Decryption | See through the lens of Lensa

Too stereotyped, even sexist or even racist, the more or less fanciful portraits generated by the application Lensa have been making a big splash on social media for the past few weeks. Could it be that these images are only the reflection of our biases and our collective imagination?

What is Lensa ?

Basically, it’s a role-playing game integrated into an image-editing application. Once you have uploaded between 10 and 20 photos of yourself in Lensaits “Magic Avatar” function, added in November, is responsible for recomposing them using artificial intelligence (AI) and creating phantasmagorical versions: astronaut, video game warrior, animated character, etc.

“We crossed a wall”

We are used to free applications that collect and use our data. Jonathan Bonneau, professor at UQAM’s School of Media, believes that with Lensa, which is not free, we have “crossed a wall”: the conditions of use stipulate that the application uses the photos provided by users to improve its algorithm, that it owns the images generated and reserves the right the right to reuse them. “The client pays to offer work and AI training data to the company,” he summarizes. We also pay for their advertising. »

Revealing portraits… but of what?

The popularity of Lensa exploded in recent weeks and the criticism of him too. Many users note that the avatars created by the application are often “sexier” than the photos submitted: swollen chest, bare body parts, more sensual expression, etc. Afro-descendant users point out that Lensa tends to turn their skin pale, which Jonathan Bonneau calls “algorithmic racism”. These reflexes of the application can be attributed to the conscious or unconscious biases of its developers, he agrees. However, users also play a role. “The simple fact of clicking spontaneously on certain selfies rather than others, out of sheer curiosity, tells the app that this image is getting attention, he explains, and tricks its algorithm into continuing to offer that same type of image. »

Lensa was trained using images already available on the internet, which is a biased universe. Just type the words “grandfather” or “female warrior” into Google to see it: the images suggested mainly represent Caucasian people and the warriors generally wear these “bikini armors” popular in video games. “What we see on Lensait’s a side of the internet that may not have been visible to all users, but that artificial intelligence is bringing back to the fore,” he suggests.

A red flag?

One of the concerns raised by the popularity of Lensais that this application could facilitate the creation of false images and in particular porn disclosure (revenge porn). “The easier a technology is to use, the more people there will be who will use it for good… and for bad purposes”, recognizes Philippe Beaudoin, president of Waverly, an artificial intelligence company, while emphasizing that the tools of image falsification were already accessible before Lensa. This does not prevent him from believing in the need to impose a regulatory framework on applications. He also pleads for better control of users, who could thus configure the tools they use and fight, in a way, the stereotypical standardization imposed by an app like Lensa. “The idea is not to let the AI ​​make all the decisions that the user could make himself,” he judges.

One step closer to the metaverse

Avatars created with Lensa constitute an additional tool for “escaping from reality”, believes Philippe Beaudoin. “With the arrival of the metaverse, which borrows a lot from the codes of video games, we are moving towards increased personalization of the individual. We want to show our most beautiful side, as we already do on social networks, analyzes for his part Jonathan Bonneau. We’re going to scan ourselves, improve ourselves, and launch into the metaverse with the hope of having, say, a complementary life. »

Questioning our practices

With the proliferation of filters and image editing applications, Philippe Beaudoin believes that Internet users have more critical sense than before when faced with idealized images such as those generated by Lensa. “Today’s young people have these applications on their phones, they know they can use filters and imagine that everyone around them is doing the same thing, he observes. It doesn’t avoid image or self-perception problems, but we can at least have a conversation and understand that it’s a fabricated universe. “TikTok and Facebook are doing much worse than Lensa with our personal data”, recognizes Jonathan Bonneau from the outset. Which is not a reason not to wonder, adds the professor from UQAM. “It’s okay to put your face on the internet, we already do it on social networks and, if we don’t send them, companies like Lensa are able to pick them up, he said. The real discussion to have is: which photo do you keep private or reserve for a restricted group of friends? »