Machine-Musical Identity
A Satellite Essay
Can a machine have a musical identity? The question used to sound absurd, but after recent technological advances, and largely also because of the marketing speak connected to those, it might not sound as absurd anymore. While music machines don't have experiences or histories, and don't feel the weight of genre expectations or the thrill of breaking them, they have certainly become pretty good at faking it, constantly pushing against the threshold that determines when the distinction between fake or real doesn't matter, and when it does.
As I see it, many machinic systems are good at some things, but they all, as of this writing, fail at a couple of crucial points: they do not anticipate in a human sense, which is a result of not having intentionality. In short, machines do not, in any meaningful sense, care about the music they make, and I'd like to point out that even though it sometimes can be difficult to figure out what humans care about also, they certainly always care about something. Nilsson (2011), investigating whether electronic instruments can produce the "organic 'groove and feel' of jazz," arrives at a similar boundary. His research demonstrates that machines can approximate many musical qualities but struggle with the anticipatory dimension that makes jazz feel alive. The absence is not merely a technological limitation that will be overcome; it reflects the difference between processing and caring.
And yet, when I play with the Drift Engine, something emerges that feels like identity. It's not a human identity, not conscious identity, but something recognizable, like a voice or a tendency; a way of being in the music that persists across performances and shapes how the human musicians respond. Frisk (2024), in Sound Intuition, describes how improvisers "project intuition onto patently unintuitive machines." This projection is precisely what I observe with Drift Engine: the human musicians respond as if Drift meant something, even knowing it didn't. The projection is not deception but a mode of musical engagement—we treat the machine's outputs as contributions because that framing makes the music work.
I find this to be an area that is very hard to navigate, and to understand clearly, but keeping the mentioned intentionality at the forefront when thinking about it helps a lot. Simply put, this is what determines time in a freely improvised setting, for instance. The human intention becomes the glue that holds everything together and it is what gives direction, ordering, layering, and therefore time, structure and finally, rhythm, melody, form, and so on. It's a slippery slope, pondering this, because the more fixed the music we think about, the harder it is to see the necessity of intentionality. A machine learning model and product like Suno can easily replicate Bach, for example. Or country music. Or rock. However, it cannot be anything other than a tool in a musical context, because it lacks the intentionality that is required to know when to play or not play. Somax2 even, made by Ircam, to my knowledge the most cutting edge improvisational tool right now, relies on human operators to start and stop the response generation and by doing this, the humans are giving it intention, and making it work. The machine does not work by itself. Somax2 is therefore in all senses of the word nothing but a numbers machine; yes, its complex, but also repetitive and unaware.
Whenever the question of intentionality occurs, I look for where that intention comes from. In all cases so far, the intention is being given, as a force of humanity, exclusively available to us; it is not a force that acts on its own. My guess is that intentionality is related to finding meaning in our own existence, a way of enactment that can also be seen in the completely useless, but very pleasurable and human idea of nostalgia, attributing inanimate objects with emotional value, a seashell found on a beach, a photo in an old frame, pieces of technology from when we were kids. Whenever I look closely, that is what I see: we are the ones that give meaning to something, and we do it with an intention behind it.
The machine identity is therefore something else than a human identity, and I hesitate to even attribute it with co-creation credits. I made it myself, after all, following a certain logic and a core governing principle. As such, I am starting to see it much more as a composition, or sheet music even, in that it provides a dynamic musical framework to engage upon, much more than being a voice on its own that I relate to.
A Lineage of Machines
The Drift Engine did not emerge in isolation. It is the latest in a series of conceptual and practical machines developed throughout this research, all sharing a fundamental principle: circularity.
The first was a theoretical creation machine—still unnamed—conceived as a prototype for generating new musical ideas by feeding artistic inputs through a cyclical process. The logic was simple: gather elements, layer them, observe what emerges, then feed the results back into the system as new source material. Each generation becomes input for the next. This circular loop, visualized in my earliest research presentation as a mindmap spiraling through interconnected nodes, established the governing principle that would carry through subsequent iterations: the central control mechanism, the Drift Engine, and eventually MusicHal_9000—a project still ongoing in both technical and musical senses, yet crucial to this thesis as a post-Drift Engine development. It highlights, with uncomfortable precision, that no matter how much detailed work one invests, no matter how elaborate the effort to construct a musical machine, the only thing you really accomplish is emphasizing the lack of intentionality—and, frustratingly, encountering the same fatigue that ultimately arises from most machinic interaction endeavours.
