Cognitive Farmsteading
= "It is primarily an idea about humans, particularly human collective attention, and the relationship between intellectual production and sociality." [1]
Description
Venkatesh Rao:
"We’re trading the biases of cognitive urbanism for the variance of cognitive ruralism. At the individual level, we’re becoming more solipsistic in our individual thinking, and more intimate in our shared thinking.
But we’re also becoming better adapted to the local conditions of our own intellectual lives, just as farms must be adapted to local topography. Social solipsism is enabling more material realism.
Our intellectual products are becoming more home-made and home-consumed. They’re being shared in more raw forms with more trusted people who share a greater degree of context with us.
We’re trading industrially organized competitive urban intellectual production on a global scale for a sort of communitarian mutualism. We’re trading polished and packaged intellectual outputs designed to propagate around the world as a cash crop for rougher ones designed to serve the intellectual needs of smaller groups, often outside of the cash economy.
As these dynamics strengthen, the general tendency of AI will be to drive divergence, not convergence. The index-fund-like convergence at the level of basic thinking, and the precommodification effects at the level of public language, give way to a redshifting universe of thought, as islands of knowledge begin to retreat from each other at an accelerating place. That they may all eventually vanish past each other’s lightspeed horizons may not matter if they have nothing much of value to say to each other.
It no longer matters if dozens of people come up with the same idea or neologism at the same time, in conversation with the same foundation models. Those parallel rediscoveries are going to shape the trajectories of small, local milieus in highly context-dependent and path-dependent ways.
When rediscoveries and reinventions of useful ideas become so cheap that they’re not worth transporting over long distances, they can be be better adapted for local use. Interestingly, something like this logic applies to biofuels — agricultural waste feedstock is not worth converting to fuel that is transported long distances, but is worth utilizing in situ."
(https://contraptions.venkateshrao.com/p/cognitive-farmsteading)
Characteristics
Venkatesh Rao:
"The idea of cognitive farmsteading rests on three specific affordances of AI — automation, bespokification, and pensievification. And these affordances are orchestrated in ways that reveal that not only is AI a technology of intellectual ruralism, it is a a technology of intellectual rural domesticity.
Ie, at the retail user level, it is much more of a highly customizable “home” technology, like handyman services, or kitchen cabinets, than a “work” technology.
Of the three, automation is the most obvious one — AI allows you to set up entire complex intellectual farmsteads, with a dizzying variety of automations and scaled operations. If you enjoyed tools like IFTTT and Zapier, you’re probably going nuts with AI automation right now.
For example, I’m finally making use of big black hole of PDFs accumulated over 20 years, and lugged from laptop to laptop (like some of my physical moving boxes that I’ve lugged unopened from apartment to apartment for a decade). With AI, I’ve been able to set up a little automation where my personal librarian sends me a random PDF everyday on Telegram, and I give it feedback to guide future choices. An entire area of highly personal and idiosyncratic memory has come alive.
Another example: I just read a book with a lot of untranslated French passages. I had Claude Code translate all of the French bits and produce a document that I read alongside the book.
Third example. I now routinely consume and produce all sorts of specialized material, well beyond my own training and intellectual abilities. Having an AI read papers you can’t read yourself, and work with your ideas to build on top of the ideas in those papers, is now a routine matter.
What is rural about this is that modern farms are also highly automated and leveraged for the few humans who remain and work on them.
Bespokification is also pretty obvious — AI allows you to create n=1 products customized to your needs, creating a virtuous cycle where your thinking can get increasingly idiosyncratic and better adapted to your circumstances.
This is perhaps easiest to see if you use agentic coding to create actual apps and other software tools for yourself, like a farmer improvising equipment, but it is visible even in advanced patterns of pure chatbot-level AI use. My wife, for instance, has been using purely chat-based AI for 3 years now to guide her experimentation with bespoke skincare products, which she makes mostly for herself and friends and family (it is worth noting that as late as the 1890s, rural households made things like soap for themselves, before consumerization turned those essentials into purchasable products).
More revealingly, she (and to a lesser extent, I) has also been using it extensively to parse and make sense of what would otherwise be an overwhelming amount of complexity in managing renovations and customizations of the old house we bought this year. Home improvement is necessarily a game of bespoke solutions to relatively idiosyncratic problems.
It is not an accident that AI and home improvement go together like peanut butter and jelly. AI is an unstoppable source of bespokification of consumption. The home is an insatiable sink of bespoke consumption. The rural home, with its greater distance from urban amenities, and higher need for autonomy, doubly so.
This is why I say: AI is a domestic technology. We use it in ways that reflect the idiosyncratic needs of n=1-4 users, the way we use various technologies and services to customize homes.
As a technology of intellectual rural domesticity, AI drives us to heavily customize the ideas we consume and produce for our own needs, with a great deal of DIYery.
