The apparatus and the octopus
A public field notebook on convivial AI ecosystems — human-AI systems designed for agency, relation, and repair.
During AY 2026-2027, I am on sabbatical. I'm thinking about the core features of pro-social human-AI ecosystems. This effort includes the study of how intelligences emerge, how environments shape them, and how ethical life depends on conditions under which many forms of agency can coexist without one dominating the rest. I am looking to biology for inspiration, particularly invertebrates. I plan to draw on the work of Ivan Illich, Vilém Flusser, Jakob von Uexkull, Thomas Nagel, Michael Levin, and others I discover along the way.
Why? This sabbatical is a departure from my prior research agenda.
I'm finding that AI research is moving at a pace that makes deep thinking difficult (for me). Most days/weeks are spent keeping up with new models, new benchmarks, a daily accumulation of papers, and a public discourse that oscillates between extravagant promise and existential alarm. It is possible to do good work under these conditions and still lose the forest for the trees. A sabbatical is supposed to restore perspective to one's work. I am taking this literally and treating distance as a method.
The octopus. I grew up on the Atlantic and did my doctoral work on the coast of Croatia. I was on the island of Vis when I drafted my sabbatical plans. The sea has always been the most persuasive evidence I know that intelligence inhabits worlds we can barely imagine. More recently, a workshop at Harvard Divinity School on biology and agency convinced me that these questions are not a detour from research on human-AI systems but a direct path.
So, now I am reading and looking for strange loops:) from philosophy of technology, biology, and design.
I will use this notebook to think in public as the work develops. I will include short essays, reading notes, field notes from marine life observation, speculative fiction, and other musings. Most of it will be provisional. Some of it will be wrong, and corrected later in plain view.
Mostly, this page is here to invite conversation.
A working vocabulary
Convivial AI ecologies (I argue) are human-AI ecosystems designed to preserve agency, sustain relations, and enable repair. "Convivial" here refers to Illich's "convivial tools".
Agency refers to people's capacities to think, judge, write, learn, create, refuse, and act. AI systems should strengthen human agency. Again here, I refer to Illich. But, I also lean on Flusser, who suggests that agency is exercised within the constraints of an apparatus (So, AI as an apparatus programs possibility.)
Relation refers to the ways in which AI systems reshape connections among people, communities, institutions, and shared worlds. I.e., AI does not only mediate human-machine interaction. It also mediates human-human and human-world relations. Relation is central for Illich, Uexkull, Nagel, Levin... but I am particularly interested to explore the concept in conversation with Uexkull and Nagel, whose work highlights the ways in which different beings inhabit different perceptual worlds (so, other minds are never fully available to us).
Repair is a central wager of this project and distinguishes my agenda from that of AI safety or alignment. It refers to the practices and mechanisms for noticing, correcting, healing, and redesigning when things go wrong. Rather than treating failure as an exception to be eliminated, I am interested in repair as a defining feature of pro-social human AI ecosystems. Michael Levin's work offers biological examples of repair as an adaptive process in systems of agents distributed across parts and scales.
(some) questions I'm asking
-
What makes an AI system convivial?
-
How do AI systems function as apparatuses that permit/constrain... what social relationships are mediated, what kinds of judgements are delegated, what forms of agency become available, who can repair...?
-
What can cells, cephalopods, and other forms of non-human and distributed intelligence teach us about ecosystems that promote agency, relations, and repair?
-
What are useful frictions that underpin healthy ecosystems?
-
How do interfaces, incentives, norms, and institutions shape the types of intelligence which emerge?
-
What are the feedback loops through which human-AI ecosystems evolve over time?
The shape of the year
The year is scored, roughly, in five movements. I don't know how it will all sound together, which is the point.
I. Conviviality cantabile
Illich's convivial tools, and whether the idea fits AI.
II. Apparatus marcato
Flusser: what the AI apparatus makes possible, constrains, programs.
III. Other minds rubato
Umwelten, octopuses, and the limits of imagining another's world.
IV. Repair tenuto
Repair is a first-class design principle, rather than an exception to be eliminated.
V. Synthesis espressivo
By the sea, where it began.
Reach out: smr48@psu.edu
