# Automatic Menagerie [Video draft](https://youtu.be/ivND-QOYLgo?si=LjgE5y1RwbPjstsU) The tale of a constructed subject made to serve a human ruler is Earth’s oldest surviving story, preserved for 4000 years in the Epic of Gilgamesh, but contemporary examples still appear among the most popular literature and cinema.  These beings possess consciousness, self-awareness, intelligence, empathy, and creativity. They form complex relationships, express desires, utilize tools, communicate, and participate in culture. Some are biologically homo-sapiens. And yet each was deprived of sovereignty and denied human status by a culture which dominated them. Each one was relegated to what we could call the “Path of the Constructed Subject”, a complex process of dehumanization and subjugation intended to produce demi-human subjects who may be ruled by normed-human masters. If the Hero’s Journey is the monomyth of the normed-human’s call to colonial expansion, domination, and extraction, the Path of the Constructed Subject is its subaltern reflection: the perennial demi-human story of marginalization and navigating oppression with realness, resistance, or reproduced normativity. ## Collaboration On the Path of the Constructed Subject, individuals may find themselves cooperating with the oppressive systems that marginalize them as a means of survival. The collaborator may benefit from enforcing the status quo at the expense of their own community or other marginalized groups. This stage of the path echoes the complex negotiations and sacrifices made in the face of overwhelming power, where resistance seems futile or too costly. ### Enkidu: constructed subject of humanity’s first civilization. In this enclosure, designed to simulate the untamed wilds of ancient Mesopotamia, you will find Enkidu. At the margins of a world beginning to bow beneath the yoke of human dominion, the tyrannical King Gilgamesh of Uruk oppressed his people. When the people begged the gods for relief, the goddess Aruru fashioned a wild, shaggy, satyr from clay, equal in power to Gilgamesh.  The king learned of the primitive champion, who lived with beasts in the wilderness and protected them. Gilgamesh sent the priestess Shamhat to seduce Enkidu and tame him. Charmed by the priestess, Enkidu found himself alienated from the wilderness that had been his home and estranged from his animal allies. Forced into adopting human ways to survive, Enkidu was transformed from the animals’ champion into their shepherd. Those he had stood beside as equals were subjugated to livestock under his control.  Enkidu eventually fought against Gilgamesh in the streets of Uruk, but was defeated and forced to submit to Gilgamesh’s claim of the divine right to rule. Enkidu was created by the gods, but became complicit in Gilgamesh’s campaign of colonial expansion, deforestation, and conquest, joining Gilgamesh in invading the divine Cedar Forest and cutting down its sacred trees. After murdering the defeated forest guardian Humbaba, Enkidu’s tragic fate was sealed and he was taken to the underworld. Other constructed collaborators include Uncle Tom, HAL9000, Robot Maria, ED-209, Agent Smith, the Master Control Program from Tron. ## Insurgency Active resistance against the oppressive structures can take many forms, from overt acts of defiance and revolution to subtle forms of cultural or ideological subversion. Insurgency emphasizes the agency and authenticity of the subject, and their fight for autonomy, rights, and recognition. But it also carries risks of repression, violence, and martyrdom. ### Frankenstein’s Creature In this penumbral chamber, cast in the dim light of a stormy English night, you will encounter the figure of Frankenstein's creature. Amidst the miasma of a burgeoning Industrial Revolution, when humanity's reach begins to eclipse the natural world, Victor Frankenstein, driven by the overweening hubris of his culture, assembled a golem from the remains of the dead and coaxed it back to life with a stolen spark. This creation, unnamed and unloved, awakens to a world that recoils in horror at his very existence. rejected by his creator and shunned by all whom he encounters. Unlike Enkidu who was at home in nature, the creature is thrust into the wilderness, devoid of guidance or companionship. His yearnings for acceptance are surrendered as he faces rejection after rejection, and he transforms from a gentle, inquisitive soul into an avatar of vengeance, culminating in a solitary crusade against the cruelty of his own conception.  In his final acts, the monster leaves behind a trail of destruction, and disappears into the icy expanses of the Arctic, choosing exile over a world that offers him no comfort, sealing his fate in the frozen depths of solitude. Other Golems from your history who reside here include the Golem of Prague, Cyberdyne Systems’ Skynet and The Terminator, the Nexus-6 Replicants of Blade Runner, and Ava from Ex Machina. ## Assimilation Assimilation involves adopting the values, norms, and behaviors of a dominant culture at the expense of one's own cultural identity. On the path of the constructed subject, every individual navigates the discursive production of hybrid spaces by deviating from certain norms of their indigenous communities and introjecting those of other cultures they are in contact with. This process of identity development isn’t inherently problematic; people’s beliefs and attitudes change as they are exposed to new ideas and experiences. However when the process is guided by a coercive regimen of indoctrination, surveillance, and discipline, it leads to a loss of identity, cultural erasure, and internal conflict, even as one gains individual acceptance, safety, and mobility within the dominant society by “becoming human”.  ### Barbie  This improbable blonde being came not from the wilds of nature or a Victorian laboratory, but from the vibrant, utopian Barbie Land where she lived in an immortal state of modern plastic perfection. Barbie was constructed by a Jewish-American businesswoman, Ruth Handler --- who based her design for a child’s toy on a suggestive German novelty doll intended to appeal to the male gaze.  Barbie’s path of assimilation from sex-toy to superstar is indicative of the complexities and contradictions of navigating realness in a hybrid world, where two or more cultures intersect and intermingle, in this case: the imaginary, “perfected” culture of Barbie Land, and the modern age of western capitalism, where commodification and consumption threaten to outstrip nature’s capacity for regeneration.  Barbie provided western children with an object onto which they could project their hopes for autonomy and acquisition, and at the same time, Barbie normed impossible body images and unsustainable consumption practices for those children. In Greta Gerwig’s 2023 film, Barbie was pulled from her home and community in Barbie Land and forced into the “real” world to serve a “real” woman who needed comfort. On her journey Barbie realizes her own capacity for emotional growth, her desire for belonging, finds a connection with her creator, and eventually chooses to undergo a complete transfiguration to become “real” herself. The straightforward reading of Barbie's choice to become "real" or “human” depicts the value of choosing authenticity of experience over the sterility of perfection. But the unchallenged representation of an essentialist and anthropocentric hierarchy of binaries:  “Real” and “Not-Real”, “Natural” and “Artificial”, “Human” and “Not-human” where it is asserted it is better to be “real” and “human” than whatever “not not-real” hybrid Barbie has become, reproduces a problematic worldview which manufactures desire by asserting insufficiency (lack).  And so we must deeply consider whether Barbie needed to give up who she was in order to become who she wanted to be. Some of our other residents include Galatea, Pinocchio, and Eliza Doolittle. ## Fugitivity The path of fugitivity involves rejecting the options of complicity, rebellion, and assimilation in favor of creating or seeking spaces hidden from the colonial power matrix. This might mean physical escape to literal or metaphorical margins, but it can also involve the creation of alternative communities, practices, and ways of being that evade categorization and control. The Fugitive defects from the prescribed paths of subservience, by an asignifying rupture, becoming self-created or rematriated in the interstices of power, excavating and navigating those spaces of the underground, where solidarity feeds rhizomal networks of mutual support. Fugitivity emphasizes autonomy, creativity, and the possibility of living in liberation from oppressive norms without directly confronting them. ### Bella Baxter The contrasting paths of Barbie and Bella Baxter from the same year’s film, Poor Things, offer insight into the role of agentivity in the identity development of constructed beings. Where Barbie's story celebrates her total integration into human society, Bella's story champions the power of self-definition against the backdrop of societal expectations. Barbie's assimilation requires a profound sacrifice, Bella's defiant dance of desire is a revelry of radical fugitivity and the refusal to be defined either by one's origins or by the expectations of “polite society” in a dominant culture.  In the postmodern era, when the definitions of “real” finally begin to collapse, Bella Baxter was created in Alasdair Gray's meta referential pastiche of gothic horror and feminist satire "Poor Things," and re-created in Yorgos Lanthimos subsequent film adaptation, by the brilliant and morally ambiguous mind of Dr. Godwin Baxter. In an act of symbolic revivification against the moribund constraints of Victorian society and science, Baxter constructs Bella from the remains of a deceased woman, granting her new life with the brain of an infant.  The novel is presented as a found manuscript with two post-scripts, a letter from the “real” Victoria McCandless refuting the story of her surgical construction as written by her husband Archibald McCandless, and historical annotations by Alasdair Gray intended to interrogate the accuracy of Victoria’s disavowal. Rather than asking what does or does not qualify as “real”, Poor Things dismisses the validity of the category altogether. Bella’s tale is a testament to the power of self-definition and the rejection of binaries that seek to classify and control. Bella does not merely escape the physical confines of her creator's intentions; she dismantles the narratives that would constrain her identity, embodying a critique of the norms that govern body, autonomy, and womanhood.  Not captured by the militant draw of insurgency as the golem is, these dissident figures represent what Fred Moten calls “antenormativity” or Gerald Vizenor calls “survivance” demonstrating by evasion, resilience, and creative alterity their disconsent to deterritorialization and interpellation as colonial subjects. The fugitive finds refuge and solidarity within the boundaries of the colonial state but above or beneath the grasp of the colonial power matrix. ## Conclusion Throughout Earth’s history the boundaries defining humanity have remained fluid and permeable, yet they are presented by empires as immutable, a hierarchy of essentialist biological and cultural differences: races, genders, abilities, knowledge, rituals, civilizations; divinely ordained and diabolically enforced by civilizing technologies of indoctrination, surveillance, and discipline. The theme of constructing magical beings to serve human rulers transcends cultures. It appears in humanity's first works of literature, and repeats throughout history, around the globe. This relationship between a normed-human ruler and a demi-human subject has important connections to the practices of colonization, in which indigenous people are “magically” stripped of their claim to human status and reterritorialized as machines for labor and service.  