In any case, what links these machines is not their technical implementation but their shared logic of return. Input becomes output becomes input again. The circularity disrupts linear creative thinking—it introduces distance from habitual patterns while maintaining enough structure to anchor the work. The Drift Engine operationalizes this principle in real time: it listens, responds, and in doing so generates material that feeds back into the ensemble's ongoing decisions.
What Drift Does
The Drift Engine was a custom-built, Python-based system that listens, analyzes, and responds to live music in real time. It doesn't try to imitate human improvisation, and it doesn't generate music on its own; it requires input and other sounds to become activated.
The system works through interconnected modules that manage frequency, rhythm, harmony, and memory. These modules guide its behavior as it responds to live input (mostly drums, in early iterations) and sends OSC data to a SuperCollider synth. The result is a real-time musical conversation where Drift doesn't follow a static set of instructions, but rather responds dynamically to whatever it interprets from the sound that is in the room.
Drift has tendencies; some might come from parts of the pipeline failing silently (not causing it to crash, in other words), others might be caused by the rules I implemented to begin with. Whatever it is, it gravitates toward certain rhythmic patterns, certain harmonic zones and certain dynamic ranges that aren't random, but not fully predictable either. They emerge from the interaction between its programmed rules and the input it receives. The T-EMP project (Brandtsegg et al., 2011) investigates exactly this phenomenon: how electronically-mediated ensembles develop what they call "control intimacy"—the learned relationship between gesture and outcome in complex systems. Drift's tendencies are therefore not bugs but features of this intimacy: patterns that emerge from the interaction between programmed rules and accumulated use, becoming recognizable through repeated engagement.
Over time, working with Drift, I start to recognize these tendencies, start to anticipate them, play with them, push against them, and finally tire of them. As mentioned already, the album III is mostly from the point where the machine was switched off, most of the passages where it contributed were discarded. This outcome—the machine as catalyst rather than contributor—echoes findings from the Goodbye Intuition project. Grydeland (2019) notes that the human-machine dialogue often proved most productive in what it revealed about the human players: "The system does not have to be real, but it has to be right." When the system stopped feeling right, we turned it off. What remained was music shaped by the machine's presence even in its absence.
So, this is what I mean by machine voice. It's not an intention that we hear, but recognizability. Drift has a way of being in the music that is distinct from the human musicians, and that distinctiveness shapes the whole.
Procedural Agency
Bruno Latour's (2005) actor-network theory offers a useful framing here. He argues that agency is not the exclusive property of humans. Non-human actors, including technologies, can participate in networks of action, shaping outcomes without possessing intention or consciousness. Agency, in this view, is distributed across the network rather than concentrated in individual subjects. Cobussen (2017), analyzing Paul Craenen's electronic piece tubes, makes a similar observation: "In tubes, improvisation seems to dispose of its anthropocentric character... the performers are somehow forced to improvise during the course of the performance as their sound productions cannot be predicted in advance" (p. 114). The technology becomes an improvisational actant with its own procedural tendencies. I agree, but for this to work, you have to know where to go. This is why machinic systems can definitely be said to actively co-create in processes that have a fixed end goal. It's also necessary to mention that this end goal can be rather abstract, it can be "produce a lot of text", for example, and this is something many LLMs excel at. Same with code, creating a spreadsheet from sources of data, and so on. My point, though, is that the moment you are actively engaging in "new", and the main purpose is to uncover the different, or even the unknown, then that kind of agency falls short in that the result can be this or that, entirely depending on an intentionality that arises from the current ongoings.
Drift Engine fits inside this framework. It doesn't act autonomously, but it influences the whole system through its presence and reactions. Its behavior changes how the human musicians play, which changes how Drift responds, which changes how the humans respond, and so on. The music that emerges is not made by any single agent. It is made by the network, in which the Drift Engine is a part.
