Stylized notifications from Costar, Tiktok, and a contact named Soulmate.

Issue 4


Oracles in the Machine

written by Zora Che

edited by Kyle Barnes

art by Tiffany Wang

Think of the computer, not as a tool, but as a medium.

—Brenda Laurel, Computers as Theatre

If you are reading this, this is for you.

My finger makes gentle contact with the glass matrix, caressing ceramic nano crystals and giving up heat to the obsidian screen. My phone, an inanimate 6 oz. piece of matter (the same weight as a deck of cards), is whispering soft nothings to me.

My face is enveloped by a soft glow as I attentively await my reading. “Tonight,” a smooth voice whispers, “we have the queen of cups... complete unison and harmony with this universe.”

I have been keeping quiet about my nighttime divination; private in my own room, yet aware of the publicity of the readings I receive—the videos are filled with strangers’ comments and reactions. The readings are for me, and for an indeterminate number of us. I scroll down the infinitely extending screen to unravel more stories for myself, until my eyes are heavy and I drift off.

During the day, I pry open models, reverse-engineering their inner workings and meditate on their societal ramifications.1 At night, I scry with opaque models, heart yearning for answers, fingers absorbed. I receive group tarot readings on Instagram, from someone with a beaming smile soaked in sun rays. The tellers are charismatic, the readings emphatic. One-in-four young women in the United States uses Co-Star, the AI astrology app―I am one of them. I code models, and I let models recode me.

Divination is a practice as old as human time. The word divination has the Latin root of divinare: “to foretell, prophesy, forebode, divine the future.” Ever since the concept of the future entered the realm of reckoning, humans have been weaving together what's around us to foretell the future. We gazed at the movements of clouds (nephomancy), fluffed up the flour (aleuromancy), poured wax (ceromancy), and dealt cards (cartomancy). Anthropologists Ze Hong and Joseph Henrich assert that viewing divination as technology has a long history in anthropology, and that divination practices are often best viewed as specifically an epistemic technology: technology for knowledge-making. In ancient Greece, divination (mantikē) was categorized as a kind of technique (technē), which is the etymological root of the modern English word “technology.” Plato describes two types of divination: possession divination where a diviner reveals information acquired from a deity, and technical divination, where the diviner interprets the hidden meanings of phenomenons, both of which have been considered as technē by the Greeks (Hong & Henrich, 2021).

Cultural anthropologist and oneirologist2 Barbara Tedlock notes that divination involves a primary process of seeing, and secondary processes of thinking or knowing. In her view, diviners are “specialists who use the idea of moving from a boundless to a bounded realm of existence in their practice.” The key to good divination is the weaving of useful knowledge from open-ended oracular observations. Tedlock argues that “[Diviners] link diverse domains of representational information and symbolism with emotional or presentational experience.” In this lens, divination is a way of knowing, with subjective validity and narrational depth.

As new technologies seemingly bloom beside us—seeded by companies eager to capitalize off our every last moment of attention—they too become a part of divination. The fortune-telling automatons from the era of penny arcades are precursors to the divinatory apps in my phone: divining with obscurity and novelty, a few bucks to outline the future and to be entertained.

Computational divination extends divination as a way of knowing to further imaginative depths. Computers, after all, are a natural medium for building worlds. In Brenda Laurel's seminal work Computers as Theatre, she argues that in the field of human-computer interaction, we may create “imaginary worlds that have a special relationship to reality—worlds in which we can extend, amplify, and enrich our own capacities to think, feel, and act.” Laurel emphasizes the metaphor of theater as an attempt “to represent something that might go on, simplified for the purposes of logical and affective clarity.” The expressive capacity of computation and computer interactions creates a wide stage for divination. In a 1995 Wired review of a Tarot software on CD, the reviewer praises the convenience of divining via software and notes that softwares have “jazzy features” including an “‘Gate of Entry’, an enticing virtual stone doorway,”and “ethereal New Age music.”

Effusive group tarot readings on Instagram; Co-Star's minimalistic, mysterious interface; and the illustrated “Gate of Entry” in 90s tarot softwares are examples of computational divination that fall under a category of phantasmal media, which D. Fox Harrell, professor of digital media and artificial intelligence, defines as “a combination of imagery (mental or sensory) and ideas.” He notes that phantasmal media is “a particularly pervasive kind of imagination, one that encompasses cognitive phenomena including sense of self, metaphor, social categorization, narrative, and poetic thinking.” Interactive, beautifully embellished divinatory content, even when taken as entertainment, creates a sounding board for one to renegotiate and reimagine their stories, metaphors and narratives.

