Core Primitive
Signal is not a property of information. It is a relation between a piece of information and a decision you own. Gregory Bateson defined information as "a difference that makes a difference" — and the operational version is sharper: signal is a difference that makes a difference to a decision that is yours to make. The same paragraph is signal to one reader and noise to the next.
The definition everything else hangs on
In 1972, the anthropologist and systems thinker Gregory Bateson compressed the entire problem of information into one sentence: information is "a difference that makes a difference." A temperature reading is a difference. Whether it makes a difference depends on whether anyone is deciding anything — to dress differently, to adjust a thermostat, to evacuate. The difference alone is just physics. The making-a-difference is the information.
Everything this phase has taught about noise sharpens under that lens. Most information is noise (Most information is noise) not because it is false or badly made, but because it makes no difference to anything you decide. Signal requires a defined goal (Signal requires a defined goal) because without a decision in view, there is nothing for a difference to make a difference to.
Signal is a relation, not a property
This is the move most filtering advice misses. People audit their sources, upgrade their newsletters, and curate their feeds as if signal were a quality grade stamped on the content — as if there were signal-bearing articles and noise-bearing articles, and skill meant telling them apart.
There are no signal-bearing articles. There are only articles in relation to readers, and the relation runs through the reader's commitments. The same earnings report is signal to the analyst covering the company, noise to the analyst covering its supplier's competitor, and signal again to the founder deciding whether to sell to them. Nothing about the report changed.
The decision-you-own clause
Bateson's definition needs one operational tightening for daily use: signal is a difference that makes a difference to a decision you own. Not a decision someone owns. Yours.
The news is full of differences that make enormous differences — to central bankers, to generals, to executives you will never meet. Consuming those differences feels like participation in consequence. It is not. If the decision is not yours, the information cannot function as signal for you, however much it functions as signal for someone. This is Epictetus's dichotomy of control operating as an information filter, two thousand years before feeds: attend first to what is yours to control, because only there can a difference make a difference through you.
Postman's test: the information-action ratio
The media critic Neil Postman gave the relation its cheapest diagnostic. He asked of any piece of information: what will you do differently because of it? He called the underlying measure the information-action ratio — how much of what you consume ever alters an action.
As a filter, it is brutal and fast. What would this change about what I do? If the honest answer is nothing — not "nothing yet," not "it's good context," but nothing — it is noise, whatever its quality. The test takes three seconds and requires no source rankings, no reading-list hygiene, no willpower. It requires only that you know your decisions.
Why "know your decisions" is the hard part
The test's difficulty hides in its precondition. Most knowledge workers cannot list the decisions they own. Asked, they produce responsibilities ("I own onboarding") or aspirations ("deciding our AI strategy") rather than decisions — choices with options, a chooser, and a date.
Until the decisions are explicit, the relation has nothing to bind to, and every input defaults to "possibly relevant." That default is the noise firehose's entire power supply. An unstated decision portfolio makes the whole world look like signal.
The classification is allowed to change
Because signal is a relation, it moves when your decisions move. The vendor comparison that was noise in March becomes signal in September when the contract renewal lands on your desk. This is not filter failure; it is the relation behaving correctly.
The practical consequence: filtering is not a one-time purge but a standing function of your current decision set. When a decision closes, a whole class of inputs quietly demotes to noise. When one opens, formerly ignorable differences promote themselves. A capture system that records which decision an input touched inherits this for free — retire the decision and its inputs retire with it.
What this does to guilt
The relational definition dissolves a specific ache: the guilt of not keeping up. There is too much worth reading — for somebody. The library's size is not your reading list's size, because the library's differences mostly make differences to decisions that are not yours.
Declining to process a difference that makes no difference to anything you steer is not ignorance. It is the definition, applied.
Protocol: the relevance gate
Make the definition a gate rather than an insight. At the point where information enters your system — inbox, capture note, reading queue — ask the two questions in order. Mine to control? What would it change about what I do? Two passes admit the input for processing. Two failures release it. One of each marks it ambiguous, and ambiguity gets a decision attached or a short expiry, never indefinite storage.
This gate is deliberately cheap and deliberately generous — it runs constantly, so it must cost seconds, and at the entry layer a missed signal is the expensive failure. The next lessons build the levels above it, where the standards tighten.
Sources
- Bateson, G. (1972). Steps to an Ecology of Mind — "a difference which makes a difference."
- Epictetus. Enchiridion — the dichotomy of control.
- Postman, N. (1985). Amusing Ourselves to Death — the information-action ratio.
Put it into practice
Exercise
Write down the three to five decisions you actually own right now — decisions where you are the one who chooses, not an observer. Real ones, with dates: "choose the Q4 architecture by the 15th," not "stay informed about AI." Then take the last ten pieces of information you consumed today and test each one against the list: does this change any decision on it? Two questions expose the relation — Epictetus's first (is this mine to control?) and Postman's second (what would this change about what I do?). Anything that fails both is noise for you, no matter how interesting, well-written, or true it is.
Watch for the failure mode
Treating "interesting" or "true" as synonyms for signal. Interesting is a property of the information; signal is a relation to your decisions. The failure looks like a well-read person who can discuss everything and decide nothing better — they optimized for quality of inputs while never asking whose decisions those inputs serve. High-quality noise is still noise, and it is the most expensive kind, because it survives every quality filter you build.
Make it stick
Put your decision list where your filters can see it — the top of your notes inbox, your capture template, wherever information enters your system. For the next week, classify inputs against it at the moment of capture: which decision does this touch? If the honest answer is none, let it go without guilt. You are not ignoring the world; you are declining to process differences that make no difference to anything you steer.