Some thoughts come at night and leave before you can share them. One of them I caught with a gesture. Then I let a machine gather it with me: it kept a part, and let another slip away. Three moments of a single thought passing from hand to hand — and a question, perhaps everyone's: when we hand machines what we think, how much of us will remain?

A study on the border between thought and transcription

Nightly Thoughts is a visual study on that thin border: the point where a thought, in order to exist outside us, must be transcribed. It begins with an automatic sketch — a continuous, pre-verbal gesture — and follows its fate once handed to a machine that can read it only in discrete units. Across three states of the same image, the study explores what is preserved and what is lost in the passage from continuous to discrete.

The process

Everything begins with an automatic sketch, reworked digitally as a luminous web on a black ground. On the original gesture, an AI-generated point elaboration was produced: the translation of the hand's continuous line into discrete units. The two components — the human line and the AI's points — were then fused in Photoshop using different blend modes, yielding three states of the same mark: the original gesture; the coexistence of line and points; the points overtaking the stroke.

The three states

The gesture — Still all mine: a thought held an instant before it vanished. The mark in its original form: a continuous line, drawn before intention or word step in. This is the territory of the Non-Computable Gesture — the mark that coincides with experience, before any translation.

Together — I let the machine touch it. For a moment, we hold it together. Hand and machine work the same mark: line and points coexist. Human and artificial overlap without one yet prevailing over the other — a momentary balance, almost a collaboration.

What is lost — The machine keeps its shape. The most alive part, it does not. The points take over and the line withdraws. The machine keeps the structure, the skeleton of the mark; but the living part — the continuous flow, the tremor of the hand — is left behind.

The concept: continuous and discrete

A machine cannot take in a continuous gesture as it is. To process it, it must translate it into discrete units — points, and ultimately tokens, the same way a language model breaks sentences apart or a digital image resolves into pixels. In this translation, the machine can only ever grasp a subset of the thought: the part that can be reduced to data. What escapes the grid — the continuous, the living nuance, what I call the non-computable — is precisely what is lost.

This is not a flaw in the method: it is its point. Discretization does not erase the non-computable; it makes it visible by difference. Looking at what the machine keeps, we see, by contrast, what it could not take. The shape remains; the life, in part, does not.

The question

Machines will learn to receive ever-larger portions of our thought. Yet however far they grow, they will always grasp only a subset: the rest will keep escaping them. That is why the study's question does not age; it only grows more urgent: when we hand machines what we think, how much of us will remain? Nightly Thoughts does not claim to answer it. But it points to one certainty: a thought is also what no machine will ever be able to hold.

Where to see the project The full project is on Behance- in the reel section @Markthought on instagram