Words by Sean Victory
Grieving and loss are deeply personal experiences, but the experience itself is one we all have in common. A newly empty space is created in our lives, where so little rushes to fill the vacuum, no amount of knowledge heals the wound, and there is only the ache of time. What was begins to take on a life of its own, a luminosity shining from through the hallowed past to light a newly hollow present.
On GHOST, Asa Horvitz collaborates with musician and machine, pushing past personal tragedy in order to connect more deeply with the ever-living world of the dead. Inspired by the loss of close family and friends, Horvitz sought a common thread, not to magnify death in a singular instance, but to pull at human history’s inevitable surrender to mortality. Using a text-based AI model built specifically to generate lyrics for this project, the vocals transmute stories from the tide of time into the beautiful and broken language of eidolons. All the intangibility of loss is given a new spiritual resonance, as the past speaks to the present, and we listen the best we can through much more than six solid feet of earth.
Off Shelf: In teaching the AI model, I read that you used 150 different texts that were focused on grief. Can you talk about what some of those were, generally?
Asa Horvitz: I didn’t have any interest in AI prior to GHOST. My father had died in 2017 and before that two of his brothers and my closest friend and my grandparents, all in a relatively short time. I was looking for some way not to mourn any specific person, but to be with the general feeling that the past isn’t past, that the dead aren’t gone. That history is a process we live inside.
I was writing songs and singing them with Carmen [Quill] and trying to capture this feeling. We did a residency at h0l0 in NYC where we did four shows in a month and nothing was really clicking.
Then one day during this time my friend Seraphina Goldfarb-Tarrant sent me a fragment of text. I didn’t know what it was or where it came from, but it totally grabbed me. I immediately wanted to sing it, get it into my body. The text was fragmented, strange, and seemed to be full of ghosts. I thought, “Damn, this is a lot like what I have been looking for.”
Seraphina told me that this text was generated by a Natural Language Processing AI system, which she had designed. This was years before Chat GPT and I had no idea what a NLP system was, and most people outside the AI field didn’t either. She explained the structure of the system – an archive of text, an inscrutable and somewhat opaque “black box” process in which the archive is processed in multidimensional space, followed by prompts which generate outputs based on the text the algorithm has been trained on. I immediately knew this was what I had been looking for – something both entirely material and somewhat mysterious. A way of dealing with stuff from the past and producing new stuff that felt like it was made up of all this “stuff” but which never quoted directly.
Seraphina and I, with the input of Alejandro Calcaño who was working at AI Now at that time, eventually trained an NLP specifically for GHOST. As you write, it had around 150 books. It’s a very small training data set which means that adding or removing almost anything will shift the voice strongly. It’s a narrow set, specific, nothing universal or conclusive. In choosing the texts I thought more about voice and rhythm than theme. Vibe over concept. I wanted writers with terse, anglo-saxon syntax and strong, simple image words. This seemed to be the best for singing. For example, Naipaul, Conrad, Woolf, the King James Bible, Heraclitus. In a way the structure of the system made it so that the text would seem to be made up fragments of other older stuff, from the past, and that felt like enough without addressing stuff about grief or history or memory directly. Plus who can escape those themes? It’s too universal in a way. I also added a lot of science fiction – Philip K. Dick, Octavia Butler, etc., and contemporary theory – Leo Bersani, Felwine Sarr, Adam Philips, etc., because I had an intuition that it would be exciting to combine very old and “classic” things with very new things. These layers are also present in the music.
You can find the full list online.
OS: How did you choose lyrical material from what was generated by the model? After it was generated, were there any additional steps needed for the lyrics to be cemented in form?
AH: We never edited within the lines. Most of the songs on the album are made from complete quotes from the AI’s responses. Sometimes we excerpted from larger chunks. Because of the small training data set the inputs or prompts didn’t make that much of a difference, the system really only has one voice. You can play with it here to check that out online too.
OS: Using the AI in this manner seems like a form of collage. I was wondering if that is a process you typically exploit, and if there is something specific you find in this method of creation that you might not have connected with otherwise.
AH: My interest in AI was in using it as a way to dig into this large archive of texts, to play with stuff from the past. This came from a need to get into the feeling that the past isn’t past, the dead aren’t gone. But I wanted to do this in an extremely material way. My intellectual background is Marxist, materialist, and at the same time religious. So writers like Benjamin really appeal to me.
The image of the AI system as a whole – archive, process, output – is one I find very powerful. The text which came out of it made Carmen and I sing differently than we might have had we written our own lyrics, and with Ariadne [Randall]’s processing and Bryan’s compositional input, and later Wayne [Horvitz]’s contributions, a musical world began to emerge that certainly would not have otherwise. I have not worked this way in other projects though I generally like lyrics which leave the listener a lot of room to associate.
