AI chatbots are coming for you: The glass is half-full, and AI can't draw it

Sex in the future can provide a much-needed respite from the capitalist mandate of “faster better more,” although this will be complicated by capitalism’s need to create market niches for exploitation, and tempting advances in AI will only complicate matters further.

What Ray Kurzweil says about remote sex

Prominent futurist Ray Kurzweil, who thinks people alive today can Live Long Enough to Live Forever, has made a number of predictions about the future of technology. Many have come true, many have not, but for me, the most interesting ones are the ones that go sideways. A hardworking Redditor has helpfully compiled a list of Kurzweil’s predictions organized around his books, if you want to check in with how he’s doing.

Obviously, my main interest here is in teledildonics, or “remote sex,” which I think is one of the most interesting sideways predictions. Kurzweil predicted in The Age of Spiritual Machines that by 2020, we’d be having internet-enabled remote sex in skin-tight bodysuits with built-in haptics that are indistinguishable from the real thing. While the AR/VR parts aren’t nearly as ubiquitous or immersive as Kurzweil said they’d be, AR and VR are here, and so are haptic feedback and internet sex—but not the way Kurzweil predicted back in 1999.

Today’s internet-enabled sex can be roughly divided into three “components”: audio-visual, interactivity, and haptic (touch-based) feedback. These are respectively represented today in the plethora of AV-capable platforms available to everyday consumers for sharing their homegrown; the numerous variations of interactive video sex on OnlyFans and such; and internet-capable sex toys that can be controlled remotely (you haven’t lived until you’ve showered on live video with a cross-country groupchat and given them control of your vibrator via poll). You can get audio-visual, interactive, or even physical stimulation, all over the internet with the click of a button.

But this is a far cry from Kurzweil’s vision, in which remote sex would be not only indistinguishable from, but preferred to, face-to-face sexual encounters. I mean, technically, with all the Love and Dating in the Time of COVID stuff we had going on during lockdown, it became preferred for a minute, but I think that contributes mainly to the overall “sideways” vibe. Our remote sex either simulates, substitutes, or supplements a sexual relationship, which is fine I guess, but “fine I guess” is leagues away from “indistinguishable from and preferred to.”

Tech makes us both more powerful and more vulnerable

This is emblematic of the central problem of technology: we expect it to make our lives better, and it does, but our expectations tend to oversell the ensuing reality, while simultaneously underestimating potential drawbacks. And so the promise of technology leads inevitably to both disappointment and unintended consequences, whatever progress we make.

“Having an AI assistant, partner, and therapist” sounds great on paper, but relying on the internet for anything comes with serious vulnerabilities. I remember back in days of yore (when dragons walked the Earth and bad checks were a serious crime), when the power or the network went down at Blockbuster, we would process rentals on paper until the system came back up—hours or days later. Today, if the internet goes down, a lot of brick-and-mortar retailers all but shut down, with no procedures in place for what to do when the technology they rely on proves suddenly unreliable.

I’m not pulling a Plato here. To remove a selective pressure just is to relegate any corresponding selective traits to a vestige, these are facts, but that happens at the timescale of eons and epochs. Our brains remember about as much as they did in Plato’s day, but most folks just fill those memories with sports stats, pop culture tidbits, brand slogans, and other trivia, instead of really long poems or local flora and fauna.

Nevertheless, we do have to worry about the more short-term factor of skill maintenance, for the pocket rectangles giveth and the pocket rectangles taketh away. Instant communication and navigation removed a large part of the need for us to plan ahead, and so a lot of us stopped doing it as much, and now basically everyone I know feels like they’re constantly flying by the seat of their pants and unable to make plans. Whenever a machine does something for you, you get better at using the machine, but worse at doing the thing by yourself, for the obvious paradoxical reason that you’re doing it less even if you’re getting more of it done.

This creates big problems when students use GPT-driven AI for academic work. The very point of an academic degree is to catch up your own brain to the state of the art, which will serve as personal background knowledge to be activated when solving novel problems in the field during the rest of your career. This background knowledge is constructed by studying and working with these ideas, in a controlled environment, where the people evaluating your work know more about it than you do. You are being tutorialized by a bunch of experts.

Using AI to do the work robs you of that experience, so you don’t actually get the background knowledge you’re supposed to have, no matter how many A+ papers you pump from the machine. Like John Searle, you take the assignment and put it through the machine’s instructions, then hand back the output. No understanding required.

AI can only articulate the paradigm

You won't be able to use AI to advance knowledge of the field, because all AI can do is “articulate the paradigm,” to borrow Thomas Kuhn's phrase. It's actually really good at interpolating connections between well-established ideas (and there is admittedly a lot of good work to be done here), but those interpolations still need to be checked by an actual human expert (even if the robot tests them autonomously), and it still can't extrapolate those ideas into new territory beyond its training.

This is why AI is really good at coming up with satisfying stories for internet communities, since its instructions were trained on a variety of good stories. Following those rules of story construction (or their representation as model weight values), as well as the softer rules (again from good stories, like developing themes and variations in pacing), outputs some real bangers that push our buttons in just the right ways.

