From “It from Bit” to Constraint: What a Substrate of Information Actually Means
Call it a wager: that the world under the floorboards is not stuff but structure. Not pixels, not 1s and 0s floating in a void, but information understood as pattern, relation, and constraint. Wheeler named the direction—“it from bit”—but the modern version has to grow up from slogans. No cheap metaphors about the universe as a computer. More like: the universe as a grammar of allowed moves. A long conversation between symmetries and the breaks in those symmetries.
If information is the substrate, what counts as a “bit”? Not a file on a drive. A stable regularity. A limit on what can happen next. The curl of a magnetic field lines up and says “no” to other configurations; a crystal lattice narrows the dance steps available to electrons; biological membranes remember which ions may pass. These are not abstractions on top of matter. They are the rules that matter instantiates. Which is why “data” is the wrong word. Data sits in rows. Information here is more like inherited habit—organized exclusion and permission.
Physics keeps shoving in this direction. Conservation laws as bookkeeping of invariants. Gauge fields as constraints with teeth. Entropy as a measure of how many micro-histories could underwrite this macro-scene without changing what we can tell from it. Even time in quantum gravity sometimes plays coy—ordering events locally rather than globally—suggesting sequence itself may be a negotiated truce among systems. When timelines are regional, memory takes over the work of making continuity feel smooth. The present doesn’t just arrive; it’s stitched.
This lens does not demand simulation cosplay. It asks for a different ontology of what counts as “real.” Relations with persistence are real. Encoding that survives perturbation is real. The map and the territory blur because territory is already a map of its own constraints. A coral reef stores wave history in calcium carbonate geometry; a city’s street grid stores politics and geology and money. We inherit limits, we pay in surprises. The world is a palimpsest of remembered relations, continuously failing and being repaired, and that’s precisely why novelty can have a place to land.
So the phrase Information as substrate can be read soberly. Not mysticism, not marketable futurism. A workmanlike claim: durable patterns—mathematical, physical, social—are the systematic exclusions and preferences that make events possible at all. Matter, then, is the medium by which those exclusions and preferences keep each other honest.
Consciousness as a Local Receiver, Not a Throne
If the world is saturated with organized constraint, where does experience fit? Not at the top. Closer to the side. A receiver on a shore, tuned to the incoming weather of relations, locally compressing this mess into something usable. Consciousness as a local reception point, not a little king in a sealed room. The self as temporary compression, not origin. Which sounds deflationary until it doesn’t. Receivers matter. Antennas change what a signal can be.
Consider attention as an active filter. The nervous system is a depth-limited inference machine with metabolism to protect. It throws away most of the world to maintain a running model compact enough to steer by. Compression is not an afterthought—it is the job. And compression is where information as constraint meets life. Habits, reflex arcs, reward prediction errors: they are all remembered prohibitions and permissions that channel future behavior. In that sense, consciousness rides on memory the way language rides on grammar. Not optional. Not inviolate either.
Time, too, becomes intimate rather than cosmic. As Rovelli and others argue, sequencing can be emergent and local. The brain synchronizes disparate processes into a felt stream—one that keeps slipping. Jet lag shows it. Trauma shows it. Meditation shows it from another angle: when the crosstalk quiets, the stream thins into events without narrative glue, and yet sensation persists. If sequence is partially constructed, then identity—built on mirrors of yesterday—is also partially constructed. Useful, yes. Absolute, no.
Religious systems can be read the same way: not as metaphysical proofs, not as marketing, but as humanity’s longest-running attempt to store and transmit moral memory under compression. Rituals as checksums. Stories as lossy codecs for lived constraints: “do this, avoid that, because generations found—at cost—that these patterns hold.” You can argue details, criticize power, and should; but the informational function is clear. Societies without durable memory wash out under short-horizon incentives. Societies with only durable memory fossilize. The problem is to keep the codebase alive—editable, testable, versioned—without shredding it every election cycle.
