Can AI Sex Toys Really Learn What We Like?
The fantasy of a high tech lover who anticipates your needs before you're aware of them is still a long way off.
Image via Shutterstock
This article originally appeared on Motherboard.
Once a new technology becomes sufficiently popular—or even just buzzworthy—someone, somewhere, will try to turn into into sex tech. 3D porn, sexy VR, and internet-enabled vibrators have all had their time in the sun. Now that AI technology has begun to feel more real life than sci-fi fantasy, a number of companies are attempting to incorporate it into their customers’ sex lives, specifically in the form of AI-enabled sex toys.
What an “AI sex toy” is depends on who you ask. Is it a sex toy that will have a conversation with you, like The Smart Lipstick currently seeking funding on Kickstarter? Or is it a product where the intelligence is concentrated on the backend, like a blowjob machine that can learn how to give you the ultimate blowjob?
For many AI experts, much of what’s being sold as AI erotic devices doesn’t truly count as AI. “You don’t need AI to respond to voice commands,” Annalee Newitz, author of the novel Autonomous and co-host of the podcast Our Opinions Are Correct, told me. She explained that many of the “interactive” toys currently on the market aren’t actually making use of machine learning. But even if we haven’t quite gotten to the point of actually intelligent sex toys, the various products that have advertised themselves as being artificially intelligent offer some idea of what consumers might want out of an AI-enhanced sex toy.
What might that look like? Kyle Machulis, the creator of MetaFetish and Buttplug.io, told me that he suspects that “a toy that you start using and without having to press any buttons, and without having to really even think about it, it starts adapting to pleasure you,” is what most people fantasize about when they think of an AI sex toy.
And indeed, one of the of the earliest attempts to use AI to upgrade a sex toy promised just that. HUM, a vibrator prototype that appeared on IndieGoGo in late 2014, teased consumers with a vision of buttonless white vibrator that would adjust its vibration patterns based on your body’s movements. Billing itself as the first artificially intelligent vibrator, it was talked up everywhere from The View to Cosmopolitan.
“Because of the sensors, you don’t even need to press buttons,” HUM’s creators explained on its Indiegogo page. “HUM feels your touch and motion, and creates vibrations that change in harmony with your body. Just turn it on and use it.” Although the product never made it to market—and its website is now defunct—the media hype around its crowdfunding campaign demonstrated a desire for the kind effortless erotic experience that the HUM team promised its programming could deliver.
More recently, other companies have latched onto the idea that advanced programming might be able to enhance our masturbatory experiences. In October, the company behind the Autoblow, a sex toy for flesh-penis owning individuals that works like an automated masturbation sleeve, released a white paper documenting its attempts to use AI to enhance its product, specifically by using machine learning to teach a sex toy how to give a porn-quality blowjob. Sloan hired a team to watch, annotate, and collect 109 hours of blowjobs in porn scenes to train its algorithm to give what, theoretically, feels like a porn-quality blowjob.
Brian Sloan, the inventor of the Autoblow, told me in a Skype call that the company’s recent project as a way to use machine learning to make an automated blowjob more comparable to the real thing — to get away from the idea of sex toy “modes” and move towards one where sex toys offer “experiences” that parallel sex with another person. In the long term, he said,he sees AI and machine learning as potentially having even more benefit for sex toy users. Sloan said he envisions “a product [that learns] how to stimulate you based your response to the product.”
Yet even as these products outline an enticing future where an amazing masturbatory experience is as easy as opening a box and applying a product to your genitals, they also face obstacles that might get in the way of achieving that experience. One challenge they face: Where will the creators of these products source all the data to teach their AI what, exactly, mindblowing sex looks like?
The Autoblow AI team’s decision to train the device on porn film blowjobs makes sense on one level: Ease of data collection. There’s no shortage of pornography out there, and it’s understandable that someone might assume that the thousands (if not millions) of hours of hardcore content that have been committed to film might be able to educate an AI about sex.
But even setting aside the question of how accurately porn reflects the sex people want to experience firsthand, there’s a significant limit to the data pornography is able to provide. Some of the most important parts of sexual stimulation are invisible to a camera or obscured from outside viewers. Watching porn (even hundreds of hours of porn) can’t tell you what a blowjob giver is doing with their tongue or, for that matter, what’s happening inside a vagina to make sex feel so good. Things get even more complicated for an act like cunnilingus, where porn performers are often forced to choose between something that feels good for their partner and something that looks good on camera. (Machulis recently shared an incredibly in-depth explanation of the issues with using porn to teach sex toys about sex.)
The Lioness, a rabbit-style vibrator that’s available on the market now incorporates force sensors, temperature sensors, and accelerometers and gyroscopes to map out a user’s orgasmic response pattern—all data that could, in theory, be used by an AI to reverse-engineer the ideal orgasm. The Lioness approach, with its reliance on biometrics over visual data, might seem like a better way to develop this sort of device, but this, too, comes with a number of problems. "I love that idea that your device might be able to pick up on arousal signals,” Newitz told me. “The problem is that the limited amount of research that there is out there on [arousal signals] shows that women don’t always experience the physical signs of arousal as arousal. They’ll be measurably aroused—there will be engorgement, lubrication—but when the women are asked, ‘Are you aroused?’ they respond, ’No.’”
Indeed, a number of researchers have found that female (and sometimes male) arousal is frequently noncordant, meaning that your brain and body aren’t always in agreement about whether or not you’re turned on. You mind may be eagerly anticipating sex while your body shows no physical arousal at all, or your brain may be totally uninterested in sex while your genitals swell, lubricate, and otherwise appear to be aroused. Some research has even shown that many women will react physically to images of bonobos having sex, without any conscious sense that what they’re looking at is sexy. This poses a significant problem for a smart vibrator: How can we guarantee that what a smart vibrator registers as a user being turned on and into a stimulation pattern truly is that, and not just some noise generated because, say, someone was shown a picture of bonobo sex?
Ultimately, smart sex toys may end up being hampered by the very problem they seek to solve. The history of AI sex toy hype suggests that we want a sex toy that’ll instantly know what we like, without any effort on our end, because we’re either unwilling or incapable of figuring it out for ourselves. But without doing the hard work of figuring out what, exactly, turns us on and gets us off, there’s no real way for us to teach a sex toy how to take us to the highest peaks of pleasure.