Machine learning has come a long way, but it needs to meet safety criteria
UTA, April 7, 2022
As artificial intelligence enters our lives tightly, very often it is given tasks that were previously given to people, for example, to make decisions about hiring people.
However, even here everything is not as simple as it might seem, because as it turns out, with the help of some tricks, applicants can increase their chances of being accepted. Shirin Nilizadeh, Associate Professor of Computer Science and Engineering at the University of Texas at Arlington, has been particularly involved in this.
It’s all about text embedding algorithms that are used in hiring. They match the words and sentences in the resume with the job description – based on this, they make a conclusion about the hiring of candidates – however, it has been repeatedly noticed that such algorithms can be deceived. The goal of ‘Attacks Against Ranking Algorithms with Text Embeddings: A Case Study on Recruitment Algorithms’ was to identify keywords from the job description that can improve resume rankings.
“We found out that you can tailor your resume for a specific job by using specific keywords that could get you pushed toward the top. It’s a kind of hack to the recruiting process,” commented Shirin.
Venture Beat, April 3, 2022
Like any new rapidly developing technology, artificial intelligence can present not only advantages, but also risks.
Adversarial attack is the one when a smart algorithm is basically used to manipulate or deceive another AI system. However, since many AI programs learn through behavioral learning, they can become vulnerable to the fact that almost anyone can then teach the AI algorithm to perform malicious actions – and attackers, of course, understand this. Despite the fact that there is a lot of talk about machine learning and artificial intelligence technologies, they are only in development and they still have a lot to do – and therefore the issue of the security of smart technologies is likely to remain open for a long time.
Despite the fact that at the moment most of these attacks are carried out in the laboratory, the situation can change at any time – and then the threat will become quite real in everyday life. For more information about the existing risks in the field of artificial intelligence, as well as ways to protect, read the article at the link.
The Verge, March 3, 2022
Facebook reported a “massive ranking glitch” that experts say is responsible for half of all feed views exposed to a potential “honesty risk” over the past six months.
The problem was first noticed back in October last year. There was a sudden spike in disinformation in the news feed at the time. What happened is that instead of blocking false messages, the algorithm, on the contrary, began to distribute them. Of course, the number of views also increased by as much as 30 percent worldwide. However, despite the rather vivid manifestations, the reason for what was happening remained unclear, so the Facebook specialist could only observe. The ranking issue was identified and fixed last month, almost five months after it first appeared.
Meta spokesperson Joe Osborne commented that “detected inconsistencies in downranking on five separate occasions, which correlated with small, temporary increases to internal metrics.” The situation is even more different in terms of the degree of vulnerability of artificial intelligence systems, as well as the fact that even errors in it can manifest themselves in serious consequences – leading to a high degree of misinformation.
Stay up to date with what is happening! Get a first look at news, noteworthy research and worst attacks on AI delivered right in your inbox.
Few days ago, the world’s first “FOOL AI ART Exhibition ” was launched and a multiple visitors have already participated, tried to hack AI and shared their feedback. “I think ...