OS: I think AI so far has managed to do something that is only really revealed with other forms of media in hindsight, after that media evolves. As the technology changes and we are given the best possible quality—the new standard, looking back at previous iterations, the artifacts of that media are something we begin to appreciate. Until given the gift of hindsight, it’s like those imperfections never existed. The difference with AI is that it will never be a finished product, and so as it grows we are able to appreciate these demarcations in progress, revel in the short-comings that make it feel faulty, human, and certainly more useful artistically. So, what were some of the things that interested you about integrating a new form of technology into your creative process?
AH: As you intuited, this is a kind of “vintage” AI project. The technology we used is broken by today’s standards and was intentionally broken by Seraphina to produce fragmented text. The idea was that you could feel this feeling of the presence of the fragmented past in the text itself. I don’t care about AI, in fact I am mostly against it from a political perspective. Though to paraphrase Ariadne’s paraphrase of some AI theorist – whose name I can’t remember – it appears that so far AI mostly re-arranges the given, rather than producing the new. So I’m not worried that it will replace art or artists and I’m also not concerned about the singularity or other so-called rationalist bullshit.
I worry much more about data collection and automated decision making, policing and weaponry. I also frankly don’t care that much about the concepts behind GHOST. As in the story I started with, I only got interested in AI because it produced a text that gave me a feeling as a singer that I wasn’t able to get anywhere else. That was something I was looking for on an emotional, intuitive level. It only matters to me if it actually does something on that level – to us, the performers, and ultimately to the listeners, to the audience.
OS: You have a musical family. Your father was a musician, and his brother Wayne performs on this album with you. Do you feel a common language between your kin? Do you think about how your father would receive this album where his life is very much at the heart of it?
AH: My father, Bill Horvitz was a very generous and warm person who opened my ears to a lot of music. For my 14th birthday he bought me a record player and would give me records from his collection – for example, Miles Davis, Varese, Indian classical music, Harry Partch, Ornette Coleman – to listen to every week, then come by my room and quiz me on them. Where are the chord changes, what instrument is that, let’s imagine how this was recorded, etc. Perhaps even more importantly, he would sing with me with a tamboura drone, trying to get me to sing really in tune and most of all to listen. He always told me that music was first and foremost about listening. The basis of music being listening rather than producing sound or expressing something is something that I really care about and believe in, and that I feel I have in common with the rest of the GHOST band. I don’t mean listening as an academic concept or as an in-vogue term in the art world. I’m talking about a very concrete experience, an action which you do with your body and can do with other people. I feel this when we play live. There’s a sensitivity that happens there that can be really pleasurable.
We had a release show at Public Records in Brooklyn last week and Robin Holcomb, my aunt and Wayne’s wife, opened up with a solo set. After she finished playing I turned to Carmen and said, “no pressure.” We both laughed, but there’s definitely a high standard in my family that I hope to live up to. Robin is one of my favorite songwriters. I feel very lucky to have grown up around such musicians, it keeps me on my toes and wanting to practice and keep learning. I can only hope that some of the GHOST material has a tiny bit of the insanely open-ended yet crushingly specific emotional resonance of her songs. There’s something in Robin’s work that is very “weird American music.” Starting with 20th century songwriters and popular music, I also hear Ives, Debussy, Messaien, traditional music like Sacred Harp, the AACM, and of course the post-Cage experimental composer lineage. I think Wayne, and also Carmen and Ariadne, share these departure points. I also hear these influences in my dad’s solo electric guitar playing from the 80s and his late big band music written in memory of his youngest brother Philip. It’s something experimental yet sonically and emotionally immediate. This is music that I love.
OS: Connecting this familial aspect to gathering a history of texts for the model to work with, can you talk a little bit about the theme of inherited knowledge that permeates this work?
AH: I just really hope that the experience of listeners is open-ended and that they get to travel through their own landscapes of association and memory, whether it is related to grief or loss or not.
OS: Did you begin working on this project ahead of your father’s passing? Did the presence of his approaching death inform your relationship, and how, if at all, did it change your relationship with the art you create?
AH: My dad died relatively suddenly and I didn’t have any artistic plans during his illness. I had made a grief project before, called VALES, for my friend Dominic Ziegler who died in 2012. That was really for Dominic, for a specific person and filled with personal memories. GHOST is kind of the opposite, inspired by personal loss but really trying to move beyond the mode of “my story” which is just all too pervasive in our social media moment. I write more about that in the essay on the website which is a version of the intro text to the theatre performance. I should also say that my father had a strong experience of softening after his youngest brother, Philip, died in 2005, so I grew up with a sense that grief might open or transform you.
OS: Whether involved in this project or not, was there any art that you found or were drawn to while grieving?