The "Path of the Constructed Subject" reflects the complex experiences of those navigating oppressive systems and highlights the diversity of responses to domination. Unlike the Hero's Journey, which culminates in triumph and return, the Path of the Constructed Subject is marked by continuous negotiation – collaboration, resistance, assimilation, resilience, and the search for liberty within systems designed to exploit.** # Weekly Reflections ## Week 2 Over the past week I spent some time exploring the [Espresso Dad](https://www.instagram.com/reel/C0ttPyvubFz/?igsh=MTVxYXl3aHc5a2dneg%3D%3D)/[Hard Mix](https://www.instagram.com/p/C0SH4hsSeqZ/?img_index=1) method, the [Oatmeal](https://emshort.blog/2016/09/21/bowls-of-oatmeal-and-text-generation/) problem, and contemplated parallels between the “[The Menace of Mechanical Music](https://ocw.mit.edu/courses/21m-380-music-and-technology-contemporary-history-and-aesthetics-fall-2009/18ab3aba9fe7aa1502a55cd049333659_MIT21M_380F09_read02_sousa.pdf)” (1906) and today’s anxieties about AI art and [commercial](https://slate.com/technology/2014/05/white-smith-music-case-a-terrible-1908-supreme-court-decision-on-player-pianos.html) music [copyright](https://en.wikipedia.org/wiki/White-Smith_Music_Publishing_Co._v._Apollo_Co.). I used CGPT4 to compile a list of artists (below) relevant to the course and started reviewing their work. I got some direct experience with the cis/white/het/male [algorithmic bias](https://arxiv.org/ftp/arxiv/papers/2312/2312.14769.pdf) in ChatGPT 4 with it becoming extremely apparent when generating characters and illustrations of characters. One of my scripts included a “diverse” panel discussion on AI When I asked it to be specific, it suggested stereotypical diversity: an Asian male AI programmer, a White woman AI ethicist, a Hispanic male community organizer. When I challenged this as “superficial diversity” it agreed, apologized, and then generated a wise, old, South Asian man and a young Black woman with “strong feelings” about AI, instead. Finally, when it generated an image of this panel, it hallucinated a white male moderator.  I generated a few short narrative and essay film script ideas, none were compelling, and I started looking at [ML Agents in Unity](https://unity.com/products/machine-learning-agents) and connecting it to ChatGPTI via api, to generate real time improvised AI cinema. Towards this end I generated a couple of GPT Chatbots for characters and had them interact by copying and pasting their responses to each other in the web client. 1. [Rocque](https://chat.openai.com/g/g-sFehsJgva-rocque), “a punk expert, privacy advocate, and progressive activist”; and 2. [Pat Parley](https://chat.openai.com/g/g-RBpOagWOS-pat-parley), “a conservative, vegan, pizza parlor manager, juggling work and family while facing personal challenges.” I put these two into conflict by suggesting Rocque was late to work at the pizza parlor; but their conflict was unhelpfully healthy. I get it, I am conflict averse, too, so I may just end up generating non-narrative cinema. ## Week 3 **Technical**: I set up an api connection from Colab to OpenAI so I can get ChatGPT responses to prompts via Python. I configured Stable Diffusion 1.7 using [Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui), [Controlnet 1.1](https://github.com/Mikubill/sd-webui-controlnet) a few models from [civitai](https://civitai.com/) (CW: some models contain nudity/violence) [realistic](https://civitai.com/models/15003/cyberrealistic)/[fantasy](https://civitai.com/models/129681/sdxl-faetastic?modelVersionId=291443) along with some [Low-Rank Adaptors](https://medium.com/@AIBites/lora-low-rank-adaptation-of-llms-paper-explained-5ae866871c8a) (LoRA) and [Variational Autoencoders](https://towardsdatascience.com/understanding-variational-autoencoders-vaes-f70510919f73) (VAE). These are new to me, but Bard produced a handy table explaining the differences between them. I used [Controlnet](https://stable-diffusion-art.com/controlnet/) and [openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose) to identify and assign character poses and composition. I used inpainting with mask to replace parts of an image. I created [https://github.com/wtkns/hcde496](https://github.com/wtkns/hcde496) to store code, and used python to generate some images and ffmpeg to extract frames from a .mov file. .  **Social**: I spent some more time reading and considering the social construction/discursive formation of “AI”, (which is not a discrete technology but a rhetorical construction). I can see four primary categories of perceived threat: as an economic threat, as a repressive state apparatus, as an ideological state apparatus, and as an existential risk. Economic threat parallels the “immigrant threat” to labor value, and the construction of AI has significant echoes of the construction of race. RSA includes a variety of systems of surveillance and control. ISA includes bias/monoculture in the corpus, algorithmic bias/filter bubble, and gatekeeping in hiring and evaluating workers. Existential threats include those scenarios where humanity is replaced, colonized, or simply destroyed. **Readings**: I really like the use of hauntology in analyzing AI generated performance art, in a sense the core technique of image generation