But it's also why AI can't draw a glass of wine that is anything other than completely empty or “glass of wine” full, and why I always get huge honkin’ bazonkers, no matter what size breasts I’ve asked it to draw. Trends in representation on its input set will train it to continue that representation, rather than speculate and explore beyond it. This is actually what enables AI to talk back to flat-earthers and other cranks, because its instructions enable it to reject a nonsensical defiance of what is otherwise very clear consensus.

And this is also, to bring it back around, the fundamental problem with AI sexbots, whether of the chat or droid variety. They can provide responses that mimic romantic and sexual interest, operate within your fantasy, and even fill the role of emotional support (one of the first breakthrough incarnations of AI in the 1960s was a therapy bot). But they cannot learn and grow with you the way a real partner can, they do not have any desires or interests they are not programmed to have, and they will not ever challenge you except in ways you initiate and allow.

For techbros who just want a live-in bangmaid, this probably seems fine. But just like simulations of child abuse lead to real-life child abuse, simulations of abusing people lead to actually abusing people, because the subject is practicing the motions and dynamics of abuse (physical with a droid, verbal with a chatbot) and thereby reinforcing cognitive-behavioral patterns of abuse.

This also happens with porn addiction, which is real but not as inevitable as it sounds. Over time, repeated exposure to supernormal novelty and activity with multiple simulated partners leads to habituation and escalation, to the point that you can’t “get your response cycle started” with only one live person. The solution is to either taper down so you’re not going cold-turkey, or to moderate your consumption to avoid becoming addicted in the first place. Both of these require mindfulness that is harder to muster when you’re using porn to compensate for a lack in your life—which is exactly the way many addictions start.

The point is not that these things are inherently addictive—even the “physical dependency” form of addiction relies on our own physiology, which means the mechanism is inside us, and not an inherent part of the thing we become addicted to. Addiction is not something “done to us” at all, it is one way that we are cognitively brittle. This means that lots of people will be able to moderate their usage of these new sex technologies as they proliferate and improve—but lots of people will not, and that is going to cause problems.

AI sexbots, in being cognitively brittle just like us, will also have seams and edges that can “glitch out” under the right conditions. They'll do things a person wouldn't do, such as tolerate indefinite amounts of verbal or physical abuse, or provide unconditional validation to things that shouldn't be validated. Again, techbros who just want a bangmaid probably see these as features rather than bugs—but the profound fundamental difference between the bot and a person means that any “relationship experience” (such as it is) will not necessarily apply to a relationship with a real person, and can in fact make you worse if you get used to treating your robot girlfriend poorly. Similarly, practicing with porn can increase your endurance a bit, or broaden your horizons, but threatens to twist your drive all out of whack (so to speak) if you don’t practice healthy moderation.

Everything in moderation, as the Buddhists say

And if there's one thing capitalism sucks at, it's cultivating healthy moderation. E.F. Schumacher wrote a paper on Buddhist Economics in 2003, questioning how we define “success” for an economy, and using the Buddhist principle of Right Livelihood in particular to propose that we measure economic success in terms not of highest amount of activity, but of the right amount of activity.

Otherwise, we will repeat the patterns measured with frankly alarming clarity in a 2004 paper studying a group of fisheries as they independently implemented variable degrees of automation over about the same timeframe. The fisheries with the highest tech were the most eager to get it, but they then had the most complaints, the most illnesses, and the lowest morale of the three groups. What’s more, job strain was rated highest in the high-tech fisheries, and employee decision power lowest. This is Karl Marx’s alienation all over again, only now with data to back it up. The farther we get from the meaningful part of the work and its importance to everyday life, the worse we feel.

Schumacher echoes this point in distinguishing “tools,” which enhance a person's labor power, from “machines” which do the work and relegate workers to technicians. The less we use tools and the more we use machines, the more work sucks, to drastically paraphrase.

The same is true of sex. We need some kind of Buddhist Middle Way, or at least an Aristotelian Golden Mean, to use technology in ways that robustly enhance our sexual experiences, without diminishing our humanity in the process. In a live video call from my shower, the people operating the vibrator inside me are actual people, and we have some kind of connection. But if you’re just activating a machine that pumps orgasms out of you, then it’s a sliding scale from there until you become one of Roger Williams’ “infinitely masturbating vegetables”: directly stimulating their pleasure neurons over and over, in cyberspace isolation, forever.

You can have lots of good, connected sex with yourself or with others. That doesn’t make you an infinitely masturbating vegetable. The difference is between enhancing and diminishing—are you doing it to get in touch with yourself, to connect with others, to enjoy being you? Or are you doing it to escape something, to replace something, or “just to do it”?

It can be a subtle and blurry difference, but ethics is hard. To drastically paraphrase Aristotle, “Anyone can fuck, that much is easy. But to fuck the right people, in the right way, at the right time, to the right extent, and for the right reasons—that is much harder.”

When you’re doing it right, you can appreciate the highs without wishing they were higher, face the lows with equanimity, notice the nuances in the everyday, and keep a joyful heart through it all. When you find a way to be satisfied with enough, it applies equally to orgies, solo orgasms, and life in general.

It takes mindfulness, it takes serenity, it takes restraint; but more than anything, it takes valuing something more than “more” itself.