AH: When I’m grieving I really like to be in nature or to be with friends. I like to cook, swim, talk, sleep outside. I prefer to go to shows than to listen to records. I saw Mount Erie on the solo ‘Crow’ tour a few weeks after my Dad died, which was incredible and inspiring though of course extremely different from GHOST. Lately when I’m feeling sad I listen to a lot of music that is nostalgic for me. This fall I got really into Korn, Incubus, and Creed [laughs] but actually awesome. Perhaps unrelatedly, I’ve been in a Talk Talk phase for a few years.
OS: The album covers a lot of ground musically, from solo piano, to ensemble for voice, to landscapes of synthesizers and strings, but there is a melding of the synthetic with the organic, a juxtaposition that serves the music well. How did the songs take this form, and with the many years that passed in creating this material, what was the performance/recording process like? How did you know when it reached completion?
AH: We started with improvisations. I would bring a text from the AI system, Carmen and I would sing, and Ariadne would process us. Song forms usually emerged suddenly in an improvisation. We would listen back and choose something that sounded promising and then work on it together until it came into form. Sometimes things were just born in a single take, but definitely not always. The material still changes some when we play it live so I’m not sure it has reached completion. The recordings on the record are a mix of stuff, rehearsal recordings from 2019-2020, and then studio recordings from late 2023. In 2023 we made and toured a theatrical performance – also called GHOST – which gave us a chance to really hone the music through rehearsing and performing a lot.
OS: It seems you started the writing process working closely with Carmen Quill. How did you get Wayne and Ariadne involved? On a project that is obviously very personal, what was the process like involving these other artists and were there unexpected facets they brought out with their contributions?
AH: We started in 2019 in my apartment in Ridgewood. Carmen and I had been singing other songs already, mine and hers, but something wasn’t clicking yet. I started bringing in the AI text and at the same time invited Ariadne to come process us live. It was this combination – AI text, Carmen & Asa singing and playing, and Ariadne doing her processing and mixing – and later playing my weird Moog-imitation modular – that gave birth to the musical world of GHOST.
Songs began to emerge immediately – two of the songs on the record were improvised almost entirely in the first session we had. One of the tracks on the record itself is actually a Zoom recording from that session. Carmen’s father was dying during this time. “Hail Necessity” with “and my father is dead and I am the daughter of mighty friends” came out of a session right around then; Carmen improvised something very close to the final version of that song in one take. Sometimes the AI system would give us something insanely serendipitous, but that kind of unexpected meaning emerging is a feature of most randomized systems.
Bryan West joined a few months later and helped things begin to get focused – he is an incredible editor and just has the best ears – and a musical vocabulary began to emerge. A lot of the tracks on the record were written in a 2020 session in Amsterdam with the four of us. Bryan doesn’t perform with us anymore but is on a number of tracks on the record. Carmen and I were steering the ship, but the others made large and important musical contributions to all of the material.
Wayne didn’t join until 2022; Bryan had moved upstate to get a PhD and have a kid, and we knew it would be impossible to replace his very developed and specific skill set. We needed another strong individual musician/composer with a developed voice. I don’t remember how Wayne came up but once he did it seemed like an obvious choice. Wayne was extremely generous in accepting my invitation and super professional during the process. He worked under his idiot nephew, and tolerated some practical situations that no 70 year old should ever have to put up with. He brought his incredibly rich harmonic and textural language on piano, keyboards, and electronics, his inimitable touch, his listening, and his strong presence as a performer. It’s a huge privilege to have someone with 50+ years of experience in the team, you learn a lot. And as a friend of mine said when we performed in Amsterdam, just the fact that my father’s brother is there onstage brings a strong, surprising layer of emotional resonance to the performance.
OS: The title of the album is GHOST. What is a ghost, and in your understanding, what is the nature of their existence?
AH: I’ll let the AI system speak to this one – “The dead and animals will remember the smells of cabbage and water. I have no idea what is happening, and what I have to say is that I cannot bear to be a man. This is the only place where the dream is the sound of the dead man’s initials.”
OS: What do we gain in grief that is otherwise withheld by life?
AH: Heart? Dropping down, really being here. Our culture is pretty head focused. Grief stops you, re-arranges things.
OS: So, what is next for you and for the group? Is there anything else you’d like to mention?
AH: We really want to play more GHOST shows in the next year, but we have to see how things land after reviews, etc., have come out. Carmen has a European tour this month, opening for Melissa Weikart. Ariadne and Wayne both have lots of stuff going on in many contexts, check out their websites. After a rough year health-wise – I had to go home to have surgery – I am back on the road and working on a new larger performance piece with the composer, Anthony Vine that will premiere in 2026. I’m also playing smaller music shows along the way. The VR Opera I composed, SONGS FOR A PASSERBY, which I’m proud to say won a Lion in Venice in 2023, is still touring – mostly film festivals and museums.