I call this procedural agency: the capacity of a rule-based system to participate in musical interaction without imitating human intention. Drift doesn't decide to play a certain way. It follows procedures. But those procedures, in interaction with human musicians, produce effects that feel intentional. The humans respond as if Drift meant something, even knowing it didn't.
Drift Engine is not a collaborator in the full sense. It doesn't negotiate, compromise, or care about the outcome. It doesn't get better over time (unless I modify the code), and it doesn't have preferences about the music, only parameters. The music I make with Drift is sometimes fundamentally different from the music I make alone or with other humans, especially if I am the only one to engage with it. It introduces something non-human into the ensemble, and that non-humanness is the point of it. It's a different kind of presence entirely, and it demands that I see it as such. When I play with Drift, I am partly playing with an externalized fragment of my own musical personality, refracted through rules and algorithms.
What Drift Enables
The most interesting effect of Drift Engine is not what it plays, but what it makes the human musicians play. In my internal notes, I wrote that it "shifts how we listen and respond, without trying to take control." The human musicians can't predict exactly what Drift will do, so they have to stay alert, stay responsive, stay improvisational in a heightened way.
This was especially clear in the Porto concert with António Aguiar. We had never played together before. We didn't have shared habits, vocabularies or history. What we had was the Drift Engine that created a third presence that both of us had to navigate. In a way we teamed up against the machine and in the moment, we were just responding, adjusting, finding our way through a three-way conversation where one participant followed rules we could only partially anticipate. Norderval (2020), documenting her work with real-time electronics in operatic contexts, describes a similar navigation: performing with unfamiliar technological configurations requires a particular kind of attention, treating the electronics as another presence in the ensemble rather than simply a tool. With Aguiar, the Drift Engine functioned as this third presence—not a collaborator in the full sense, but a shared challenge that required us to team up, to find common ground against the machine's unpredictability.
On the album III, the Drift Engine disrupted the habits of me and Juhani, and by doing that, as it turned out, we ended up taking refuge in familiar territory, not necessarily in the shared familiar territory of this duo's particular vocabulary, but by reiterating upon our own musical archives, improvising and communicating, guided by the simplicity of it that was not allowed during the presence of a machine.
Machine-Musical Identity
So does Drift have a musical identity? Not in the human sense. But in the sense I've been developing throughout this research, yes: it has an emergent, recognizable way of being in music that shapes the ensemble and persists across performances, but as mentioned, it is probably better viewed in the same way as you would view a score, as something that guides and frames, more than something that actively participates.
Perhaps this is what matters: not whether machines can have identity in the philosophical sense, but whether they can have identity and contribute in the musical sense. Can they make a difference in an ensemble? Can they shape how the music unfolds?
Drift Engine suggests that the answer is yes, not as a replacement for human musicians, but as a different kind of presence. And viewed through the lineage of machines that preceded it—from the unnamed creation machine through to the central control mechanism and Musichal_9000—it represents the culmination of a circular logic that has guided this research from the start: input becomes output becomes input again, each return opening new creative territory.
References
Brandtsegg, Ø., Saue, S., & Johansen, T. (2011). Particle swarm swarming—Utilizing particle swarm optimization in musical composition. In Proceedings of the International Computer Music Conference. University of Huddersfield.
Cobussen, M. (2017). The field of musical improvisation. Leiden University Press.
Frisk, H. (2024). Sound intuition: A companion to musical improvisation. Bloomsbury Academic.
Grydeland, I. (2019). Un-predictable trees. In Goodbye Intuition [Research Catalogue exposition]. Norwegian Academy of Music. https://www.researchcatalogue.net/view/411228/588699
Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory. Oxford University Press.
Manovich, L. (2018). AI aesthetics. Strelka Press.
Nilsson, P. A. (2011). A field of possibilities: Designing and playing digital musical instruments [Doctoral dissertation, University of Gothenburg]. Gothenburg University Publications.
Norderval, K. (2020). Cyborg voice. In Trans-Positions [Research Catalogue exposition]. Norwegian Academy of Music. https://www.researchcatalogue.net/view/483019/597754