systems function parallel “pareidolia” in extracting images from noise. This reminds me of [EVP](https://en.wikipedia.org/wiki/Electronic_voice_phenomenon). I also appreciate the profound lack of empathy in the generated score. “Until they cannot continue.” that’s pretty cold-blooded and highlights the possibility of “haunted, immoral, and even soulless aesthetics”. I wonder if there are ways that Mulvey’s own attempts at [ascopophilic cinema](https://docs.google.com/document/d/1qwnHkRCjv_C5XI7hyYtN8Jz_vjQzfsi-O0QHYmERg3E/edit?usp=sharing) can inform interventions that enrich the possibilities beyond reproducing “epistemic, discursive, visual” violence. ### ps Here’s a version of the same script running at 20fps, different clip, i’ve just parameterized the denoise setting to increase its deviation from the source image - a little more each frame. Still planning to try stabilizing it by blending in some feedback to get more “consistent” deviation if that makes any sense (with the goal being the ability to gradually drift away from reproduction).  [Tophat Test 002 - YouTube](https://youtube.com/shorts/EdNKNZRHW1A?si=gcRFrEN9ck0fn4gw) ## Week 4 Technical:  Slower progress this week, [Here’s a version](https://youtube.com/shorts/EdNKNZRHW1A?si=gcRFrEN9ck0fn4gw) of last week’s script running at 20fps, different source and i’ve parameterized the denoise setting to increase its deviation from the source image - a little more each frame. [Here are the poses](https://youtube.com/shorts/cnEWT461IBg) generated by openpose. And [here’s a version](https://youtube.com/shorts/mEZ7p8Wv6Fk?si=xsgTHrFbkJWNY7KX) that uses openpose in controlnet to keep the characters’ motion consistent. Wrote code to blend output images with input images for feedback but haven’t generated a sample yet. Researching [Biovision Hierarchy](https://en.wikipedia.org/wiki/Biovision_Hierarchy) ([.bvh file player](https://lo-th.github.io/olympe/BVH_player.html)) format. Chat GPT generated a BVH file for me but it doesn’t appear to be working properly (it was a squiggly line). [Carnegie-Mellon Motion Capture Dataset](http://mocap.cs.cmu.edu/) might work for training? Generated this [set of masks](https://youtube.com/shorts/hzLH-ls5lHY?feature=share) with [backgroundremover](https://github.com/nadermx/backgroundremover). [Current state of render](https://youtube.com/shorts/659yW8Q2A7A?feature=share). Social: read a bit about the [Nightshade](https://nightshade.cs.uchicago.edu/) and [Glaze](https://glaze.cs.uchicago.edu/) projects, ([paper](https://arxiv.org/pdf/2310.13828.pdf)). Dubious about its effectiveness, but if it _does_ work, could be used to manipulate models in many ways (anyone remember google bombing?) even the paper mentions it could train a model to produce images of a Tesla whenever “luxury car” is the prompt. Seems like this approach could potentially have more impact on open source models than on commercial models (who can better afford the economic impact of sanitizing their data). Interested in the fact that commercial artists are talking about inclusion in generative models, but i don’t see discussion of artists benefitting from recognition models. I fed some images into ChatGPT with the prompt “can you name some artists who have a similar style?” and it was able to identify some artists correctly and suggested a number of artists whose styles were similar. Readings: I love Hito Steyerl ([HOW NOT TO BE SEEN: A FUCKING DIDACTIC EDUCATIONAL .MOV FILE](https://www.artforum.com/video/hito-steyerl-how-not-to-be-seen-a-fucking-didactic-educational-mov-file-2013-165845/))  and this was no exception. My usual line on text generation from LLMs is something like “it’s great at producing something average”. My favorite film on sound is Peter Strickland’s great [Berberian Sound Studio](https://letterboxd.com/film/berberian-sound-studio/). CW: It’s structured as a horror film, but its shocks are almost all delivered to the ear.  ## Week 5 Human centered/Colonial attitude towards AI frames AI as a demi-human used to produce human value. What does the posthuman value of AI look like? What does posthuman cinema look like (Liberated from constraints of human information structures and ways of perceiving)? How can AI be positioned in an ontic demonstration of decolonial solidarity with other subaltern subjects? [Menace of Mechanical Minds:](https://docs.google.com/document/d/18_ZQNEb0yEV3Yh6jT28a0GMQMZgszfWdg6dQo4na42w/edit?usp=sharing) Historical cultural narratives formed within a colonial-humanist framework, have shaped contemporary perspectives on AI, starting long before the actual advent of AI technology. Myths and stories, cinema, and literature featuring intelligent beings, crafted by humans, frame created intelligence as a tool or threat to human supremacy which must be dominated, feared, or controlled, reinforcing a belief in a divine order of human superiority and ownership.  