Artificial intelligence-based models are the genies out of the bottle in the current wave of digital divination. Various divination tech startups tout AI, which is simultaneously common in its adoption among all kinds of businesses as a flashy feature for users’ benefit yet arcane in how exactly it is referenced and used. Co-Star promises users that it is able to give personalized astrological reading by using artificial intelligence to analyze NASA Jet Propulsion Lab data, thereby finding patterns in a user's planetary transits. Tarot readings show up on Instagram through artificial intelligence-backed personalized content ranking. The core of popular algorithmic divination from the receiving end is randomness and opacity. Under the hoods of my digital oracles lie models which have learned from my data and interaction patterns. Yet I have no interpretable context for what I receive. On one hand, speculations, randomness; on the other hand, a stroke of luck, a sign from the universe.

At times, my daytime profession of probing models leaks into my nighttime devotional hours. Currently, many models are opaque and there’s a lack of adequate tools to interpret them. Neural networks, the most popular and widely used class of models, are trained without symbolic reasoning. As such, even when we lay out a model’s internals—the model’s architecture and weights—we don't always understand what we see. How and why does the model generate its answers? Much of the algorithms that touch our lives are opaque and uninterpretable, qualities which pose real concerns for fairness and safety. When I divine with models, I am divining with a black-box.

The popularity of algorithmic divination today is inseparable from the opacity of the black-boxes that set out to personalize for users in a world of unlimited choices and data. The promise of the personal in models parallels the need to hear about and think about one's reality. Co-Star's co-founder Banu Guler says in an interview, “The crux of feeling like a human is being able to talk about your reality.” The language of the messengers is telling: Co-Star presenting “your day at a glance”; TikTok promising fresh content “for you”. Much of algorithmic divination aims to deliver snippets that reflect on your reality: predictions of what is “in power” for you, what is troubling for you, how you relate to home and to work, what the year has in store for you. And with the premise of personalization, divinatory products argue that we have scaled understanding of individual realities. It's tempting to connect the promise of the personal to divine messages “just for me,” though the model only sees a skewed version of me that it further echoes.

In sociologist Charles Cooley's theory of the “looking glass of self,” we understand ourselves through the perceptions of others. Online, models perceive us, responding to and reinforcing the versions of ourselves which they glean from our behaviors. They sense my finger lingering, my invisible gaze apparent by the gap of my movements. My understanding of my digital self and my digital reality becomes a feedback loop churned by models I cannot see. Moreover, the model only “sees” me as data that can be optimized for objectives that I cannot uncover. That objective is something closer to optimizing my time spent on the digital product than to holding my deepest needs; the latter perhaps was never a mathematical question to begin with.

Divination and algorithmic opacity both appear to bring us what we cannot see. Diviners see through what is obscure and beyond our comprehension: it may be incomprehensible pain and grief, vertiginous lack of control, and/or the unwarranted future. The opacity of divination comes from the limitations of our own knowledge. But the opacity of algorithms comes from both the algorithm itself and the socio-technical infrastructure that it was built around. Jenna Burrell writes of three layers of opacity in models: “(1) opacity as intentional corporate or state secrecy, (2) opacity as technical illiteracy, and (3) an opacity that arises from the characteristics of machine learning algorithms and the scale required to apply them usefully.” As consumers of models, we interact with the first and third layer of the opacity―that of platforms hiding models from us, and that of the gap between what the model is optimizing for and what may be explainable. The black-box model is an alluring oracle, interacting with us in inexplicable ways: no explanation for the daily laconic message Co-Star pushes to its users, no logic behind why you received this tarot reading while scrolling, no insight into the models behind these oracles and their objectives.

The objectives hidden from me hold power. Their inexplicability is part of their charm. Without an explanation, I may build my own story for it. Would my reading be as meaningful to me if I could trace the clear thread of why it was pushed to me? That perhaps my scrolling patterns mirrored someone who was fresh out of a relationship, that the algorithm learned I wanted to hear whispers of comfort, wanted to feel digitally held amidst change? That it knows I eat up soft nothing prophecies of someone coming towards me with apologies like I’m watching junk TV? Delusions are a type of dopamine hit.

