The fembots of Ashley Madison

Content note: this post includes some explicit sexual material which readers may find offensive and/or distressing.

‘Life is short. Have an affair’.

That was the sales pitch for Ashley Madison, the website for people seeking ‘discreet’ extra-marital sex that recently came to grief after hackers dumped a load of its users’ personal data on the web. It turned out that the website was basically running a scam. Straight men, the majority of site-users, were paying to hook up with women who did not, for the most part, exist. Real women did use the site, but they were massively outnumbered by fake ones.  Profiles were cobbled together by employees, and then animated by an army of bots which bombarded male subscribers with messages.

The bots’ opening gambits were merely banal: ‘hi’, ‘hi there’, ‘hey’, ‘hey there’, ‘u busy?’, ‘you there?’, ‘hows it going?’, ‘chat?’, ‘how r u?’, ‘anybody home? lol’, ‘hello’, ‘so what brings you here?’, ‘oh hello’, ‘free to chat??’. But if a man responded (using his credit card as instructed), they started to sound distinctly bottish. ‘Hmmmm’, they would confide, ‘when I was younger I used to sleep with my friend’s boyfriends. I guess old habits die hard although I could never sleep with their husbands’. Or: ‘I’m sexy, discreet, and always up for kinky chat. Would also meet up in person if we get to know each other and think there might be a good connection. Does this sound intriguing?’ No, actually—it sounds like you’re a bot.

Some men did suspect fraud. In 2012, one site-user complained to the California state authorities, though nothing came of it at the time. What tipped him off wasn’t, however, the bots’ clunkily-scripted lines. It was being contacted in a short space of time by multiple women who supposedly lived in his area, who hadn’t looked at his profile, and who sent him identical messages. All things that might have passed unnoticed if the bots hadn’t been operating on such an industrial scale.

The Ashley Madison bots were pretty basic. But the sex industry is a serious player in the world of AI bots—more sophisticated programs that can learn from their interactions with humans, and produce novel, unscripted messages. David Levy, who has twice won the Loebner Prize (a competition based on the Turing test for machine intelligence, in which a computer has to convince human judges it is also human) is the author of a book called Love and Sex with Robots, and president of Erotic Chatbots Ltd, a company whose name is self-explanatory. Recently it has gone into business with an enterprise that makes high-end sex dolls. At the moment, sex dolls are designed to satisfy their owners’ physical and aesthetic requirements: few of them talk, and none of them could be said to converse. Chatbots, conversely, can do talk, but they’re not usually physically embodied. Bringing the two things together in one package—a doll that looks and feels realistic and can also make human-like conversation—seemed like an obvious (though technologically ambitious) business proposition.

When I first read about this I was sceptical, for reasons that are succinctly summarized in this comment left by a man:

Don’t you realize, the whole reason to get a doll is so we DON’T have to listen to them talk after sex?

But while this may be the prevailing attitude among the minority of men who regularly fuck inanimate objects, there are reasons to think it is not how most men feel. In surveys of men who buy sexual services, a high proportion typically claim to want some kind of human relationship. Silent, sullen prostitutes who make no effort to get to know the client, talk to him or pretend the encounter is enjoyable for them are apt to prompt complaints from punters, even if those punters also describe them as physically attractive and compliant.

You might think this issue would also deter men from having sex with robots: you can’t have a human relationship with a non-human entity. However, many experts believe otherwise. Studies of people who work with robots in other contexts have found a strong tendency to anthropomorphize them, projecting personality traits and feelings onto them which, outside fiction, they do not have. The military sometimes uses robots to do dangerous tasks like disarming bombs, and sometimes the robots get blown up. Human soldiers reporting these incidents say things like ‘poor little guy’. One group whose robot got blown up held a funeral for it.

So, there could be a market for talking sex robots. But what kind of conversation will they make?

The less ambitious developers are just hoping to improve on the current generation of ‘unintelligent’ sex chatbots, programmed to spew out the sort of random messages Ashley Madison’s subscribers got. You could do this by giving them a sexed-up version of the capabilities displayed by Virtual Assistants like Siri and Cortana. They wouldn’t pass a Turing test, but they’d be able to, as one developer puts it, ‘follow simple instructions’.

Erotic Chatbots Ltd. has more ambitious plans. At the moment it’s developing a bot that can ‘talk dirty’. Levy explained in an interview how you train a bot to do that:

You give them lots and lots of examples and they generalize from those examples and they can make the whole of their conversation sound like somebody who talks dirty in a loving way. We teach [the bot] and it generalizes, but it will talk about any subject. You can talk to it about Italian food and it will interject about lasagna. “I could have a great time with lasagna!”

His business partner Paul Andrew chipped in:

We’ll be using erotic writers to help us program the language, so we’re actually going to work with people who do this for a living, as it were. That way we can give the chatbot a good understanding of the vocabulary and the… talk. I’m trying to think of a good word to use there. Basically, we will give them a really good grounding, and then the chatbot learns. Once they have a vocabulary, once they have a basic brain, they grow themselves. They’re quite competent. We also work with some people who do [sex] chat lines; we’re going to pick their brains, too.

In other words: ‘we’re going to teach our bot to emulate the linguistic characteristics of porn’. In the circumstances that’s not a big surprise. But there is, perhaps, a certain irony in it. Levy and Andrew want to use cutting-edge science and technology to make machines capable of producing one of the most predictable and stereotypical linguistic registers in existence—so clichéd that its human users often sound like bots themselves.

