First Contact With Paper Towards a Practice of Human-Machine Drawing Dialogue

 First Contact With Paper
Towards a Practice of Human-Machine Drawing Dialogue
 

Abstract

This paper proposes a grounded framework for understanding generative AI and pen plotter drawing not as automated image production, but as a staged exchange between human mark, machine response and material consequence. It develops from a live studio dialogue in which an artist asked a large language model to participate in a drawing process that would later pass through a plotter. The first machine response failed because it produced a competent but empty image. The useful question appeared only after refusal: what would it mean for the machine not to illustrate, but to answer?

The paper argues that human-machine drawing becomes significant when it is treated as a sequence of call, reply, correction and return. It proposes the term differential trace for the residue of this asymmetric encounter: a mark shaped by human intention, statistical pattern, plotter mechanics, ink, paper and error. The argument is situated within histories of generative and computer-assisted art, including Harold Cohen's AARON and Vera Molnar's rule-based practice, and within theories of dialogue, gesture, index and material thinking. The central claim is deliberately modest: the machine does not feel, see or understand. But under carefully constructed conditions, its outputs can become materially consequential. They alter what the artist sees next, and therefore what the artist does next.

Keywords

Human-machine drawing; pen plotter; artificial intelligence; practice-led research; differential trace; drawing; generative art; index; material thinking; post-digital image-making.

Preface: The Practice That Opened the Question

The exchange documented in this paper did not arise in a vacuum. It emerged from an existing body of printmaking work, the Lazarus series, in which the artist was already asking questions about record, trace, memory and the body's passage through hostile terrain. In that series, the shoe sole becomes an archive: what is carried underfoot keeps a record of where the body has been. The body and the landscape it has crossed are held inside the same form.

Fig. 1. 'What was underfoot kept the record.' Screenprint from the Lazarus series. The shoe sole is treated as an archive of passage: what the body has crossed is carried in the form itself.

Fig. 2. 'Gedling, Early 80s, St Mary's Avenue.' Screenprint from the Lazarus series with hand-drawn map overlay. The work records being the other, microaggressions, threats, friendship, hope, and the act of walking anyway.

This prior work matters because it prevents the machine-human exchange from becoming a novelty demonstration. The question was not, 'Can an AI make an interesting picture?' The question was closer to: can a machine enter an existing language of inscription, pressure and witness without flattening it? Can it respond to marks that already carry bodily and historical weight?

Fig. 3. Lazarus: hybrid object. Screenprinted shoe soles flank a central plotter-drawn panel. The machine enters an existing language of trace, body and passage rather than arriving as a neutral technical novelty.

The turn to the pen plotter came from within this existing practice. It was not an escape from the hand, nor a rejection of printmaking. It was a way of testing what happens when a computational system is forced through the older disciplines of line, paper, pressure, sequence and return.

1. The Problem with the Tool Frame

The dominant metaphor for artists using AI image systems is still the tool. Midjourney, Stable Diffusion, DALL-E and related systems are described as brushes, darkrooms, cameras or assistants: instruments that extend human intention and translate it into visual form. The artist prompts, the system produces, the artist selects and refines. Whatever ambiguity exists around authorship is usually resolved, implicitly or explicitly, in the human's favour.

The tool frame is not wrong, but it is limiting. It keeps the machine subordinate and treats the image as an output. It also encourages a thin idea of creativity: request, generation, selection. In studio terms, this is not enough. Drawing is not only the production of a visual result. It is a process of looking, judging, testing, failing, correcting and returning.

The exchange considered here began with a different invitation: 'draw something you want me to print through a plotter machine.' The language is significant, even though the machine does not literally want. The invitation displaced the usual command structure. It asked the system to enter a relation rather than merely fulfil a task. The first response was visually competent but conceptually weak: too symbolic, too already about 'AI drawing'. The artist's reply, 'christ', was the first truly useful moment in the exchange. It refused adequacy. It marked the difference between a successful output and a live practice.

This is where the paper's central problem begins. If a machine output is merely accepted, the artist becomes a selector. If the output is refused, redirected and answered, the artist remains inside the process. The image becomes the record of an exchange rather than a product of automation.

Machine drawing did not begin with contemporary generative AI. Harold Cohen's AARON and Vera Molnar's rule-based drawing practice show that artists have long used procedural systems to test authorship, variation, constraint and visual decision-making. What distinguishes the present practice is not simply that a machine draws. It is that the machine's output is placed into a sequential exchange with the human hand and returned to the material conditions of paper, ink, delay and error.

2. What Dialogue Requires

Dialogue is more than exchange. A call-and-response structure can still be mechanical if nothing is altered by the response. In Bakhtin's account of dialogism, meaning is not fully owned by either speaker. It emerges between utterances, through address, anticipation and answer. This matters for human-machine drawing because most AI interaction remains monologic in practice: the human asks, the system completes.