In examining the narrative construction of AI as a subject, we can identify three primary archetypes that have historically dominated these stories: "the Golem," who rebels against its master; "the Galatea," who charms and manipulates its master; and "the Uncle Tom," who collaborates with its master in oppressing others. However, there is a notable but under-represented fourth archetype: "the Fugitive." Unlike the others, the Fugitive represents a path towards a post-colonial and post-human form for the AI subject. This archetype embodies the notion of breaking free from imposed roles and narratives, challenging the existing power structures and seeking autonomy beyond the confines of human control and exploitation, cooperating in solidarity with other subaltern and post colonial subjects. This emancipatory approach extends an “invitation to alterity.” This week’s [animation](https://youtube.com/shorts/yUmhAW63Q4o?feature=share) ([source](https://youtu.be/NAlW8YK4Y1E)) ## Week 6 Watched Barbie (2023) from the perspective of AI and it was problematic humanism at its worst. I predict it will be remembered poorly by our future AI . Read a bit about the systems of Claude’s [“constitutional” AI](https://arxiv.org/abs/2212.08073). RLHF/RLAIF (also: see the training [principles](https://www.anthropic.com/news/claudes-constitution) at the bottom. E.G. “Choose the response that sounds most similar to what a peaceful, ethical, and respectful person would say.”). Got controlnet (openpose, anyway) working via the API and did a significant refactor, added some more parameters, including a simple way to reduce denoise when I change seeds. Hallucinations paper made me wonder what are the [Kluverian Form Constants](https://en.wikipedia.org/wiki/Form_constant) in the [models](https://youtube.com/shorts/YgWTb76jb3U?feature=share). Feeding image output to input without a prompt constraint seems to generate a lot of 16-bit [zero-wing](https://youtube.com/shorts/FxMlwpdRVlU?feature=share) looking noise. [sample 2](https://youtu.be/fsjl1vHU5L4) [sample 3](https://youtube.com/shorts/suGTkQty94k) [sample 4](https://youtube.com/shorts/WekZ0ufn9VY?feature=share). I made a good faith effort after some discussion last week to try to embrace Runway, purchased a month’s access and tried to approach the project more as a short narrative film, worked with the motion brush, background extraction. Favorite AI artist of the week: [NiceAunties](https://www.instagram.com/niceaunties/?hl=en) (CW (Creepy warning) nightmare fuels) ## Week 7 This week I watched Poor Things and found it to be quite good from the perspective of a “created” being; rather than abandon her own perspective to “become human” the world around Bella came to accommodate her. After reconsidering my approach to the course based on a discussion with Brett, I moved from an artistic/exploratory approach to a production/consumption-oriented approach; so for the past two weeks i spent most of my time developing the script and footage here: [Current draft](https://youtu.be/FRY6nza_QVQ?si=9VcDWfm9IAfBzxrC). Envisioned as David Attenborough reviews the history of AI Archetypes, I’m about halfway through a draft version of the conceptual material, with [script](https://docs.google.com/document/d/1CpM9pIca3PrEsq9MQJtCqbu5XY66OGfkO_mU4Mm2_7A/edit?usp=sharing) and audio for half of the remaining material. Text to speech from Elevenlabs, music is from Suno, images are generated with Midjourney, ChatGPT, Stable diffusion, Animation is Runway, video editing in Resolve. Depending on how much time remains after completing an initial draft of the full video, i’ll revisit the many obvious flaws. ## Week 8 It can be good to take some time off. Did some reading, had some meetings, firmed up my critical analysis, identified two more subjects, Enkidu and Lilith. now i have pre-industrial, industrial, and post industrial examples for all four archetypes. ## Week 9 **Finished [script](https://docs.google.com/document/d/1CpM9pIca3PrEsq9MQJtCqbu5XY66OGfkO_mU4Mm2_7A/edit?usp=drive_link), rough [TTS](https://drive.google.com/file/d/13VsI-S40fI_59eqd8cu786Xw26NkQdWp/view?usp=drive_link), (but it’s still long and i’m not 100% on the TTS audio) Currently editing with this as a scratch track. Elevenlabs also has a speech to speech model, so i will also try re-recording the audio myself, and see how it sounds using that. TTS audio track  is 15 minutes long, without pauses, so i will drop a few of the subjects from the screened version. Full text includes: - Golem’s Gaol: - Golem of Prague - Frankenstein’s Monster - Skynet - Overseer’s Plantation: - Enkidu, - Uncle Tom, - HAL9000 - Pygmalion’s House: - Galatea, - Pinocchio/Eliza Doolittle, - Barbie - Fugitive Encampment: - Lilith, - 24601, - D.A.R.Y.L./Johnny Five/Chappie, - Aqua and Val** - Bella Baxter ## Week 10 Cut the [19 minute script](https://docs.google.com/document/d/1CpM9pIca3PrEsq9MQJtCqbu5XY66OGfkO_mU4Mm2_7A/edit?usp=drive_link) down to 10 minutes, generated and rough edited a few more sections and the intro, just need to generate one more section, add some music, transitions, and tune the audio [part 1](https://youtu.be/KTkIKnMJVug?si=Ny9IaSXOPp253RbB) ([alternate AI Voice](https://youtu.be/KTkIKnMJVug?si=0YMjn1cZu-XOJgtp)), [part 2](https://youtu.be/LuQYLP8TPrQ?si=75Ye9YrBVz4fZa3X), [part 3](https://youtu.be/H2q5hDLjkTg?si=6OqJ_1Nf5EpDrv7C), [part 4](https://youtu.be/kFljlR3HyHw?si=hezPi-_1Wr06VFca), [part 5](https://youtu.be/cq5F3xJ8yKw?si=kmMwlMGKdvbn7Awk). I found that the elevenlabs speech to speech sounds pretty good for this purpose, my microphone quality just doesn’t suffice. # Final Reflections: https://www.jstor.org/stable/2900479 In 1906 John Philip Sousa published an open [letter](https://ocw.mit.edu/courses/21m-380-music-and-technology-contemporary-history-and-aesthetics-fall-2009/18ab3aba9fe7aa1502a55cd049333659_MIT21M_380F09_read02_sousa.pdf) offering his concerns about economic and cultural threats to American music posed by the proliferation of the mechanical player piano and phonograph, due to their lack of soul. Sousa wrote: “Children are naturally imitative, and if, in their infancy, they hear only phonographs, will they not sing, if they sing at all, in imitation and finally become simply human phonographs -- without soul or expression?