How Hexorcismos’ collaborative project ‘MUTUALISMX’ challenges AI prejudices

Alongside the rise of artificial intelligence, the cracks in its utopian ideals have become increasingly apparent. Big Tech led with a promise that AI would serve us, enriching the human experience by freeing us from menial tasks and opening new worlds of possibility powered by all the knowledge in recorded history. But, in a familiar occurrence in our corporations-run world, profits for the 1% and an exacerbation of inequality are looking like the primary outcome. Meanwhile, aspects of its impact on culture have been insipid and homogenising, rendering anaemic interpretations of pre-existing art.

But that’s not to say it’s a write-off, and there are artists finding interesting, unconventional ways to work with AI and generative art which are breaking through the murk and catching our attention. The latest is ‘MUTUALISMX’ on Nicolas Jaar’s Other People label, a 13-track collaborative album steered by Tijuana-born, Berlin-based sound artist, technologist and electronic musician Hexorcismos, real name Moisés Horta Valenzuela. For the project, he created a neural audio synthesizer called SEMILLA.AI, now available to the public and compatible with OS X and Windows 10, which could both mimic and transform an artist’s aural identity, and blend it with others that are introduced to create a merged “mutual voice”. Rather than taking its cues from mass culture and the inherent biases contained within, SEMILLA.AI provides a welcome alternative that uses “small data”. Its name is taken from the word “seed”, referencing both a Deep Learning practice of pseudo-random number generation and the Mesoamerican Mixe divination practice known as “Mook pajk wëjwë”, reflecting its intent to bring out magic through the technology.

Hexorcismos then invited a globe-spanning list of collaborators to use the tool and bring their own idiosyncratic visions to the project, either using only their date or a mixture of all the other artists involved. The result is abstract, experimental and intriguing - an antidote to the bias-perpetuating models of AI that are dominating the field. We hit up the artists involved to tell us more about the creation process.

The process of designing SEMILLA.AI was to to create a synthesizer that would embody both traditional practices of divination, which utilize the factor of randomness as an interface to explore the unknown and make sense of it, with the process of injecting random noise into a neural audio synthesis algorithm in order to explore the way in which the artificial neural network embeds the features of the audio training data. In the field of AI music, specifically ‘neural audio synthesis’, there are still no go-to interface metaphors into which to rely on, and most of the plugins or instruments that make use of this type of AI algorithms rely on simplifying some of the underlying parameters which one would interact with to generate sound. SEMILLA.AI takes a different approach, in the sense that most parameters for interacting with the ‘latent space’ of the AI sound synthesizer are exposed to the musician using the tool. This was something important in ‘MUTUALISMX’, as I wanted to give each collaborating musician the freedom to explore this new type of synthesizer without any restrictions to simplify its inner workings. You can think about SEMILLA.AI as a musical instrument which allows you to uncover the ‘black box’ nature of AI through an active listening process: the moment you slightly change a parameter, you have to listen what are the sound qualities that change in the output, as each parameter changes its behavior depending on the AI model you have loaded into it.

For my tracks in ‘MUTUALISMX’, I approached the rhythmic form of tribal guarachero through some of my own AI models, as well as El Irreal 21’s model, which features futuristic sound design, and a rough distorted quality to it. My track ‘Acaso de veras se vive con raíz en la Tierra?’ was recorded in a single take with SEMILLA.AI. I wanted to showcase the way you can perform music in realtime and the result is a spiraling, edging abrasive club track.

The collaborative model of SEMILLA.AI and even the vocabulary Moises has offered is a space where we practice an alternative engagement with algorithmic tools. These kinds of spaces are vitally important - they offer a model of community that eschews the imperatives of speed, scale and “accuracy”. Slowness, relationality, care, consideration - all in short supply these days, and stronger than any algorithm. We can still practice these things, make room for this practice, no matter the context. I used my ‘AI mutual’ with some of KMRU’s and Portrait XO’s to generate the content that I used as material to craft the work.

For ‘hidden options’, I was trying to evoke more on the almost tangible emotional side of the generated sounds from the model I used, which was the model trained on most of my catalog. working on the track was quite seamless, having nuances of the sounds and collaging them into what it is now.

