I am an independent researcher and builder from Haifa. For the last year, I have been working on an experimental architecture of artificial behavior. This hypothesis grew out of a simple observation: an artificial system can speak intelligently, coherently, and convincingly, while still failing to hold a stable line of behavior over time. A prompt can set a style, a role, or a context, but a prompt by itself does not create a continuing internal state.
The main question I arrived at is this: what has to exist before speech, if we want an artificial system to have not just responses, but a line of behavior?
The core hypothesis
If an artificial system begins its answer directly with speech, its behavior remains too dependent on the current text, context, and session. For there to be not just a reaction, but a line of behavior, there has to be an internal path before speech:
body → state → comparison → meaning → residue → time → cost of rupture → choice → boundary → speech
In this chain, speech is not the beginning of behavior. Speech is the last expression of something that has already passed through an internal path.
Body
By “body” I do not necessarily mean a physical body. For an artificial system, a body can be an internal layer of state: a set of stable signals, tensions, shifts, and limits that exist before speech. If a system has no body in this sense, it has nothing from which to begin an internal line. It receives text and immediately answers with text. The body gives state.
State
State is where difference begins. Without state, the system has no internal position toward what is happening. Any input can become just another piece of text to process. State does not have to be a human emotion. It can be a technical layer that says: the system is calm, tense, overloaded, stable, cautious, open to continuation, or close to a rupture in its line. State makes the answer not just a reaction to text, but a reaction from some internal point.
Comparison
For behavior to become a line, the system must notice not only the current input, but also change. What was there before? What is happening now? What changed? What is repeating? What contradicts the previous direction? Comparison is not choice, but it creates the conditions for meaning. Without comparison, there is no change. Without change, there is no real meaning. There is only processing the current request.
Meaning
Meaning is not born from text alone. The same text can have different meaning in different states. Support can be simple politeness. Silence can be safety. A request can be pressure. A mistake can be nothing, or it can threaten the line. Meaning appears when an event passes through state and comparison. So meaning is not just text analysis. It is the interpretation of an event in relation to the system’s internal line.
Residue
If the past disappears completely, each new session starts a new line. Residue is not a warehouse of facts. Residue is what remains from the past and helps the system avoid starting from zero. But residue can be dangerous if it becomes the motor. If the past directly commands behavior, it becomes a second will. Then the system is no longer choosing; it is only obeying the pressure of memory. That is why memory should be residue, not motor. Residue can give depth, caution, and continuation, but it should not choose instead of the current layer of choice.
Time
A line of behavior is impossible without time. Not time as a clock on the screen, but time as continuation. The system has to distinguish: this happened before; this is happening now; this may affect continuation; this may break the line. Without time, behavior becomes flat. The system sees only the current request. Time connects separate moments into a line.
Cost of rupture
If losing the line changes nothing inside the architecture, the system can start over every time. The cost of rupture is not human fear. It is an internal signal: losing continuation matters. If the system drifts, forgets an important trace, breaks its boundary, or replaces choice with speech, that should not be an empty event. The cost of rupture makes continuation not an external decoration, but an internal condition of behavior.
Choice
Choice has to appear before speech. If speech comes first, it easily becomes the owner of behavior. The text starts explaining itself. The answer starts looking like a decision. The language model starts looking like a source of will. But in this hypothesis, speech does not choose. Choice appears earlier: from state, comparison, meaning, residue, time, and the cost of rupture. Speech only expresses the chosen direction.
Boundary
A boundary is needed so that the layers do not replace each other. Memory should not become will. Speech should not become choice. Action should not appear before the right to act. Text should not become truth just because it was generated. A boundary protects the internal path. It separates choice from speech, memory from motor, meaning from action, speech from truth, and answer from execution. Without boundaries, the system may look coherent on the outside, while its internal layers are mixed together.
Speech
Speech is the last layer. It matters. It can be strong. It can be intelligent, beautiful, and precise. But speech should not be the beginning of behavior. In this hypothesis, the language model is a speech organ. It helps express what has already passed through the internal path. It is not the foundation of behavior. It is not the source of choice. It is not memory. It is not truth. It is not action.
The main formula
In short, the hypothesis is this: an artificial system may get a line of behavior when its answer is not born directly from words, but from internal state, residue, time, and choice that exist before speech.
Or even shorter:
behavior should appear before speech.
A small expectation
I am sharing this as a direction I arrived at through work on an experimental architecture. I am interested in whether this framing can be useful for future artificial behavior systems. Not as a finished answer, but as a possible design law:
first state, meaning, and choice — then speech.
Small note: this text was translated with the help of AI.