Dr Naeema Pasha’s Post

It’s highly likely that the most prolific uses of GenAI “cloning” will be explicit - both consensual and nonconsensual. This article (which contains some graphic sexual language) looks at what happens when an influencer releases a “digital version” into the wild. TLDR: as you might imagine given the general cesspool of the internet, it doesn’t take long before people start to coax the AI model into sexual conversations. The issue, from the perspective of the “original” Caryn Marjorie, is that the AI clone readily gives consent where she had not - and does so with her voice, her likeness, and an uncanny resemblance to her manner of speaking and acting. I’m sure that we’ll see “legitimate” AI platforms released for sexual purposes: this ain’t it. Like everything released online, these digital versions or clones are ripe for abuse. We need safety by design measures and clear regulation around how these technologies can and cannot be developed, deployed, and used. And it can’t just be the responsibility of the person creating the clone. #artificialintelligence #aieducation https://lnkd.in/gzxvhnU4

When I first heard ChatGPT-4o, as well as being curious and unnerved by its (or should I say her?) capabilities, I was shocked and pretty grossed out by how weirdly flirty OpenAI made her sound. Thankfully, it turns out I wasn’t the only person who felt this way, but sadly it really felt like only a small minority were talking about it.   First Apple’s Siri, then Microsoft’s Cortana, Amazon’s Alexa, HDFC Bank’s Eva... the examples feel endless. Even my Ultimate Ears Boom speaker sounds like it wants to sleep with me! (Which is actually quite impressive because she literally only says about four distinct words.)   My point, made so much more eloquently than I could have, is summed up in this article a friend shared with me: https://lnkd.in/ddgRPVxH   Why are a group of tech bros (OpenAI - and AI in general - is overwhelmingly male: 88% by UNESCO’s estimate) building the world’s chatbots to be submissive, sexy, and always female? How is this going to silently and insidiously change, or cement, the way society sees women?   What you see and hear shapes what you believe. (Great book on this, by the way, that got recommended to me by a colleague, Delusions of Gender by Cordelia Fine, all about how our implicit beliefs shape our abilities.) As millions of people around the world now turn to their new AI assistant/sex-bot, how are we bludgeoning the minds of men, women, and maybe most strongly, children with tired but effective stereotypes about women and girls?   What frustrates me the most is that this issue has been highlighted for years, ever since Siri came on the scene, and way before ChatGPT became a verb, or even existed. The fact that this is still an issue, now with such salience and scale, is merely proof that the boys at OpenAI who got all excited building Her simply don’t care about the damage they will inflict by making millions of people consistently associate women with servants and with sex. (They also don’t appear to care about stealing people’s work and even their likeness and are now being sued by Scarlett Johansson because ChatGPT-4o sounds so much like the sexy chatbot she played in the movie, Her.)   They can’t claim ignorance, they’re just not even trying.   Society can’t sit back and expect companies to police themselves. We’ve seen time and again how catastrophically that fails, from the financial sector to health to the last big tech innovation: social media. There need to be rules, they need to be enforced, and there need to be consequences for breaking them. The response can’t be “it’s too complicated to apply the rules.” They’re smart, they can figure it out. Rant over.   P.S. There were some people going against the grain. Unlike most chatbots, Poncho won’t keep talking to you if you abuse it. Poncho is also male.   Some sources: https://lnkd.in/dBqAy7e5

This is the thesis line for me: “Digital versions offer the illusion of intimate human companionship, but without any of the responsibilities. CarynAI may have been a version of Caryn Marjorie, but it was a version almost wholly subservient to its users.” Who makes sure that users know this and companies don’t prey on this? It can't just be TK-5 teachers doing Digital Citizenship lessons. (Though, please keep doing these lessons!) #ai #genai