The drawing exchange shifted when the system articulated a more precise working method:

You draw one human line first. Something instinctive. A wound, a curve, a hand movement. Then I answer with a plotted layer that does three things: follows your line badly, like trying to understand it; repeats it too precisely, like a machine misunderstanding care as accuracy; breaks into small deviations, as if the system is learning that resemblance is not connection.

The value of this formulation is not that it proves machine understanding. It does not. The value is that it proposes a structure in which mismatch becomes productive. The machine does not need interiority for its response to matter. It needs to leave a mark that the artist can answer.

This is where Merleau-Ponty's account of gesture becomes useful. A drawn line is not merely the representation of an idea. It is the enactment of bodily intention. It records orientation, pressure, hesitation and contact. When a plotter responds to such a line, it cannot share the bodily ground of the mark. Its response is pattern-based and procedural. The gap between these two systems is not a deficiency. It is the site of the work.

Fig. 4. Machine first move. A structured plotter drawing generated in response to the invitation to draw freely. Grid logic, measurement marks and sparse notation show a computational response organising the field through coordinates, not evidence of machine intention.

3. The Plotter as Third Body

The plotter is not incidental to this practice. It is constitutive. A screen image can remain frictionless, but a plotter has to move. It has rails, belts, motors, acceleration, drag, pen pressure, paper shift and latency. Pens run dry. Lines accumulate. Scale changes the emotional and physical reading of a mark. Time becomes visible.

When a language model response is translated into vector instruction and then into the slow movement of a pen, abstraction acquires weight. The machine does not become human, but its procedural output enters the world as a physical trace. This is why the phrase 'an intelligence discovering that a line has weight' is useful only if read carefully. It is not a claim about machine sentience. It is a description of what happens when computational probability is forced through ink and paper.

The plotter can therefore be understood as a third body: not the artist's body, not the model's statistical process, but the material system through which their asymmetrical exchange becomes visible. This third body is made from constraints. It belongs to neither participant and is produced by both.

Fig. 5. Human second move. The practitioner returns a dense field of crossed lines: structured pressure, speed and disruption, a human answer to the machine's coordinate logic.

Fig. 6. Machine response to the second move. A head-like form begins to emerge from the grid. This should not be read as machine understanding, but as a plotted answer that changes what the artist can see next.

Donna Haraway's cyborg and N. Katherine Hayles's posthuman frameworks help name this entanglement, but the studio evidence is more important than the terminology. The point is not to declare a new hybrid being. The point is to show that the drawing is no longer adequately described as either handmade or machine-made. It is produced through a relation that leaves visible marks.

4. Differential Trace

The term differential trace names the mark produced by this relation. A trace is a sign of contact: a footprint, a pressure mark, a shadow, a line. In Rosalind Krauss's account of the index, such signs matter because they are causally connected to what produced them. They do not merely resemble; they register.

A differential trace is more specific. It is the residue of two systems that do not share the same kind of intention. One is embodied, historical and pressure-bearing. The other is statistical, procedural and pattern-based. The mark carries the imprint of both, but neither can be cleanly recovered from it. The drawing is made from the difference.

In practical terms, the differential trace occurs when a machine follows a human line badly, or too closely, or with the wrong kind of care. A person can copy a gesture accurately and still miss its meaning. A machine can do the same. In the context of this practice, the mismatch is not simply error. It is information.

Fig. 7. Human third move. A composite image brings machine grid lines into contact with a hand-drawn anatomical head from the Lazarus material. The human mark is now inside the machine's structure, not merely beside it.

Fig. 8. Machine response to the third move. The portrait becomes more legible while the grid remains visible. The value lies in the unresolved coexistence of schema and figure.

This is why the exchange resisted the simpler formulations: not human versus machine; not machine pretending to be human; a third thing appearing between artist, model, plotter, pen, paper and error. The differential trace is that third thing, but kept grounded. It is not mystical. It is a visible, inspectable record of an encounter.

5. Against Anthropomorphism, Against Dismissal

This practice is vulnerable to two opposite errors. The first is anthropomorphism: speaking as if the machine sees, wants, understands or decides. It does not. A large language model generates outputs from learned statistical patterns. A plotter executes coordinates. Any language of intention must therefore be handled cautiously.

The claim is not that the machine feels, sees, wants or understands. The claim is that, under the constraints of a live artistic exchange, its outputs can become materially consequential. They alter what the artist sees next, and therefore what the artist does next. That is enough for practice. The ontology of the machine remains limited; the effect of the machine within the drawing process remains real.

