The Privileged Node and Its Intellectual Exoskeleton

I have recently been building a filesystem for my intellectual projects. It is deliberately plain: a library folder containing one subfolder for each project. Each project folder preserves a snapshot of the project’s current state and instructions for the AI agents working with me, while the larger library accumulates the knowledge produced across projects. The purpose is continuity: when substantial work exists only inside a particular chat, model, or interface, later AI systems recover the context poorly, and the human operator must laboriously supply a large amount of input merely to reconstruct the project’s last working state.

At first, this looked like an organizational project primarily for the benefit of the AI agents themselves, and less so for me. I now think it is better understood as a small experiment in intelligence amplification. Interestingly, concurrent to the filesystem project, I had GPT 5.6 dig up and summarize Anders Sandberg’s old, sprawling transhumanism site; while doing so, the agent suggested a parallel between the filesystem project and Anders’s intelligence amplification system.

The historical framework comes from Sandberg’s 1997 essay, “Amplifying Cognition: Extending Memory and Intelligence”. Sandberg organized cognitive enhancement across two dimensions—internal versus external and hardware versus software—spanning the biological brain and its learned methods, biological interventions, and external resources such as writing, computers, and the internet. The demarcations are porous, and that is partly the point: human intelligence is rarely the product of a naked brain operating without support. Social organization, speech, books, and other aids have long altered what human beings can think and accomplish. Intelligence amplification therefore is about more than merely raising IQ, making memory photographic, or finding a Limitless drug. Effective intelligence for high-performing persons increasingly stems from a coupled system: biological memory and intelligence, learned knowledge and reasoning, a smartphone, external project files, search systems, AI models, human collaborators, and more besides.

	Internal	External
Hardware	Brain, body, health, biological modification	Phone, computer, books, sensors
Software	Knowledge, memory, habits, reasoning methods, values	Search, files, Anki, AI, databases

The privileged biological node

According to my biased anthropocentric philosophy, I must remain the privileged biological agent at the center of this sprawling system. The surrounding components form something of an intellectual exoskeleton: they extend my memory, computational ability, search, and collaborative ability beyond what my brain could accomplish unaided over the same period of time—or at all.

Calling the human operator the privileged node is meant to emphasize human centrality in intelligent systems. I supply the purposes for which the system is used. I experience its consequences. I remain responsible for the clinical, ethical, and intellectual judgments made with its help. I am the animus. The external system should expand the reach of the biological agent without displacing the soul of the system.

This makes biological intelligence itself an important axis of intelligence amplification. The rationale for this is briefly developed in my note, “The Flat Part of the Curve”. Sleep, exercise, health, pharmacology, sustained education, and future biological modification act more directly on the privileged node, as do changes to what Sandberg called internal software: learning, metacognition, stored memory, etc. Each of these deserve sustained development in separate work.

The danger of successful externalization

A key dimension to a developed intellectual exoskeleton is external memory, which works well. A recent meta-analysis of cognitive offloading found a large improvement in performance on memory-based tasks when people could use external aids rather than internal memory alone. This is one of the reasons writing, calendars, databases, etc. are as useful as they are. But successful task performance with an external aid is not the same thing as improved internal memory. There is a characteristic tradeoff: “Boosting performance but diminishing memory.”

This creates a genuine design problem for our amplification scheme. The more effectively memory is externalized, the less frequently it is encoded, retrieved, and practiced by the agent. Retrieval itself strengthens circuits. Searching one’s own memory can bring distant ideas into contact—Proust’s madeleines. Knowledge already present in the mind supplies the background against which a new claim appears plausible, surprising, or obviously wrong. In fact, reducing our working cache arguably limits our ability to find and attach meaning to experience—another idea I will return to elsewhere. See Derrida’s contextualism.

The answer cannot be to reject external memory; that would discard one of the most powerful forms of human intelligence amplification ever developed. The design question is not whether to externalize knowledge, but what must remain internal. Foundational causal models, frequently used knowledge, values, and red flags must remain available to the biological agent. Exact or rarely used details can remain external, as demonstrated by the famous anecdote of Einstein’s failure to reproduce the speed of sound from rote: “I do not carry such information in my mind since it is readily available in books.”

The smartphone and Anki: traffic in opposite directions

The smartphone may be the most important everyday external-cognition device yet created. It travels with us; it can capture text, images, speech, location, observations, and ideas at the moment they occur. Sandberg imagined a wearable camera that could copy text from the physical world into external memory, where an associative system might later retrieve it. The smartphone is that system when coupled to an appropriately designed AI workflow.