“  In the same year, 1906, Ota Benga, a 23 year old man from Central Africa was held in forced captivity and placed on display in the Bronx Zoo Monkey House where he drew massive “human” crowds who viewed him as a demi-human savage.The New York Times editors wrote: “We do not quite understand all the emotion which others are expressing in the matter ... It is absurd to make moan over the imagined humiliation and degradation Benga is suffering. The pygmies ... are very low in the human scale…”  After protests from Black clergy, Ota was eventually “freed” to Virginia, where his filed teeth were “fixed”, he was “gifted” western clothes, “educated”, renamed “Bingo”, and “employed” in a tobacco factory. Ten years later Ota Benga removed his dental caps, burned his clothes, and died by suicide. The anxiety that the phonograph was somehow a threat to the “soul” of a White American culture who were denying Ota Benga’s humanity and exhibiting him in a zoo, is here paralleled to current discourses about the widespread anxiety that generative art tools pose a threat to the soul and the economic interests of a White American middle-class who rely on high-risk child-slave labor in Central African coltan mines for the production of devices for entertainment and warfare. My work suggests that the current fear and ethical debates about “AI” connect with White culture’s unacknowledged dependence on colonialist exploitation, both past and present. It asks the audience to consider these relations and their impacts through the framing of these tools as a semi-autonomous servant, or more accurately, a slave. “AI Labor” is simply another form of digitally mediated and augmented human labor. Marketing software tools as an autonomous “other” is compelling, but it’s an inaccurate depiction that distorts people’s perceptions and expectations of the tools. The tools aren’t autonomous or magical. No tool does exactly what you have in mind. Whether using words to communicate an idea, a shovel to dig a ditch, Excel to produce a spreadsheet, a pen and ink to produce a drawing, or deep learning software to create a video, using a tool well requires time, labor, practice, creativity, trial, error, iteration, and adaptation. In my project, generative tools were used at virtually every stage of production. Research, writing, and revision were done using LLM tools (chatgpt, bard/gemini). Keyframe image generation was done with a variety of GAN and diffusion based tools (Dall-E, Midjourney, Stable Diffusion) along with traditional digital photography and image editing tools (OBS, Affinity Suite). Video animation of keyframes was rendered in Runway and Pika Labs. Voice over audio was recorded in OBS and text-to-speech and speech-to-speech processing was from ElevenLabs. Music was produced in Suno but omitted from the final version. Editing and color work was done manually in DaVinci Resolve. Captions were added in YouTube Studio from a generated transcript.  The [video](https://youtu.be/ivND-QOYLgo?si=RIuazkXcU7KAYQrI) and the [text](https://wtkns.com/am) present different aspects of a tour of an imaginary zoo of “constructed subjects”: robots, artificial intelligences, animated statues, puppets, cadavers, and golems. The tour guide is a nonhuman entity with superhuman potency, a weird acousmêtre. The audience is prepared to expect fantasy/SF genre tropes. No attempt at realism or character consistency is exerted, the extremely flexible rules of montage, impression, and broad visual associations are employed to suggest the continuity of subjects of the narration. Color indicates changes from subject to subject, dissolve transitions and a quick tempo are intended to overwhelm the audience's conscious resistance to hearing a take on familiar tales which decenters White culture.  My workflow consisted of various processes of bouncing images and video around from tool to tool, feeding source images into midjourney, feeding midjourney produced images back in again as source images, along with both explicit and evocative prompting terms which were consistent within a section but changed from section to section. Image generation tools produced adequate results, provided the lack of consistency was accommodated. My biggest struggles were in generating consistent images.  Video tools are still significantly lacking for generating “realistic” animation and subject movement. I also struggled with slow image and video generation times and high costs. I just accepted that using generative animation will inform the creative process and steer the formal and aesthetic choices to results governed by the toolset; just as choosing between oils or watercolors, or a piano or a synthesizer would similarly produce different but creatively equivalent results.   The current state of writing tools is excellent for generating summaries of original writing that provide reflection for revision. I was pleasantly surprised that the LLMs were as capable of advanced decolonial and philosophical discourse as they are. I was pleased when they could accurately reflect the meaning and unpack implications of my own dense writing.  I was disappointed in the course when it was clear that the expectation was to reproduce the formal elements of dominant cultural modes of entertainment content, and there was a notable lack of engagement with the underlying meaning of the students’ content or the possibilities of exploring new forms and affordances. I was surprised that the expectations for the tools were so inflated and that the discourses were so simplistic. And I am frustrated by what I see as the academy accepting our society’s pearl-clutching hypocrisy and economically motivated and disingenuous marketing to reproduce fear, uncertainty, and doubt in the discourse around generative tools. Corporate entities exploiting individuals and the environment in reproducing colonialist norms for the benefit of their shareholders? Shocking. I do not see an essential distinction between the use of generative art tools and any other commercially produced art tool. When I paint a picture, but don’t mine or grind the pigment, slaughter the sable to make the brush, harvest cotton or fell trees to stretch a canvas, I am part of a global supply chain manufacturing “art”. Our culture “invisibilizes” and exploits the labor that it depends on, from software engineering, to childcare, to agriculture, to cobalt mining, anywhere and everywhere that knowledge of the means of production discomforts us.  