A frictionless future, today!
The problems of an AI-powered future are already here; what we should worry about; the value of friction in AI development
This essay is a sequel to one I wrote in April – I’d recommend you read it first!
We can already see what reduced friction does to us on a societal level. It’s easier than ever to stay in touch with our friends, yet we see them less than ever. When there’s an effortless option that appears to meet the same need, we will often choose that: send a text instead of calling or seeing someone in person, even if the latter would add more value to our lives. As we continuously opt for the easier option, our ability to withstand friction for a greater payoff atrophies. We put off sending texts in favour of scrolling. Watching YouTube is less effort than watching an actual movie. This is what it looks like when the technology we use determines our values. The less I use my phone, the more I’m astounded that we’ve decided it’s acceptable to drop out of a conversation to look at our screens.
It’s possible that a future AI that performs tasks as well as we can – one definition of artificial general intelligence (AGI) – exacerbates this problem. Anthropic has found that engineers who outsource their coding to AI stunt their skill development: duh. Are we doomed to hand over everything we do to AI until we know longer have the skillset to do it anymore, or, worse, don’t want to?
What does a frictionless, post-AGI world look like?
In my previous post on friction, I broke down activities into 3 types:
totally pointless activities
compulsory, meaningless activities
meaningful activities
In an ideal world, AI becomes so good at handling the stuff we don’t care about that we don’t have to do it at all, leaving us completely free to do the work we want to do. Maybe if an AI agent just kept every file automatically organized on my computer the way I asked it to I wouldn’t have the temptation to spend an hour clearing out my downloads folder instead of writing. This is the foundational argument behind ‘fully automated luxury communism’, a techno-optimist model for a post-work, post-AGI future. It doesn’t matter if AI takes all our jobs, because nobody will need to work anymore, and we can all reap the profits of a post-scarcity, mega-abundant society through ‘common ownership of that which is automated’.
What a post-AGI labour market looks like is beyond me to project (good pieces here, here, here), but my most hopeful vision is of a more open, accessible space, with more time for what matters. In many ways, AI is just another instalment in a long history of labour-saving devices and accessibility tools. I also think advanced AI will shift how we think about effort and what value we ascribe to certain kinds of work. We don’t think people with personal assistants are lazy or living mindless, easy, Wall-E people lives. If everyone has a personal AI-assistant in their pockets, will we see people who offload meaningless, time-suck tasks as lazy or savvy?
The answer lies in what we do with the time we gain, and again there are parallels in the way we live now. I’m so glad I don’t have to scrub all my clothes in a tub of lye by hand; now I have so much more time to watch TikToks! Cal Newport talks about this a little in Digital Minimalism: it isn’t enough to just stop going on your phone. You have to fill the non-phone time with stuff that matters, otherwise that thing that should give us more free time becomes the thing we use our free time for, in the absence of anything else to do.
What would it look like for us to assert our values and then shape our technology use around that, rather than let the technology set the values for us? For me, that looks like pushing for AI that frees up time for creative pursuits – like writing – instead of handing writing as a practice over to AI because that’s the thing it happened to be good at first. I could definitely see a future where we rely on AGI to model what decisions we should make in critical situations because it’s better at predicting outcomes, but do we then hand over the decision-making to that AI? C-3PO tells us the odds; do we let him pilot the ship? These aren’t rhetorical questions – I can think of arguments where very advanced AIs make better decisions than people currently in charge of making decisions. But we should wary about giving up the meaningful work; the dystopian permanent-underclass future I fear is one where we’re stuck with the meaningless stuff that AI can’t do, instead of the other way around.
While sometimes it does feel like living a full and meaningful life involves a lot of swimming against the current, I’m encouraged by this return to analogue ways of doing things. It suggests that we ultimate crave friction and its accompanying meaningful outcomes, so much that we voluntarily seek it out. (Maybe not all of us, but as a species it seems we generally do.) Judith Dada describes Renaissance aristocrats turning leisure into elaborate, rules-based forms of work, suggesting that in the absence of real friction we invent our own.
Honestly, it feels like we’ve reached peak brainrot. We’re already sick of being fed endless, mindless content. When I started outlining this post, I included a bullet point that ‘maybe only some of us will get addicted to Sora’, and in the intervening months (oops!) OpenAI has binned it. That’s probably a good sign.
So what should we worry about?
The things to watch for in a frictionless future will be what we should be watching for already. We’re addicted to our phones despite our innate cravings for meaning, connection, and effort because enormous corporate entities make money by subverting those desires to keep us looking at screens through which they can sell us stuff. Unless there’s a real shift in how we create and deploy new technologies, the current we’re swimming against will probably keep getting stronger.
I think there are roughly three things to watch for as we go forward with this technology, and they’re just as relevant now as they ever will be:
1. How are people who want your money or attention trying to use AI against you?
How are the Ray-Ban Meta AI glasses pushing you towards effortlessly doing something against your own interests? Is ChatGPT offering suggestions that increase the shareholder value of OpenAI? Are these tools enabling you to live a better, fuller life in the real world, or creating a system where we pour our resources back into the digital space in a way that benefits them? The corporations developing these artificial intelligences are not neutral parties, and many of them are (or will be) beholden to external stakeholders. They will need to make money and they will be making that money from you. This is already true and only going to become subtler and more pervasive.