While this can and does fuel lots of wild, solipsistic crackpottery, it also allows the pursuit of original ideas with the kind of leverage that would have required a great deal of compromise with other humans in urban centers.
The third attribute, pensievification, named for the fictional pensieve in Harry Potter (a device that lets you externalize and share raw memories with others), is perhaps the least obvious “rural domesticity” affordance of AI. The simplest form, sharing an incomplete chat, is already well-known. More complex forms of pensievification are emerging every week. This is like trading excess produce from your farm with neighbors.
Not only can you surround your brain and life with bespoke, customized automations, you can share their outputs with others. You can go from n=1 or n=family to n=few.
This is where a new kind of collective culture is emerging; one that is surprisingly more intimate because of AI use. Pensievification allows a kind of raw, vulnerable sharing of work-in-progress artifacts that both calls for higher trust, and engenders it.
I’m genuinely surprised by the degree to which others are using AI tools I’ve built, and I’m using tools others have built. And often, these are just barely skinned with enough of a robust UI for people other than the creator to use."
(https://contraptions.venkateshrao.com/p/cognitive-farmsteading)
Discussion
Venkatesh Rao:
< "My claim is, AI enables a powerful mode of intellectual rural domesticity — the ability to be highly intellectually productive in “rural” modes, a historically unprecedented condition." >
"Artificial minds are just as capable of participating in the patterns of collective attention that humans need to think at their best. You can construct crucibles and trigger the crucible effects with AI replacing human peers to a significant degree.
It doesn’t quite happen out of the box. Like devoted dogs, LLMs naturally attend to you attending to something, not to the something itself. But this is relatively trivial to fix so that you and the LLM are attending to the same thing together, and in usefully independent and uncorrelated ways. You don’t need elaborately theatrical and performative system prompts to school your artificial collaborator in Right Think. You just need some minor discipline to head off sycophancy and over-eager affirmation and transform it into useful outward-directed artificial curiosity distinct from your own.
And in this mode, you can strength-train your brain with AI just as well as with other humans, and often far better. But the consequences of this capability are complicated.
Because we need peer minds to think with, intellectual identity is a kind of relational identity, similar to sexual identity, with the full range of attendant anxieties and accompanying social performances and displays.
...
What will change though, is the nature of collective attention, and the kinds of minds we use to create crucible effects to drive our own thinking. We will do more and more of our intellectual production in collaboration with AIs, and require fewer and fewer of the affordances of “urban” intellectual milieus to do so. And with the humans we retain in our intellectual lives, we will build different relationships, significantly mediated by AI.
...
The most AI-positive corners of the cozyweb, like a few discords and slacks I participate in a lot, are already cheery little villages with mixed human/AI populations. This is an early version of cognitive farmsteading, featuring simple AI bots actively constructed and empowered to participate alongside humans. I myself have built and deployed three such bots.
This sort of thing is spiritually the antithesis of AI-hostile inquisitorial tooling of the sort that has cropped up on Substack.
In this analogy, obviously, places like Substack, which are institutionalizing AI-hostile defaults, are like Medievalist strongholds in Asimov’s novels.
The old internet (RIP) could be regarded as cognitive urbanization on steroids. In its heyday you could find your intellectual milieu at whatever scale you wanted, online. Even with the cozyweb turn and the evolution from dark forest to dead forest, the internet largely retained its hyper-urban character. Even without the AI bots, my Discord communities are more intense intellectual crucibles than the various university campuses in which I spent over a decade of my life (universities are surprisingly intellectually isolating places; they’re better than villages and provincial small towns where nobody reads, but not by as much as you’d think).
A year and some months into reading, writing and thinking with AI in the loop by default, I’m come to the conclusion that the hard part is not getting AI to do what I want.
The hard part is getting humans I like to think with to play new kinds of collective attention games together.
...
Today, chances are, for anything worth thinking about, everybody capable of thinking about it is already doing so with AI in the loop, developing similar understandings, down to similar metaphors and neologisms. While useful to the individual, it does make it harder and more pointless, to talk to other humans about your thinking in the usual ways. Everybody already thinks the same thoughts in the same ways, and the race is over before it begins. A great many thoughts are now common knowledge at birth (a phenomenon some determined deists are interpreting luridly as evidence of AIs becoming “hyper-persuasive”). So you have to find new ways to talk to people.
Equally, there is a precommodification of curiosities too. It is harder to think up truly original questions. LLMs act not just like index funds of past knowledge, but as index funds of present curiosities. If you explore an unformed curiosity about some contemporary matter with an AI, chances are, others are doing so too, and will converge on similarly framed curiosities.
The situation is similar on the reading front. It is increasingly easier and more valuable to have AIs summarize a great deal of reading material for you, reserving raw reading bandwidth only where there is either pleasure on offer, or value to close-reading original texts. If an author offers neither, and also locks down the text, I typically skip it."
(https://contraptions.venkateshrao.com/p/cognitive-farmsteading)