain that artificial intelligence will not replace the doctor, Van Wesberg believes. “AI can help make a good diagnosis more quickly and take over certain other tasks,” the privacy lawyer says. “This leaves the doctor time, for example, for more personal contact or to help more patients. This can make care more efficient and therefore cheaper, and ensures more confidence in care.”