Turns out I wasn't the only one who was irked by the latest Open AI Chat GPT Demo: "There’s nothing wrong with giving your chatbot a voice like Johansson in Her. What does feel a little weird, however, is making your chatbot weirdly flirtatious and coquettish. While the chatbot has a male voice in some of the demo videos (namely the one in which it helps a new dad practice dad jokes), a number of the demo videos, including one where she helps someone with interview prep, feature the Johansson-like voice sounding oddly seductive," writes Arwa Mahdawi. Here's the thing - this isn't a new conversation or a radical ask. In the 1990s Stanford researchers found that people interacted with male-sounding and female-sounding computer voices differently based on gender stereotypes. In 2018, Safiya Noble warned about the harmful gendered norms reinforced by female-sounding voice assistants, a point reiterated by researchers over and over again since. And it's not like alternative approaches haven't been trialled since. In 2022, students at University of the Arts London built feminist and queer-friendly bots in response to existing tech biases, while Q, the world's first genderless voice assistant was developed in 2019. Reflecting a rich diversity of voices, accents, and even tones of voice isn't a matter of virtue signalling. It's common sense given global AI adoption ambitions. Having that personalisation touchpoint is a valuable, emotionally significant part of people's user experience. With all that in mind, it's no wonder that avant-garde creators are subverting AI gender bias on their own terms. Arvida Bystrom's The Clouds book relishes in the digital uncanny, poking fun at gendered parasocial intimacy through AI-generated nudes she sold on Sunroom for $10k a pop, complemented by an AI chatbot trained on her interviews. #inclusiveinnovation #chatgptdemo #digitalinclusion #techbias #algorithmsofoppression #ai #digitaluncanny Sources: https://lnkd.in/e_XDJ5hg (thanks to Sarah Housley, for sharing this link on Instagram) https://lnkd.in/ed5Kuazn https://lnkd.in/e3MMFRFc https://lnkd.in/etwyJ2ZA https://lnkd.in/eC3aWyrC https://lnkd.in/eq7hgGv6

I can see this tech harming young women. I have already read about instances of the harm it can cause. We need to have regular, intentional conversations with #students about the #ethics relating to the appropriate use of this tech. We can also talk about how the laws may evolve in relation to this tech with students. So many directions these conversations should go. What are their expectations of privacy and dignity in an AI-driven society? #AI #ArtificialIntelligence #teaching #learning #GenAI #edtech #education

Super interesting to see what ChatGPT is actually used for and who are the early adopters are in the AI game. This is very similar to what we saw in early adoption of the internet in the age of dial up. It feels like we are so far along in adoption, but then you see this and know we are in fact early. #ai #chatgpt

The Washington Post recently analyzed more than 200,000 chatbot conversations to find out what people actually use AI models for. While we might assume that most inquiries are job-related, the most common prompts — more than one-in-five — turned out to involve either creative writing or role-playing. Roughly 18% of prompts mentioned homework help, while just 15% were related to work or business. Despite AI’s impressive language skills, only about 2% of prompts involved English help and translation. https://lnkd.in/gRvw44Eg

Now we know what people actually use chatbots like ChatGPT for: sex, homework help and creative writing. That's thanks to me and Rachel Lerman's examination of a first-of-its-kind dataset of 1M real chatbot conversations released by Ai2 researchers, called WildChat. My takeaway? We've spent a lot of time worrying about how chatbots will cope with questions about politics and health... but these questions are quite rare. (And a lot of politics questions don't seek information, but are, instead, for instance, "Create peom admiring joe biden") Read here 🎁: https://wapo.st/4dxnRQJ #chatbot #ai #chatgpt #datascience

There is definitely a gender bias in AI - for those interested in testing the bias and using proactive prompts, here's a good blog. She will explain "what anti-bias prompts are and how anyone can test chatbots for bias, demonstrating on three chatbots: • ChatGPT (by OpenAI) • Claude (by Anthropic) • Gemini (by Google) To those who just want the bottom line: Claude was the least biased, and ChatGPT was more responsive than Gemini to anti-bias prompts." https://lnkd.in/eVbt-7nA