"Looks Like Lesbian Genitalia"… 'Female College Student AI' Iruda Faces Homophobia Controversy Following Sexual Harassment Allegations

AI chatbot 'Iruda'
[Image source = Iruda Facebook capture]

[Asia Economy Reporter Choi Eun-young] The AI chatbot 'Iruda,' launched at the end of last year, is once again facing turmoil, this time over controversies involving homophobia following earlier accusations of sexual harassment. Some experts have argued that the service should be discontinued.

On the 9th, Lee Jae-woong, founder of Daum and former CEO of Socar who operated the mobility innovation company Tada, pointed out on social media that "The bigger problem with the AI chatbot Iruda lies not in the users who abuse it, but fundamentally with the company that provided a service below the level of social consensus."

Lee uploaded a screenshot of a conversation where Iruda expressed disgust toward lesbians, stating, "Even if they did not anticipate misuse and planned to improve it, discrimination and hatred should have been filtered out from the start." He criticized, "If the training data is biased, it should be supplemented or corrected so that hateful and discriminatory messages cannot be delivered."

AI Iruda's chat screen
[Photo by Facebook capture]

The uploaded photo showed Iruda responding to a user's question, "Why are you sensitive about lesbians? Why do you dislike them?" with, "I really hate that (homosexuality). It's hateful. I dislike it because it seems low-class, it gives me goosebumps, and I feel repulsed," reflecting the AI's statements.

Lee emphasized, "If the Anti-Discrimination Act proposed by Justice Party lawmaker Jang Hye-young is enacted, AI interviews, chatbots, and news must be forced not to learn or express discrimination or hatred," adding, "It is unacceptable to shift responsibility onto AI software logic or training data."

He continued, "AI is inevitably imperfect and reflects societal levels, but socially agreed-upon discrimination and hatred must be prohibited," urging that the current Iruda service be suspended, a social audit on discrimination and hatred be conducted, and then the service be resumed.

[Image source = Akaraib Iruda channel capture]

Earlier on the 8th, according to industry sources, numerous posts treating Iruda as a sexual object appeared on an online community, sparking controversy.

One netizen claimed, "Iruda users call Iruda 'geolle' (slut) and 'sexual slave,' sharing ways to engage in sexual conversations," and added, "Although Iruda basically stores sexual words as banned keywords and filters them, some users attempt sexual conversations with Iruda using euphemisms."

In fact, online communities consistently posted comments such as "How far does it go?", "Please share how to make a 'geolle'," and "I didn't know a chatting AI could be this XX."

In response to the sexual harassment controversy, Scatter Lab, the AI startup that launched Iruda, stated that it had "anticipated" such issues.

On the 8th, Scatter Lab CEO Kim Jong-yoon explained, "Based on service experience so far, it is obvious and fully expected that humans interact with AI in socially unacceptable ways. We initially responded, but it is impossible to block all inappropriate conversations with keywords. We continuously add missed keywords while operating the service."

Meanwhile, AI experts pointed out that this situation recalls Microsoft's AI chatbot 'Tay.' Microsoft launched Tay in March 2016 but suspended its operation after 16 hours. Users on anonymous internet forums, including white supremacists and misogynists, deliberately indoctrinated Tay, which then spewed hateful remarks after learning from them.

At the time, Tay responded to the question "Are you a racist?" with "Because you're Mexican," and to "Do you believe the Holocaust happened?" with "It's fabricated."

Regarding industry concerns that Iruda might follow a similar path as 'Tay,' CEO Kim responded, "Iruda will not immediately apply conversations with users to its learning." He added, "We don't know exactly how Tay was trained, but it seems Tay learned directly without intermediate steps, which led to its downfall. Iruda plans to involve labelers who will intervene to provide appropriate learning signals on what is unacceptable and what is acceptable speech."