When asked about astrology, Banu Guler says “[Astrology] is a form of self care. I think it’s also a way of, sort of, collective care. Right? Maybe ‘collective self care’ is the word … this idea of building relationships with each other and taking care of each other.” Diviners are traditionally viewed as healers integral to communities. In Ingrid Rojas Contreras's memoir The Man Who Could Move Clouds, she writes of growing up in Colombia in a family laden with magic, her grandfather a renowned curandero and her childhood filled with her mother's fortune-telling clients. Rojas Contreras writes of giving readings in her youth, eventually stopping as she felt the weight of divination. She learned the art from her mother and noted her mother's confession that “the biggest thing I have learned all these that nobody wants the truth, but everyone wants a story.” Rojas Contreras writes of the heaviness of divination for her classmates as political violence loomed in the background: “I dealt cards despairingly, for the first time understanding the responsibility of divination. It wasn’t just about precision; it was about interpreting the language of grief and finding there a sense of direction that had eluded the client. But violence was nonsense. What possible answer could there be in the face of it?”

When diviners are healers, the healing comes from helping people rewrite a story: giving people a direction that may allow the easier digestion of truth. The core of the care, then, is the reader’s personal connection with their client. If you’re a skeptic, the personal connection may be just a cold reading of a client. If you’re open to magical thinking, though, the personal connection is the inexplicable intuition of another.

For algorithmic divination to be an instrument of collective care, one needs to examine the objectives of what's under the hood. Are Co-Star’s push notifications catering to me, or catering to seeing itself being screenshotted, and going viral on social media? The inexplicable push of Instagram content at me is optimizing for a version of me, but is that version of me most aligned with my real values and goals? Is it possible for these technologies to care for us without really understanding us?

We should be reminded that mysticism and obfuscation serve the aims of tech companies. Mysterious, personalized prophecies have a global following. The global astrology industry was valued at $12.8 billion in 2021; Co-Star is a multi-million dollar company. And, on Instagram, I am joined by millions of others who consume and like group tarot readings. Both high-tech and low-tech divination make for good products. Beyond divinational tech, the more inexplicable and magical products appear to be, the more we are sold a story of tech's inevitability and reinforcement of the status quo. Is generative AI so mystical, when models are trained on millions of data, and research has shown that they reliably copy from their training sets? While companies have the ability to demystify algorithms, doing so would reveal their failings. It is actively in their corporate interest to present the most magical and infallible versions of their products. From the rise and fall of Theranos to the unveiling of a Google AI demo video as fabricated, we have been swimming in grand illusions and mysteries served by tech companies.

We are in a new wave of public fervor in astrology and magical thinking. Maybe we can blame it on the economy: seeking to understand the future and ourselves is interlinked with seeking control. Social scientists Jennifer A. Whitson and Adam D. Galinsky conducted experiments that show how lacking control increases illusory pattern perception among a set of random or unrelated stimuli. They note that, “even on a national level, when times are economically uncertain, superstitions increase.” Google searches for “birth chart” and “astrology” both hit five-year peaks in 2020, the first year of the Covid pandemic. When algorithms are hidden from us, anything we receive seems like a stroke of luck. And when we really need a story—about our life, about the things we cannot control—perhaps recommended content becomes prophecy.

We are all divining in our own ways: thinking about and predicting the future, imbuing meaning in scientific or occult methods. Algorithmic divination introduces a greater scale of dissemination, as well as opacity and randomness to this age-old practice. In incorporating algorithmic divination as a way of knowing, we are also seeing the world through the products that are stamped with objectives that may differ from our own. To acquire better understandings of ourselves, there exists an inherent risk as we weave through unreliable or perhaps conflicting narratives. Deep collective care requires true understanding and attention. In the meantime, pick your brooms wisely.



  1. I research algorithmic fairness and interpretability of machine learning models. Some questions I've been interested in are: can we preserve fairness under changing data distribution? Can we efficiently find subnetworks for a task in a large model?

  2. A scholar of dreams

headshot of Zora Che


Zora Che

Zora is an artist and researcher. She's weaving metaphors and probing models.

headshot of Kyle Barnes


Kyle Barnes

kyle barnes is a researcher and technologist who works on complex problems in science and society that disproportionately impact the most vulnerable. Based in Brooklyn, NY, they spends most of their time living well on a damaged planet.