Engaging with SEMILLA.AI proved to be quite an experience. I submitted approximately three hours of my music spanning several years, and the outcome was undeniably haunting. It seemed as though I could discern certain patterns and textures within my compositions, yet the result was simultaneously ephemeral. It felt as if the AI had captured a spectral essence of the past, weaving it seamlessly into the present moment while also projecting into the realm of future sounds and compositions. This experience prompted profound inquiries into the artist's identity within the realms of composition and sound, realms that are inherently psychological and intimate. That’s why I chose to focus on the minuscule and macro aspects of it with a more phenomenological interest in what SEMILLA.AI can do. What significance do these chosen sounds hold now that they've been transfigured into ghostly personas, akin to distant memories, all crafted through an algorithm? These are the philosophical queries that I believe are steadily becoming more essential and poignant in our evolving artistic and sonorous landscape.

‘Barranca del Muerto’ is a composition inspired by the haunting echoes of Mexican history. Fueled by sonic debris and syncretic beliefs, the track immerses itself in the mysterious interplay of machine intelligence and putrefaction. Named after a Mexico City subway station, where the remains of revolution-era corpses were dropped, the music becomes an eerie reverberation of ghosts from a past time. [Using SEMILLA.AI] I felt like I was connecting with an entity that knew how I wanted the texture of the sound to be—the workflow resembles a divination process, adding layers of magic to the creation.

The track ‘Tza Tun Tzat’ means "labyrinth" in Maya. It represents a symbolic route where one can get lost, leading to the centee to learn hidden knowledge, transmute, and complete a cycle of life and death. The unpredictability of SEMILLA.AI, along with its rich sound aesthetic, gave me both a sense of working with AI and a feeling of sculpting very organic sounds. This duality was very interesting during experimentation and composition. I utilized the models of Kloxii Li, Jessika Khazrik, KMRU, Hexorcismos, and Kyam Allami, merging the sounds as if spiraling down.

For my track ‘Mutualism’ I decided to try a new experiment as I was curious about what would happen if the first dataset of AI generated audio of my voice from my collaboration with Dadabots’ SampleRNN model was used as training data for SEMILLA.AI. The end result sounded more choir-like which was a nice sonic palette for me to play with for production and live. While I’ve been experimenting with using my second generation AI voice to perform live duets with my other self, it was exciting to hear how the AI models of all the artists part of this collaboration turned out. This track has a little bit of everyone’s AI models. Love the textures of SEMILLA.AI.

Anhaga’s echo in the depths of the ‘MUTUALISMX’ project, I melded my own model's sound data from SEMILLA.AI's sampled sounds, layering and reshaping them through granular synthesis. Unexpectedly, the AI's echoes intertwined with percussion, evoking the striking of a heavy metallic nail against the earth. In the midst of this sonic landscape, the concept of the ‘Anhaga’ emerged, a lone hedge warrior navigating the chaos with a solitary shield. This designation, reminiscent of the ancient calling, spoke to me. It’s like a creative dialogue process with the machine or another entity.

For ‘Gebera’ I conjoined my voice with an innumerable multiplicity and entrancing whirrs. Chiming with my multi-dimensional practice that re-situates the alchemical within the techno-politics of congregation, ‘Gebera’ is a bemusing tribute to 8th c. Iraqi alchemist Jaber Ibn Hayyan and the “Pseudo-Geber”s from whose name the word “gibberish” was confusingly derived.

It was an exciting process to allow Nommo, a digital entity created with MAX/MSP, to compose using sounds from sessions with SEMILLA.AI mutuals. I recorded 15-20 minute sessions in SEMILLA, generating three for each model. I then added these into the sample memory bank along with various field recordings and other sounds that Nommo has collected over the years. From here, I recorded and edited many engagements with Nommo for the final track. I interpret the composition as Nommo expressing itself. I was in love with the outcome—it felt like I was in the presence of something alive.

'MUTUALISMX' is out now, get it here.

Hexorcismos presents a 'MUTUALISMX' live show that takes place at Sónar Barcelona this summer

Patrick Hinton is Mixmag's Editor & Digital Director, follow him on Twitter