Join 41,895 users and earn money for participation
read.cash is a platform where you could earn money (total earned by users so far: $ 171,548.58).
You could get tips for writing articles and comments, which are paid in Bitcoin Cash (BCH) cryptocurrency,
which can be spent on the Internet or converted to your local money.
Takes one minute, no documents required
The photos you posted on the Internet are being used to feed AI
In 1989, along with Christmas, there was a new Internet milestone: the World Wide Web (WWW).
Berners Lee, the inventor of the World Wide Web, believes that the most valuable part of the Internet is to give people equal access to information. Based on this concept, on April 30, 1993, the World Wide Web was freely opened to the outside world, making everyone on the planet fishermen.
Influenced by the Internet sharing culture, it is normal to take photos and share likes, but in the era of artificial intelligence, the photos you personally post on social networking sites are likely to push yourself into a surveillance situation.
A considerable number of people have become victims without knowing it.
1. The "dirty little secret" of face recognition
In early 2020, Clearview AI, a facial recognition technology company based in the United States, broke the scandal.
Clearview AI uses software to quietly extract face images from Facebook and Google and other websites, and creates a database. Reliable data shows that there are more than 3 billion face images in the database. Clearview AI enriches the database while authorizing the database for use by law enforcement agencies.
Currently, hundreds of law enforcement agencies around the world have obtained authorizations, including some private companies. With such a huge data set transaction, the owner of the photo has no idea.
Such things are far more than this.
In 2019, the New York Times published a report titled "How your child's photo powers surveillance technology."
Chloe and Jasper’s mother is one of many mothers who are keen to document their children with photographs. In 2005, after registering on the photo-sharing site Flickr, she uploaded many photos of her children.
Unexpectedly, 14 years later, photos of Chloe and Jasper, along with hundreds of thousands of photos of other families, were stored in a database called "MegaFace". There were nearly 700,000 portraits in the database. It has been downloaded by dozens of companies and used not only to train a new generation of facial recognition algorithms, but also to track protesters, monitor terrorists, and monitor the general public.
The hundreds of thousands of owners of these photos had no idea before the scandal was exposed.
We all know that face recognition is inseparable from machine learning, and the top priority of machine learning is data samples.
In the early days, the only way for researchers to get face photos was to buy them: post ads to attract interested people to the laboratory, and after signing the consent form, let these people take pictures in different light and postures. , Use this as a sample for training AI.
In the 21st century, the Internet has begun to emerge, and faster and more convenient ways to obtain photos have appeared. You can download all the photos about them by entering their name (mostly celebrities at this time) in the search engine.
As time went on, the photos of celebrities could no longer meet the demand, so the researchers turned their attention to ordinary people on social media.
As a result, "dirty little secrets" began to appear in the field of face recognition: secretly stealing online photos for training AI and even monitoring.
2. Poison is also an antidote
There is a famous European proverb called "Achilles' Heel".
According to legend, Achilles was the son of the mortal hero Peleus and the sea goddess Thetis. He was loved by his parents since he was a child. Later, it was predicted that when Achilles grew up, he would become a great hero and participate in the Troy War, but the city of Troy would fall and Achilles would die in battle.
Mother Thetis heard this and began to think of a way.
In order for Achilles to have the impervious body of King Kong, she tightly pinched Achilles’ heels and dipped him upside down into the Styx. In this way, the heels exposed to the water became Achilles. The only fragile place in Sri Lanka.
Later, everything was the same as predicted. Achilles, who had an immortal body, was stabbed in the heel by a poisonous arrow and crashed to the ground.
This proverb explains: No matter how strong a person is, there must be fatal weaknesses. The "Achilles' heel" of deep machine learning is the adversarial sample.
As long as a little bit of noise (interfering pixels) is added to the machine learning image, the panda in the eyes of the machine will instantly transform into a gibbon. This means that deep learning models are actually very fragile.
This vulnerability is already a hidden danger in the era of artificial intelligence.
For unmanned driving, the attacker only needs to place a few special pieces of paper on the road that are difficult for the driver to detect. When the unmanned vehicle passes here, it will suddenly turn to the opposite lane next door, which is difficult for the driver to make in time. reaction. This kind of attack is called data poisoning.
However, poison can also be turned into an antidote under certain circumstances, such as the aforementioned secret theft of online photos.
Not long ago, researchers at the University of Chicago designed a tool Fawkes that uses the principle of data poisoning to put a "cloak of invisibility" on photos, which can prevent ordinary people from posting photos on the Internet from being used to train AI.
In the eyes of the machine, the photo is just a bunch of pixel numbers. As long as some pixels in the photo are changed very slightly, the machine cannot recognize the person in the image. In the eyes of humans, there is almost no modified photo and original image. Any difference.
Researchers have made Fawkes free to the public. Before you post photos on various websites, you only need to modify them with Fawkes to prevent the photos from being secretly used to train AI. After testing, Fawkes can make the facial recognition algorithms of Questyle, Microsoft, Amazon and other companies 100% fail.
The poison of AI is also the antidote to "humiliation".