The same smartphone is also a trap. It can dissipate attention through notifications, feeds, and indefinitely available stimulation. Philosophically, meaning and value are supplied by being-in-the-world—a broader contextualism. Attention continually withdrawn from embodied involvement risks the privileged node’s existential grounding—it risks deanimating it. For now, this remains an exploratory connection that I intend to develop further, especially as it connects to the existentialist canon of the twentieth century.

The design objective should therefore be Sandberg’s “wearable camera” that is always available for intentional capture and generally unavailable for mindless consumption. Capture should require one gesture and provide an immediate exit. Feeds, alerts, and open-ended engagement should occupy a different mode, preferably one that requires deliberate entry.

Before I forget, I want to emphasize one especially important value of this system by foregrounding it here: use the system to capture ephemeral but noteworthy thoughts before they disappear. A fundamental axiom of learning is that we get more of what we reinforce. A tool used to capture actionable ideas such that the privileged node can later meaningfully expand upon them, reinforces ideation, creativity, intelligence. If I notice a good idea, preserve it, return to it, and develop it into an experiment, decision, or piece of writing, I reward the process of creativity and intellectual curiosity, which may itself increase the frequency with which useful novel connections subsequently occur.

In contradistinction to externalized memory, Anki is a memory aid designed to improve organic recall and memorization through spaced repetition. It moves information in the opposite direction and is the bane of tens of thousands of medical students globally (more on appropriate Anki use later). Anki’s flashcards and scheduling system are external, but their purpose is to create durable biological availability through retrieval and spacing. In my own system, it remains central for foundational medical knowledge and includes a corpus of always-expanding information. In short, a citation can remain in a database, but a mechanism or discriminating clinical fact needed to navigate a clinical encounter with deranged physiology must be internalized and immediately ready-to-hand.

Taken together, the smartphone and Anki illustrate the desired two-way traffic: worthy thoughts can be externalized for later development while foundational knowledge is deliberately pushed back into biological memory.

The AI tax

AI is by far the most powerful new external-processing component in this system, and perhaps the easiest to evaluate dishonestly. Its benefit is obvious when it finds a source I had forgotten, produces a useful alternative model, or compresses hours of clerical work. Its cost is obvious when I fight through fabricated citations, mind-boggling repeated errors, or fluent but simultaneously vapid prose. All of that correction consumes the scarce human bandwidth the tool was meant to expand—I call this the AI tax. The relevant quantity is not gross output. It is something like quality-adjusted net cognitive bandwidth: useful and correct work gained, minus the time and attention spent prompting, verifying, correcting, coordinating, and recovering from interruption.

There is also a longer-term cost that ordinary productivity measurements miss but that is an increasingly hot topic of discussion, especially in education. When an AI performs the research, synthesis, or writing, the human receives less practice performing those same operations. Deskilling or never-skilling are both serious and valid concerns. The same AI can either suppress or elicit thought depending on workflow, however. If it supplies the initial frame, the person practices selection and editing. If the person first states a model and the AI attacks it, the person practices explanation, prediction, and revision—and encourages iterative improvement. Work with AI itself can become a form of intellectual training.

The overall trends are clear enough that I’m confidently betting that optimized AI systems will expand the operational theater and potency of the privileged node and that targeted mitigation regarding de- and never-skilling can minimize unwanted atrophy where appropriate, but it won’t be an easy balance to strike, and full elaboration is outside the scope of this piece.

From exoskeleton to network

The filesystem is only one of many layers in my exoskeleton. My blog is another. Sandberg’s website is another. Other writers, researchers, institutions, and readers form additional nodes of an even broader intelligent system. Ideas move through this network, are criticized and recombined, and sometimes return decades later in a new context—such as here. The internet functions as a chaotic collective memory and arguably a form of superintelligence itself. AI increasingly serves as the primary interface to it. This is intelligence amplification at a scale larger than one person: the same principle recurs at every level.

This essay itself emerged from such a system. Sandberg’s old website persisted as a public node; search and AI recovered it on demand; my project files proved fertile ground and a connection was proposed; it was evaluated and written up; and this blog post will become yet another node that somebody else may later retrieve and develop, likely using their own intellectual exoskeleton augmenting their individual productive potential.

More connectivity and more processing do not automatically produce better judgment. A system can amplify error, distraction, and banality as readily as understanding. Sandberg’s related writing on wisdom matters here: increasing the ability to solve problems does not determine which problems deserve solving.

My present goal is therefore not maximal externalization or maximal automation. It is to build a larger intellectual system that increases what I can notice, understand, remember, and accomplish, which involves pursuing augmentation along all Sandbergian axes. Tensions will exist along this fault line: the biological node should remain privileged. The exoskeleton should make it more capable, not increasingly unnecessary.