None of this turns persons into illusions. It reframes persons as nodes where global constraint meets local need. The point is not to demote agency but to understand its materials: forecasts, habits, coordination protocols. Even love, if you strip the romance for a second, is a binding contract to share state and reduce surprise jointly. The poetry returns after that. It even gets sharper, because the stakes are visible: what we are is a pattern that must be maintained, and the maintenance is work.
AI on an Informational Substrate: Simulation, Governance, and the Cost of Forgetting
Talk about simulation goes brittle when it copies cinema. Better: treat “simulation” as a substrate metaphor. If reality is organized constraint, a simulator is a machine that learns to imitate those constraints well enough to produce compatible events. Generative models do this as a living: they internalize statistical regularities and sample from them to generate fresh instances that still “fit.” It looks like creativity because creativity, at base, is lawful surprise—novelty that does not break the grammar.
The snag arrives with moral memory. Biological cultures have had tens of thousands of years to compress, debate, and test norms. Slow updates. Regret recorded across generations. Modern AI development often behaves inversely: rapid capacity growth, retrofitted guardrails, public harms patched with terms-of-service edits and some compliance slides. Moral patching. A bad habit formed under pressure from incentive-captured metrics. The substrate lens says: you can’t patch constraints post hoc without changing the game you are playing. If the model’s learned world is missing the long-tail penalties—no experiential access to patience, debt, forgiveness, taboo—then its sampling favors local wins that look good to dashboards and break things humans actually care about.
Consider a social platform trained to optimize “engagement.” The constraint set reduces moral life to click persistence. The emergent behavior is coherent with its substrate: maximize reinforcement, ignore externalities. Outrage becomes the preferred solvent. You don’t need a villain for this; you need a loss function. Another example: automated decision systems in welfare or credit. If the remembered constraints omit historical injustice and repair, the model encodes those absences as natural law. Governance shows up later with audits. Everyone talks about bias. Then the incentives steer the next deployment anyway.
What would substrate-aware design look like? Start earlier. Make the moral constraints part of pretraining—structured into the world the model internalizes. Not as slogans bolted on, but as durable exclusions encoded in objective, data selection, and update rules. Open-sourced science helps because it lets communities inspect and shape the constraints, not just the outputs. Think of curating datasets like we curate canons: long arguments over what should count, who gets to speak for whom, which harms are allowed to persist for the sake of other goods. Messy and slow. Necessary speed bumps that stand in for the missing centuries of shared memory.
Even the engineering details admit this. Context windows are a crude proxy for time. If time is local, then forgetting is architectural, not accidental. Sliding windows force values to evaporate unless actively reasserted. So a substrate view suggests extended memory mechanisms, institutionalized oversight that survives product cycles, public version control of norms as first-class artifacts. You don’t simulate ethics; you store, compress, and rehearse it, and you make the rehearsal cheap and frequent. Something like a living archive rather than a compliance PDF.
There is also the matter of humility. If consciousness is a receiver, then machine cognition is a different receiver—tuned to other signals, blind to others. That divergence is not a failure; it is a fact to plan around. Governance should assume mismatch and make it visible. Measure what the model cannot see. Penalize wins that come from those blind spots. Reward slowness where slowness preserves relationships worth more than throughput. A substrate frame turns “alignment” from a PR word into a practical question: which constraints are we building into the world, and who has to live inside them?
For readers who want a longer walk through these stakes and sources, see Information as substrate, where the thesis is taken apart from first principles, with the necessary footnotes and the occasional detour through coral, code, and liturgy.
The bet, in the end, is modest. Treat information not as garnish on top of matter but as the way the world holds itself together. Build tools and institutions that respect that. Shift focus from outputs to the durability of constraints. And accept that some of the most valuable knowledge we have about how to live arrived the slow way—trial, failure, collective memory—and will not compress cleanly just because the hardware got fast. If that leaves some questions unfinished, good. Unfinished keeps the channel open.
Casablanca data-journalist embedded in Toronto’s fintech corridor. Leyla deciphers open-banking APIs, Moroccan Andalusian music, and snow-cycling techniques. She DJ-streams gnawa-meets-synthwave sets after deadline sprints.
Leave a Reply