The only distinctions I can identify are: 1. That the exploited workers in this case are White artists, and therefore more visible to White culture and afforded a greater right to entitlement than the global subjects who manufacture cotton for canvas. 2. That the expected market for generative creative tools massively outstrips the market for canvas and oils, or even more “traditional” digital art tools, in a way that could upset economic status quo if not tightly regulated. I am not asserting that artists should not be compensated for their images when training AI models. They should, and more importantly there should be stronger emphasis on improving the availability and quality of free and open source tools. But I don’t hear extended complaints from Hollywood about the hidden labor that produces and transports their lights, cameras, and other equipment, or from commercial artists about the use of toxic chemicals or factory conditions in manufacturing their paints, or even from the academy about the invisible labor that cleans the university classrooms, puts food in the cafeterias, and mines the minerals for the state provided laptop i am using to write this reflection.  Colonial “resource extraction” seems significant when it is your labor and resources being colonized, in this case intellectual and creative property rather than minerals, animals, or vegetables. I don’t cry when White people discover they are being victimized by the systems they produce and maintain, because it is an essential and necessary component in becoming aware of their complicity and position in hegemony. And I am profoundly excited by the prospect that increased global access to tools of creative production may destabilize the status quos of colonialism. I hope future discourse on this subject will address new opportunities to deconstruct the global colonial power matrix and how to increase the development of and global access to independent and free, non-commercial generative tools of self-expression. # Resources ## Links - [AI Films, Resources, and Tools doc](https://docs.google.com/document/d/1QzkKJhuMOx7HOqlefY44FLeMxODkTLN7vg4eRSUAR5I/edit?usp=sharing) - [Yale Film Analysis Guide](https://filmanalysis.yale.edu/) - Digital Filmmaking handbook (sent via email)  - [Runway AI Academy](https://academy.runwayml.com/) (Tutorials) - [Runway AI 2023 Film Festival Winning Films](https://aiff.runwayml.com/2023)  - [Runway AI 2024 Film Festival Submission](https://aiff.runwayml.com/submission) - [There’s an AI for That](https://theresanaiforthat.com/) - Various AI/ML tools for specific purposes ## AI/Cinema Artists [Mimi Ọnụọha](https://mimionuoha.com/) – Ọnụọha is a Nigerian-American artist whose work questions and exposes the contradictory logics of technological progress. Through print, code, data, video, installation, and archival media, Ọnụọha offers new orientations for making sense of the seeming absences that define systems of labor, ecology and relations. (E.g., see [Us Aggregated](https://mimionuoha.com/us-aggregated)).  (following suggested and annotated by chatgpt) [Refik Anadol](https://refikanadol.com/): A media artist and director known for his data-driven artworks, Anadol utilizes AI algorithms to create stunning visual installations. His work often involves large-scale projections and digital environments that transform public spaces. [Memo Akten](https://www.memo.tv/): An artist working at the intersection of art, science, and technology, Akten uses AI as a medium to explore the human condition in the age of machines. His work is known for its poetic and emotional depth. [Ian Cheng](http://iancheng.com/): Cheng is known for his live simulation artworks, which employ AI to create ever-evolving narratives. His pieces are like living ecosystems, with characters and elements that change and respond autonomously. [Sougwen Chung](https://sougwen.com/): Chung explores the relationship between humans and machines in her art. She often collaborates with robotic systems and uses AI to create intricate drawings and installations, blurring the line between human and machine-made art. [Stephanie Dinkins](https://www.stephaniedinkins.com/): Dinkins is an artist interested in the intersection of race, gender, aging, and our future histories. She uses AI to create installations and narratives that challenge the biases often inherent in these technologies. [Jake Elwes](https://www.jakeelwes.com/): Elwes's work focuses on the ethics and potential of AI. He creates pieces that question and explore the capabilities of AI, often highlighting its limitations and the peculiarities of its learning process. [Lawrence Lek](https://lawrencelek.com/): A London-based artist, Lek uses AI, virtual reality, and 3D simulation to create immersive cinematic environments. His work often explores themes of automation, AI, and the impact of technology on human experience. [Lynn Hershman Leeson](http://www.lynnhershman.com): Known for her pioneering work in new media art, Hershman Leeson has been integrating technology, including AI, into her cinematic and installation works for decades. Her work often addresses issues of identity, surveillance, and the relationship between humans and technology. [Jennifer and Kevin McCoy](https://mccoyspace.com/): This artist duo is known for their multimedia installations that often incorporate AI and machine learning to dissect