Top 8 Artificial Intelligences That Aren’t That Smart

1. Racist facial recognition software

February 2019, Nijeer Parks, a 31-year-old African American, is arrested for crimes he did not commit. Among the charges: shoplifting, illegal possession of weapons and narcotics, possession of false identity papers, or even abandoning the crime scene and resisting arrest. Imprisoned for 11 days, he risks much more. Fortunately, the accused has a solid alibi: that day, he was 48 km from the scene of the crimes, sending money to his fiancée from a Western Union.

So why was he worried? Well… Because of an artificial intelligence software dedicated to facial recognition. The police report explains that the only evidence used is “a strong profile comparison between the photo of the fake driver’s license and his own, found in a database”. However… The two men have nothing to do physically, except for their skin color. The AI ​​seeming to consider that “all people of color are the same”, is unfortunately not its first racist action: Parks is the third black man to bear the brunt of these biases and to be wrongly accused. A 2019 study also shows that facial recognition algorithms are much less accurate for black and Asian faces.

2. The Microsoft chatbot spreading hate messages…

Brilliant idea from Microsoft in 2016: the launch of the chatbot (a conversational robot) Tay to interact with Twitter users. Its specificity: it is designed to learn from these exchanges with humans. A short-lived experience, since it only took a few hours for the whole thing to spin out and start writing horrors like “Feminists should all burn in hell”. Nice.

3. … and who becomes a Nazi!!!

… And that’s not all, since Tay then proceeded to declare that “Hitler would have done a better job than the current monkey” before denying the existence of the Holocaust. Nice showcase of machine learning, right? Fortunately, Tay was taken out of the circuit after only a few hours. Definitely the best idea of ​​the whole project.

4. The Facebook algorithm that confuses people of color with monkeys

In 2021, a recommendation tool used by Facebook asked people if they wanted to see “more primate videos” after watching a Daily Mail video showing black people. Even though the message was deleted and Facebook quickly apologized, it’s going very badly. Yeah… It just doesn’t work at all.

5. Google‘s photo app, and more racism

Lalalalala of racism, do you want it, here it is. Clearly, Artificial Intelligence software has a serious problem with diversity. In 2015, Google‘s “Photos” application reproduced the same pattern as mentioned in the previous point by assimilating an African-American couple to… Gorillas. In response, the designer simply removed the categories “gorilla”, “chimpanzee” and “monkey” from the algorithm. On the surface, it works. Basically, this in no way solves the question of racist biases. There’s still work.

(Google-photos-confond-deux-afro-americains-avec-des-gorilles-1363518″>Source)

6. Amazon’s sexist AI

Amazon is a pioneer in AI-assisted recruiting. Problem: the engineers who worked on the algorithm omitted a few details, like the fact that the machine was biased in favor of men (what a surprise). Technically: artificial intelligence is trained with already biased data. The employees of the box, at the time of design, being mainly men, the algorithm considers gender to be a main selection criterion. If even the algorithms start to be macho, we are not out of the hostel, I tell you!

7. The AI ​​that confuses bald heads with balloons

For once, it’s funny. In 2020, during a Scottish football championship, the AI ​​of a camera supposed to follow the ball, finally followed… The bald head of the linesman. Spectators must have had a hell of a time.

8. Alexa advising a little girl to electrocute herself

It all starts with a TikTok challenge as stupid as it is dangerous, consisting of half plugging in your phone charger, and placing a coin on the visible part of the pins. Possible consequences of this stupid move: electrocutions, fires and multiple bodily injuries. And precisely… At the end of 2021, a 10-year-old girl plays with the voice assistant Alexa, asking her to give her fun challenges to complete (lie on her back and turn around while holding a shoe on her foot, for example ). And… Do you see where I’m coming from? The speaker ended up asking him to realize this aberrant Tiktok trend. Fortunately, the child’s mother was nearby and stopped him. Otherwise, the consequences could have been terrible.