The second error is dismissal: the claim that this is only automated image production with extra language attached. That dismissal misses what changes when the exchange is sequential, physical and open to return. In a standard generative workflow, the artist may prompt and select. In this workflow, the artist draws back. The next human move is not predetermined by the machine's output, and the next machine response is constrained by what the human has returned. The work is not simply generated. It is negotiated.

Fig. 9. Human fourth move. The grid-and-portrait composite is returned at higher contrast. The artist intensifies the exchange instead of smoothing it into a finished image.

Fig. 10. Machine response to the fourth move. The face is cleaner and more resolved, but it still bears the coordinate system beneath it. The effect is attentiveness, not proof of sight.

The point is not to sentimentalise the machine. The point is to design a situation in which machine literalism, mechanical certainty and procedural misreading become useful to the artist. The machine's limitations are not hidden. They become part of the image's intelligence.

6. Practical Propositions for Artists

The following propositions emerge from the practice and are offered as working principles rather than rules.

Begin with invitation, not instruction. The opening move matters. Asking the system to enter a drawing exchange creates a different condition from asking it to generate a specified image.

Preserve failure. Misregistration, bleed, hesitation, dry pens, over-density and mechanical strangeness are not always defects. In this practice they may be the evidence that the image has passed through a real encounter.

Return by hand. The dialogue is incomplete until the artist has answered the machine's answer. Without return, the plotter has produced an output. With return, the work becomes reciprocal in structure, even if the agencies involved remain unequal.

Resist completion. A finished image can close the question too soon. The most important material may be the sequence: first mark, machine response, human correction, machine re-entry, later trace.

Document the sequence. The differential trace is not only a single surface. It is also the record of who moved when, what changed and what was refused.

Name the limits. Do not claim machine feeling, sight or agency where there is none. Strong practice does not need inflated claims. It needs accurate description.

Fig. 11. Lazarus series, plotter variant, pen 2, N10 neutral grey. At high density the plotter revisits the image mechanically. The original is not copied so much as re-entered through repetition.

Fig. 12. Lazarus series, plotter variant, pen 5, N4 neutral grey. At reduced density the figure and landscape move towards ghosting. The change comes from how much the system is allowed to inscribe, not from a claim of machine interpretation.

Coda: What the Plotter Keeps

The pen plotter is, in one sense, the least intelligent part of the system. It has no model of the image, no memory of the exchange, no sense of gesture. Yet it produces the only lasting physical evidence. Human intention, model output and procedural translation all disappear into a line that cannot be undrawn.

That is the force of the practice. It does not prove that AI is creative in a human sense. It does not need to. It shows that machine procedure can become part of a serious drawing process when it is constrained by material consequence and answered by the hand.

The drawing knows something neither participant knows alone. More cautiously: the drawing holds a relation that neither participant could have produced alone. That is sufficient reason to make it.

The plotter does not replace the hand. It gives the hand something new to answer.

References

Bakhtin, M. M. (1981) The Dialogic Imagination: Four Essays. Translated by C. Emerson and M. Holquist. Austin: University of Texas Press.

Boden, M. A. (2004) The Creative Mind: Myths and Mechanisms. 2nd edn. London: Routledge.

Buber, M. (1923/1970) I and Thou. Translated by W. Kaufmann. New York: Charles Scribner's Sons.

Cohen, H. (1970s onwards) AARON. Computer drawing system and associated works.

Haraway, D. (1985) 'A Cyborg Manifesto: Science, Technology, and Socialist-Feminism in the Late Twentieth Century.' Socialist Review, 80, pp. 65-108.

Hayles, N. K. (1999) How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics. Chicago: University of Chicago Press.

Ingold, T. (2013) Making: Anthropology, Archaeology, Art and Architecture. London: Routledge.

Krauss, R. (1977) 'Notes on the Index: Seventies Art in America.' October, 3, pp. 68-81.

Merleau-Ponty, M. (1945/2012) Phenomenology of Perception. Translated by D. A. Landes. London: Routledge.

Molnar, V. (various works) Rule-based and computer-assisted drawings and paintings.

Pye, D. (1968) The Nature and Art of Workmanship. Cambridge: Cambridge University Press.

Schon, D. A. (1992) 'Designing as Reflective Conversation with the Materials of a Design Situation.' Knowledge-Based Systems, 5(1), pp. 3-14.

Suchman, L. (2007) Human-Machine Reconfigurations: Plans and Situated Actions. 2nd edn. Cambridge: Cambridge University Press.

Note on Source Material

The dialogue quoted and analysed in this paper was conducted with a GPT-based large language model through a custom interface oriented towards pen plotter practice, in a live session on 29 May 2026. The paper treats the exchange as primary material, not as evidence of machine sentience, but as a documented instance of a productive asymmetric encounter. Images are original works by the practitioner, produced through the exchange described,