Paul Andrew mentions ‘picking the brains’ of people who work on sex-chat phone lines. Back in the mid-1990s, the linguistic anthropologist Kira Hall did some research on the language used by phone sex workers (their own term was ‘fantasy makers’) around San Francisco. She found their performances traded heavily on stereotypes about women’s language. Like speaking in a lilting, breathy voice, ‘using lots of adjectives’ when describing yourself, and dropping in plenty of elaborate rather than basic colour-terms (your imaginary underwear wouldn’t be ‘pink’ or ‘black’, it would be ‘peach’ or ‘charcoal’). The workers knew these were stereotypes: the language they produced on the phone was nothing like the way they talked when they weren’t taking calls. But stereotypes, in their experience, were exactly what their customers wanted.

Since different customers were into different stereotypes, a skilled fantasy maker needed to be able to produce a range of female personae on the phone—schoolgirl, southern belle, dominatrix, bimbo, Asian woman, Black woman, etc. They prided themselves on being able to ‘do’ personae which were remote from their own real-life identities. One of the individuals Hall spoke to wasn’t even a woman, he was a man who could pass for a woman on the phone. On the question of race, the view was widely held that white women made the best Black women, and vice-versa. As one worker explained to Hall, the Black woman of the (mainly white) callers’ dreams was a two dimensional racist stereotype which white women were actually better at producing (not to mention less uncomfortable with).

Women (and men) who work the fantasy lines are like human fembots, performing a version of femininity that callers will pay to spend time with. Not only does this performance not have to be authentic to be convincing, in the context of commercial sex an authentic (i.e., non-stereotypical) performance of femininity would risk destroying the illusion which is the real object of desire.

But you might wonder, what is sex-talk like when the parties are not in a commercial relationship? Is the language less clichéd? Are the personae constructed less stereotypical? The short answer is, not necessarily. The researcher Chrystie Myketiak has analysed cybersex encounters between peers in a virtual environment which those who study it refer to as ‘Walford’. (It’s an online community which did a deal with a university: the university would host and maintain it in exchange for being able to observe and analyse what went on in it. Its members all consented, and their consent is sought again every time they log on). Here’s a typical extract from Myketiak’s data, in which the two parties have taken the roles of a male and a female (most likely this reflects their offline identities, but we don’t know for sure). In the transcript ‘F’ and ‘M’ identify the female and the male participant.

(M) [his] hot seed fills every crevice of your womanhood…
(M) Keeps fucking you hard, jolting your entire body with each thrust.
(F) Grinds you by twisting and turning, faster and faster… she really wants it rough.
(M) Gives it to you so hard your ancestors feel it.
(F) Is pleasured senseless, she has tears coming to her eyes.
(M) Reaches around and rubs your hardened clit, violently.
(F) Whispers “Know any other wild positions? Hehe…”
(M) Whatever comes to mind is good for me.
(F) Same here… surprise me…

Linguistically, what stands out about this extract is the way the participants mix third-person narrative, second-person address and occasional use of the first person. You don’t get that in other kinds of porn. But the narrative itself is full of porn clichés, and the whole thing is organized around the heteropatriarchal proposition that in sexual encounters, men lead and women follow. Men are dominant, women submissive: whatever men desire is also pleasurable for woman. If he’s violent, that’s OK, because ‘she really wants it rough’.

In a paper she gave at a 2012 conference on robots (there’s a written draft version available here), the lawyer Sinziana Gutiu argued that if AI sexbots are successfully developed they will further entrench these ideologies of gender and sexuality. She thinks this will be a serious problem, because the combination of verbal and physical interaction which intelligent sex robots permit will have an even more powerful effect than porn does now on men’s real-world interactions with human women.

Gutiu points out that the advanced capabilities designers hope to give future robots will make them seem human, and that perception will be reinforced by the anthropomorphizing tendency mentioned earlier. However, some key human qualities will be deliberately left out of their design—like the ability to verbalize pain or emotional distress. Above all, there will never be any question about whether a robot consents to sex. It is there for its user to have sex with as and when he wishes. She goes on:

By circumventing any need for consent, sex robots eliminate the need for communication, mutual respect and compromise in the sexual relationship…allowing men to physically act out rape fantasies and confirm rape myths.

And she believes men’s experience with these nearly-but-not-quite human entities will lead at least some of them to assume that real women can legitimately be treated in the same ways.

Her paper also hints, however, that intelligent sex robots could in principle be designed to do the opposite of what she fears. If they were trained to engage their human user in talk which emphasizes negotiating consent, communicating your desires and feelings, respecting others’ boundaries and being willing to compromise, they could be used to teach a different way of interacting from the one which is modelled in porn. I’m no AI expert, but that sounds to me a lot more difficult than making a bot that ‘talks dirty’. And also, of course, a much less attractive proposition for investors whose aim is to make a profit.

It’s not only the money angle which makes me think that Gutiu’s educational sexbot is less likely to materialize than the pornified fembot of her nightmares. Therapists tell us that the most important reason why intimate relationships fail is a lack of open and honest communication, particularly about sex. But intimacy with another person doesn’t seem to be what a lot of men are looking for. If what they wanted was an intimate encounter with a female human being—a unique, complex individual with her own thoughts, feelings and desires—how could so many men have fallen for the fembots of Ashley Madison?