[ad_1]
In May, Twitter said that it ends to use artificial intelligence algorithms found to favor white and feminine faces as you rotate on images.
Now, a a strange competition to highlight the AI-only program they have found that the same algorithm, which identifies the most important areas in images, also selects height and weight, and favors text in English and other Western languages.
High inheritance, provided by Bogdan Kulynych, a computer security student at EPFL in Switzerland, demonstrates how Twitter’s photo-sharing favors young and seemingly young people. Kulynych used a deep approach to create different faces, and experimented with how to respond.
“In fact, a very small, small, and feminine image, is very popular,” says Patrick Hall, senior scientist at BNH, a company that works with AI. He was one of four competing judges.
The second judge, Photo by Ariel Herbert-Voss, a security researcher at OpenAI, says the infidelity experienced by the students reflects the bias of the people who provided the information used in the teaching. But he adds that the documentation shows how a thorough analysis of algorithms can help marketing teams solve problems with their AI models. “It makes it easy to respond that if someone is like ‘Hey, this is bad.'”
The “algorithm bias bounty challenge,” which took place last week at Interpretation, a computer security The conference in Las Vegas, shows that allowing foreign researchers to re-evaluate how they can be unstable can help companies deal with problems without actually causing harm.
Like other companies, including Twitter, Encouraging experts to look for security bugs in their code by rewarding them for certain things, some AI experts believe that companies should give outsiders the opportunity to use data and resources to identify problems.
“It’s exciting to see these ideas explored, and I hope we’ll see more,” he says Amit Elazari, Director of International Security Policy at Intel and a lecturer at UC Berkeley who has promoted the use of error detection methods to address AI bias. He says the quest for bias in AI “could benefit and empower the masses.”
In September, Canada the student explained how Twitter algorithms cut images. Algorithms are designed to focus on the face as well as other interesting areas such as texts, animals, or objects. But algorithms often preferred white faces and females in images where several characters were displayed. Twittersphere soon discovered other examples of racism that reflect racism and gender.
In an impressive competition last week, Twitter made the code that would take the photos available to the participants, and awarded prizes to groups that showed evidence of other bad habits.
Others revealed additional features. One pointed out that the algorithms were biased against people with white hair. Another revealed that the algorithms prefer Latin words over Arabic characters, giving it a western preference.
Hall of BNH says it believes some companies have followed the Twitter path. “I think there is hope for this to be removed,” he says. “Because of the approaching laws, and because the number of AI events is growing.”
In recent years, much of AI’s content has been marred by examples of how unfair algorithms can be. Identification features are displayed racial and ethnic discrimination, photo support number has been found to be sexually promiscuous, and a program that verifies that a person can be rehabilitated has been proven to be favoritism Black fighters.
This issue seems difficult to erase. Recognizing justice is not easy, and some procedures, such as those used for medical X-rays, may racism in ways that people cannot see.
“One of the biggest challenges we face – which every company and organization faces – when it comes to our interests or our performance and how do we do it?” he says Rumman Chowdhury, leader of the ML Ethics, Transparency, and Accountability team on Twitter.
[ad_2]
Source link
