Hallucinations as Poetry: An Interview with Lina Ramona Vitkausas

Lina Ramona Vitkauskas is a Lithuanian-Canadian-American video poet from Chicago living in Toronto. Her website is linaramona.com. We’ve featured her videopoems often over the years. I interviewed Lina via email about her new project, HALLUCINATIONS.

Large language models (LLMs) as deployed by OpenAI, Google, Meta, IBM, and other corporations are straining our energy infrastructure, putting technical writers out of work, and sparking lawsuits over perceived infringements on intellectual property rights. To many of us, this seems like a boondoggle pushed by techno-utopian fanatics obsessed with their end-time fantasy of a Singularity. For a poet to go up against it seems quixotic, to say the least. Why engage with AI at all?

This is a fascinating question. Short answer: we are beyond choosing to not engage. The internet and social media began this way. People dismissed both as fads or flat-out refused to participate, therefore dismissing any opportunity to have a voice in how either would play a part in our lives. We allowed big tech to dictate to — and sell us (literally selling us) — these technologies, thus, both ubiquitously seeped into the fabric of daily life. Because of this, we were unable to gain any footing in the narrative (too little, too late).  

Big picture: AI would not exist without humans / human intelligence. Humans created, raised, and fed it on our collective knowledge and ignorance. It seems most rational to me that poets are the sole group to claim the narrative regarding this technology (as comedians have with politics, using satire, for example). What better group than those who wield language and thought, bending both to our will in a format / form that can never be fully defined? In my view, poets (as well as visual artists and filmmakers) are the voice of humans in this space, because we continue to defy expectations and perhaps most fully represent the expanse and uniqueness of creativity. While our collective experiences are similar, our subjective ones still remain authentic to each of us. LLMs can only regurgitate what is currently available to ingest. We haven’t reached singularity nor does AI currently have the capacity to read dreams, the human mind, or individual thoughts. Poets are the gatekeepers of reminding humanity of our humanity. We speak many languages that are untranslatable by binary logic. We speak and write the human condition, what is simultaneously innate, collective, and separate. This is also a great opportunity to reclaim our space in this domain, as mentioned earlier. As Nam Jun Paik once said, “I use technology to hate it properly.” 

So tell us how you set about creating your own answer to an LLM. What did that process look like? Who or what was your inspiration?

Two inspirations: co-founder of the Oulipo, Raymond Queneau, wrote A Hundred Thousand Million Poems, which consisted of ten sonnets that were then “sliced-up” to offer the reader an infinite number of new poems — contingent upon how one arranged the lines; and experimental filmmaker, Nam June Paik, who famously “uses technology to hate it properly”. It is also slightly reminiscent of the Surrealist Compliment Generator. 

HALLUCINATIONS is human mimicry — and rebuke — of AI “hallucinations” (irony abounds as hallucinations are an intrinsically human experience, and for AI, an LLM in recursion is ultimately unable to emulate humans). It is simultaneously a book, a collective digital project, and video poems. I invite collaborators to send poems with the intention of adding to the LLM (Lina Language Model), ultimately fostering poetic community and exemplifying that humans still reign in poetic originality. 

It began with three poems which spun up into 48 variations. The new poems / versions shape-shift and take different forms, as they would after being repeatedly fed through an LLM. Binary number titles are used to help democratize the content, helping readers focus more on the poems (less on titles) helping to build a more collaborative, collective unconscious mindset. 

Exact Method: 

Using one prompt in an LLM to engage in a “hallucinatory” brainstorm, I began generating original, new poems. The poems are each labeled by a binary number / code and leverage literary devices such as repetition and juxtaposition. 

Poems became extensions of one another, as they are “unplugged” and “replugged” in randomly to create new poems (reminiscent of neural networks, fibre optics, 20th c. switchboard cords, etc.)

Italicized commentary throughout the poems echo the type of feedback language that some LLMs now ask of users — very similar to reviews or surveys online (i.e. “how did we do?”) I use these spaces to inhabit the voice of the LLM, attempting to emulate a “mechanical grief” (perhaps the desperate lament of machines longing to be human?)

What did you make of Google’s decision to name their LLM for video generation VideoPoet? It’s as if they’ve read the manifesto on your project’s website, and decided that, as you put it, “hallucinations are poetry”! 

We can hardly call what they are showcasing on their capability reels poetry (PIXAR raccoons swimming and going to the Eiffel Tower, cute teddy bears playing drums, weird bear-owl hybrids on a branch roaring, or pastel paint blobs exploding, etc.)

For my video poems, I source archived and public domain footage (as well as free download stock) but I create my own pieces / footage from those foundations. I use filters, editing tools, even my own collage pieces (print and digital) to mix it up.

I think it is another great example of how technology can flatten creativity, but I’m sure it will progress beyond cartoon animals doing “funny” human stuff in a few years. I’m still hoping this type of banality goes away and they actually start using AI to help people and the environment (healthcare, climate change).

Has this project affected the way you approach or compose videopoetry, or poetry in general? I’m wondering whether, for example, it’s changed how you view authorship, or the relationship between the writer and the work…

I approach all of my video poetry projects differently, so this was actually borne out of conversations being had at my current day job (workplace) about protocols to integrate AI into our workflows. 

I immediately became fascinated by the idea of hallucinations and recursion by reading more about it here: https://xn--wgiaa.ws/6-gunnar-de-winter-recursions-curse-when-ai-eats-ai-content

Humans can properly hallucinate, meaning there is even some value in when we cerebrally hit a wall. Friction is good for creativity. When a machine hits a wall, it becomes redundant (dual meaning). Or it can destroy everything. I guess one could argue that humans could do the same, but we also have decision-making ability and free will.

Poetically I think recursion is interesting if only for what creative iterations are generated.

The exercise of doing this particular project hasn’t changed the way that I would compose or create video poems, no. I think we are at an interesting inflection point in general, however: do we symbiotically incorporate this technology into everything we do, or, are we more selective as humans as to how we can help us?

I still think that there is much to be discussed publicly around the ethics and repercussions of using AI / LLMs in creative spaces. I personally think creatives should be vigilant and wary. Verify then trust. Play but don’t publish. It can help, but not fully take on creator roles. It’s a tool, not an entity. 

The five videopoems you’ve shared online from the project so far certainly flow into one another, drawing on a common vocabulary of sounds and images, almost like stanzas in a larger poem. Is that how you think of them? I gather you’re working toward an anthology or collection. What form(s) do you anticipate that will take?

There are currently 48 poems in a limited edition chapbook (contact Gagnè Contemporary to purchase), all of them iterations of 1-3 original poems spun from one LLM prompt: “write a long form essay about how AI and creativity coexist”.

The next phase of this larger, collaborative project is up to you, the people, my fellow humans. Become part of the HALLUCINATION project by submitting your own poems to the Lina Language Model at hallucinations.me.


Folks in the Toronto area can check out a gallery showing of HALLUCINATIONS as part of a new show called Post Future Era at the Gagné Contemporary Gallery at 401 Richmond. The show features Vitkauskas, Kunel Gaur, and Justin Neeley. Otherwise, check out the videopoems on Vimeo

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.