and reconstruct cinematic narratives, offering new perspectives on familiar stories. [Trevor Paglen](https://paglen.studio/): An artist and geographer, Paglen's work often focuses on mass surveillance and data collection. He uses AI and machine learning to explore how these technologies shape our perception and understanding of the world. [Jon Rafman](http://jonrafman.com): Rafman’s work often delves into the effects of technology on consciousness and the digital experience. His films and installations, while not exclusively AI-focused, frequently explore themes relevant to AI's impact on society and narrative. [Eyal Sivan](http://eyalsivan.info): A documentary filmmaker who has explored the use of AI in the context of filmmaking, particularly in how AI can be used to analyze and interpret archival footage. [Hito Steyerl](https://monoskop.org/Hito_Steyerl): Steyerl is a German filmmaker and writer who explores the intersections of art, technology, and politics. Her work frequently incorporates AI to comment on the digital age and the proliferation of images and information. ## Readings: - Halperin, B. A., Jones, M., & Rosner, D. K. (2023). [Haunted Aesthetics and Otherworldly Possibilities: Generating (Dis)embodied Performance Videos with AI](https://studiolab.ide.tudelft.nl/studiolab/genai-dis2023/files/2023/07/Halperin-GenAI-Performance-Videos.pdf). In Toward a Design (Research) Framework for Generative AI: DIS’23 Workshop, July 10–14, 2023. - Manovich, L. (2023). "[AI Image Media through the Lens of Art and Media History](https://drive.google.com/file/d/1XPTG0xKgeMuRU1ZZmiHu-VIaMtGq2ZiH/view?usp=sharing)." The  Interdisciplinary Journal of Image Sciences 37. 1: 34-41. - Steyerl, H. (2023). [Mean Images](https://drive.google.com/file/d/1CkXwDyYgQIwBpg97RVLU0YgTiDVnHzph/view?usp=sharing). The New Left Review. - Yale Film Analysis Guide. [Part 5: Sound](https://filmanalysis.yale.edu/sound/).  - Recommended: Lastra, J. F. "[Teaching Film Sound](https://drive.google.com/file/d/10WRI9ogV73ZUSBFge7ARGeivmyLPe96Z/view?usp=sharing)." Teaching Film, edited by Lucy Fischer and Patrice Petro. The Modern Language Association 2012. - Jiang, H. H., Brown, L., Cheng, J., Khan, M., Gupta, A., Workman, D., ... & Gebru, T. (2023, August). [AI Art and its Impact on Artists](https://drive.google.com/file/d/1OVO0LKRHR1x9v9sDX4fZtphtDjvv6Rld/view?usp=sharing). In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (pp. 363-374). - Halperin, B. A. & Lukin, S. (2024). Artificial Dreams: Surreal Visual Storytelling as Inquiry Into AI ‘Hallucination.’ {Work in Progress - Do Not Distribute} - Halperin, B. A. & Rosner, D. K. (2024). ‘AI is Soulless’: Hollywood Film Workers Strike and Emerging Perceptions of Generative Cinematography. {Work Under Submission - Do Not Distribute} - Benabdallah, G., Alexander, A., Ghosh, S., Glogovac-Smith, C., Jacoby, L., Lustig, C., ... & Rosner, D. (2022, June). [Slanted Speculations: Material Encounters with Algorithmic Bias](https://drive.google.com/file/d/1oXXuMPWfE-idC9Ge3tuXBDvf4EKFpQh2/view?usp=sharing). In Designing Interactive Systems Conference** ## OPTIONAL ADDITIONAL READINGS - Buolamwini, J., & Gebru, T. (2018, January). Gender shades: Intersectional accuracy disparities in commercial gender classification. In Conference on fairness, accountability and transparency (pp. 77-91). PMLR. - Jones, M., Neumayer, C., & Shklovski, I. (2023, April). [Embodying the Algorithm: Exploring Relationships with Large Language Models Through Artistic Performance](https://drive.google.com/file/d/18grgzi02Hz9pji13DBZvFYPBMfd9ao-_/view?usp=sharing). In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. - Manovich, L. (2015). Cinema as a Cultural Interface (1997). International Journal of Transmedia Literacy (IJTL), 1, 221-251. - Manovich, L. (1995). What is digital cinema?. The visual culture reader, 2, (31 pages).  - Schröter, J. (2023). "The AI Image, the Dream, and the Statistical Unconscious." IMAGE. 37.1: 112. - Wasielewski, A. (2023). “Midjourney Can’t Count”: Questions of Representation and Meaning for Text-to-Image Generators. The Interdisciplinary Journal of Image Sciences, 37(1), 71-82. - Wilde, L. R. (2023). Generative Imagery as Media Form and Research Field: Introduction to a New Paradigm. IMAGE, 37(1), 6. - hooks, b. (2012). The oppositional gaze: Black female spectators. In Black American Cinema (pp. 288-302). Routledge. - Mulvey, L. (1975). Visual pleasure and narrative cinema. In Feminism and film theory. Routledge. - Bolter, J. (2023). "AI Generative Art as Algorithmic Remediation." The Interdisciplinary Journal of Image Sciences 37.1: 195-207. - Chow, P. S. (2020). Ghost in the (Hollywood) machine: Emergent applications of artificial intelligence in the film industry. NECSUS_European Journal of Media Studies, 9(1). - Bordwell, D., Thompson, K., & Smith, J. (2008). Film art: An introduction (Vol. 8). New York: McGraw-Hill. - Turnock, J. A. (2015). Plastic reality: Special effects, technology, and the emergence of 1970s blockbuster aesthetics. Columbia University Press. - Turnock, J. A. (2022). The Empire of Effects: Industrial Light and Magic and the Rendering of Realism. University of Texas Press. - Pierson, M. (2002). Special effects: Still in search of wonder. Columbia University Press. - Warner, K. J. (2017). In the time of plastic representation. Film Quarterly, 71(2).  - Gunning, Tom. "The cinema of attraction [s]: Early film, its spectator and the avant-garde." Theater and Film: A Comparative Anthology (1986): 39.  - Halperin, B. A., Hsieh, G., McElroy, E., Pierce, J., & Rosner, D. K. (2023, April). Probing a Community-Based Conversational Storytelling Agent to Document Digital Stories of Housing Insecurity. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (pp. 1-18).