2. What specifically do we lose when we automate something or hand a task over to AI?
I think removing friction from unimportant stuff is probably good, but we should keep an eye on unintended consequences. Media theorist Marshall McLuhan once said of technology that every augmentation is also an amputation. The current return to analogue technologies makes me wonder if we can’t see what we lose by making things easier until that thing is gone. What we automate away might not actually be the thing we think we’re getting rid of – and some things we automate out of existence (e.g. reading comprehension) are probably more significant than others (e.g. the fun spinny wheel of the iPod Nano).
On the flip side, does making that friction optional therefore give it meaning? (See again: the fun spinny wheel of the iPod Nano.)
3. Who is deciding what tasks are meaningful? Can individuals choose what they automate?
Something I landed on while writing my last essay is that while I think we should automate meaningless tasks, what tasks are meaningless and what adds value to your life is a deeply personal thing. The example I gave was that I find going to the grocery store a time-suck that prompts me to overspend; that said, I think a world where there are no grocery stores and only online grocery delivery is an objectively bad world. Again, we need to push for our values to determine the technology and not the other way around.
We need to be careful that we aren’t unilaterally and uncritically automating away tasks that are meaningful for some people just because the people making the decisions (or the AIs making the decisions) don’t think they are. As described above, we should also be mindful of secondary, unintentional losses when we make a task easier. In principle, we should to create AI tools and systems that are applicable to a wide variety of use cases for individuals to deploy at will.
But there’s nuance here as well: widespread adoption of technologies like AI-powered self-driving cars will displace human-driven taxis and personal car ownership. There’s probably not a realistic halfway house, other than perhaps carved out spaces for car hobbyists and luxury cab rides. Autonomous vehicles are a polarising topic, but in terms of safety and increased access, they seem a net social gain. (I’ve found Search Engine’s reporting on this topic to be especially helpful for how I think about self-driving cars.)
So while I think we should work to preserve individual autonomy and subjectivity in our increasingly frictionless future society, there will be cases in which we can’t have mass-adoption of a technology alongside total free consumer choice. In those instances we’ll need to be able to debate the social vs. individual good – and we will want nuanced critical thinkers who have not succumbed to online discourse brainrot in the room where those debates happen!
At any rate, I think worry is valuable here. This substack is about fighting against the darkness, so I’m not endorsing doomerism. I think a positive, post-AGI future is possible and could be really good. But that won’t happen automatically: there are too many people who stand to make money from it being really bad. The function of worry is to keep us talking about these issues, trying to figure out how we can be the ones determining the values AI conforms to. Whatever we think about AI – environmental nightmare, useful tool, potential existential risk or saviour of humanity – it is here, it is coming, and it will fundamentally change things. Fundamentally, I think Brian Christian’s idea of the human-machine interface is critically important: while the machine will likely be able to outstrip human capabilities (and probably sooner than we think), it’s the humans who create the heuristics of that capability. We get to decide what’s important and what good looks like; I think we should push to make that ‘we’ as expansive as possible.
A final word on friction
One of the myriad problems with our current technological landscape is that it advances far faster than we’re able to adapt to it. We’re hooked on screens way before we find out what the negative effects of so much screentime are. This problem is even more salient in AI development, where recursive self-improvement would allow AI to use its own capabities to advance itself. (Even if we haven’t achieved true RSI, Anthropic has been transparent that it encourages its engineers to use Claude Code to accelerate their work on Claude Code.)
Reducing friction in AI development feels both inevitable and scary. Re-introducing friction, or maybe not pushing so hard for RSI, at least, would give us more time to adapt to this oncoming wave of ultimately disruptive technologies. We need time to figure out what our values are that these technologies to support – hell, to figure out what values to encode into the AI models themselves. The recent Anthropic/Pentagon showdown shows that we are very far away from an agreed ethics of AI deployment – and I’m not sure if I want those ethics to be determined unilaterally by either a) the companies building AI or b) the Pentagon.
Regardless, we should be wary of the accelerationist view: the perspective that we should push for advanced technologies as quickly as possible as the key to unlock greater social goods. That an advanced enough AI might eliminate poverty is not a reason to bring about that advanced AI as quickly as possible, especially as doing so means we’re more likely to entrench frankly shitty values in the AI we build, or build AIs that are engineered to boost the short-term profits of the companies building them. Even as things stand, we risk getting so far ahead of ourselves that we unleash something Too Dangerous or beyond our capabilities to deal with, which we then have to try to roll back or impose ethics or sanctions on after the fact, as with nuclear weapons – presuming we even can by then. We can’t put the genie back in the bottle but maybe we can try to coax it out… really… slowly…
Thanks for giving me lots of good stuff to think about here! I tend to just worry about/avoid/preach against ai use in a blanket way, but you’ve given me some good ideas for how to direct and discuss those worries in a more productive way.
I especially liked and will need to think more about this:
“it isn’t enough to just stop going on your phone. You have to fill the non-phone time with stuff that matters, otherwise that thing that should give us more free time becomes the thing we use our free time for”
and this
“What would it look like for us to assert our values and then shape our technology use around that, rather than let the technology set the values for us?”
Also I can’t remember if I’ve recommended the book “Rewilding Motherhood” to you before, but there was a really interesting chapter in there that I talked about what we lost with labor saving devices in a similar way that you do here. Bonus that it also ties in to pretty much all of your other articles about creativity and motherhood!