Travis DeShazo is, to paraphrase Cake’s 2001 song “Comfort Eagle,” building a religion. He is building it bigger. He is increasing the parameters. And adding more data.
The results are fairly convincing, too, at least as far as synthetic scripture (his words) goes. “Not a god of the void or of chaos, but a god of wisdom,” reads one message, posted on the @gods_txt Twitter feed for GPT-2 Religion A.I. “This is the knowledge of divinity that I, the Supreme Being, impart to you. When a man learns this, he attains what the rest of mankind has not, and becomes a true god. Obedience to Me! Obey!”
Another message, this time important enough to be pinned to the top of the timeline, proclaims: “My sayings are a remedy for all your biological ills. Go out of this place and meditate. Perhaps some day your blood will be warm and your bones will grow strong.”
This image was algorithmically generated based on a GPT-2 Religion A.I. passage that reads: “Sacrifice and pagan rites, indulgence, and lust, pleasure and violence; these threefold shall I ever worship, with no other motive than reverence for the Gita and Truth.”
Before we get any further, no, these aren’t real holy verses. Instead, DeShazo, a 30-year-old lab technician whose day job involves “working predominantly in lubricant base stocks refining,” has created a bot that’s trained to spit out pseudo-biblical verses. In the same way that a Benedictine monk lives a life of service and religious study, allowing themselves to be shaped and molded, the entire raison d’être for DeShazo’s GPT-2 Religion A.I. is to learn from its massive corpus of religious training texts and to turn these into new insights for its followers (currently a still-niche 3,174 on Twitter). It’s powered by OpenAI’s GPT-2, the impressive learning model that preceded the more recent — and much larger — GPT-3.
“The GPT-2 Religion A.I. is a Twitter bot that publishes curated passages generated by a natural language processing model that has been trained on English translations of ancient and modern world religious texts and myths,” DeShazo told Digital Trends. “The outputs adopt the style, themes, and diction of the source material and combinations thereof, but are also able to extrapolate and impart unique insights given the capability of the base model upon which it was trained. At one point I employed the tagline: ‘Revealing humanity in the latent spaces of the divine.’”
The Benedictine bot
For training data, DeShazo’s digital deity absorbed a corpus of data including the Bible (both Old and New Testaments); the world’s second-oldest religious text, the Epic of Gilgamesh; the ancient Indian collection of Vedic Sanskrit hymns known as the Rig Veda; fragments of the religious texts of Zoroastrianism that are the Avesta; the Bhagavad Gita; intact portions of the Nag Hammadi; the Tao Te Ching; assorted Neoplatonist texts; and others. As with other examples of computational creativity — whether that’s the A.I. that wrote a Scrubs script or the one that tried to beat George R. R. Martin to finish his A Song of Ice and Fire saga — the output is an intriguing mash-up of the original works and something entirely new.
Like with any good religious founder, DeShazo’s work has inspired others. One, a 22-year-old student named Bokar N’Diaye, who is studying the anthropology of religions and history of arts in Geneva, recently built an image generator that’s able to conjure up painterly images from any line of text you prompt it with. One of those images (below), which was shared with the world on Twitter, was inspired by a line of the GPT-2 Religion A.I. bot: “As you chant, so I weave my voice. I craft my strings of fire, I fill the world with my own sweet sound. The stars themselves, the ends of the land, the wind, even the place where the light goes out all bend to worship me.”
As seen by #latentvisions by @advadnoun https://t.co/WRx2BlsIc3 pic.twitter.com/tbO8KPt1bY
— Bokar N’Diaye (@bokar_n) June 3, 2021
The resulting figures look like a pair of Hindu Goddesses. One resembles Lakshmi, the goddess of wealth, fortune, beauty, and fertility, only holding Sarasvati’s veena instrument.
“Both art and religion are great to elicit subjective input from the spectator, because we’ve been primed to interpret them far more subjectively and with a different set of expectations than mundane text,” N’Diaye told Digital Trends. “When a program like gods_txt mentions ‘crafting strings of flame,’ you’re not likely to interpret it as the A.I. failing to understand realistic human sentences. Instead, you’ll strive to decipher the secret meaning of the sentence, or relate to its aesthetics or the feelings it rouses within you – just like you’ve been taught to do since childhood, when it comes to religious texts and/or art.”
Machines of loving grace
Applying machine smarts to religious iconography is controversial territory (you could say that it puts the A.I. in “sacrilege” territory). But it’s a link that, perhaps, isn’t as tenuous as it could be. On one level, technology, and specifically artificial intelligence, is hyperrational. It assumes that the mind can be recreated in hardware, as opposed to wetware, by modeling the behavior of individual neurons and other brain-based apparatus inside a computer. There is, it suggests, no such thing as a soul, since that would presumably cause no end of headaches on the road to human-machine parity.
But there is another side to the argument that is inherently religious. There is no shortage of tech figures whose vision of the future of technology carries a distinctly religious zeal: Steve Jobs with his cathedral-like Apple Stores and products that trigger a religious experience in the brains of fans. Google engineer Ray Kurzweil’s bestselling non-fiction book, The Age of Spiritual Machines. Kevin Kelly, the founding executive editor of Wired, who titled his 2010 book, What Technology Wants and, in it, described the force he calls “the technium,” a “global, massively interconnected system of technology vibrating around us.” And Richard Brautigan’s poem, “All Watched Over by Machines of Loving Grace.”
Even the broadly accepted trajectory of A.I. — that a tool born of humble means to carry out the dull, dirty, dangerous jobs will eventually become our master — has faint echoes of an historical figure born in a stable in Bethlehem who rose to become humanity’s savior. In short, are we building robot servants or overlords?
Heck, A.I. even has its own end-times judgment day promise. Some A.I. advocates eagerly await a world in which humans, freed from their corporeal, fleshy prisons, “live” as downloaded consciousnesses inside machines, achieving immortality through digitalization; a paradise of work-free, worry-free existence lived in perfect virtual bodies. Marvin Minsky, one of the founding figures of A.I., once wrote that: “Eventually, we will entirely replace our brains using nanotechnology. Once delivered from the limitations of biology, we will be able to decide the length of our lives — with the option of immortality — and choose among other, unimagined capabilities as well.”
“It’s interesting to think about these unknowable, uninterpretable, and potentially malevolent entities that can make massive decisions affecting our lives,” Ryan Murdock, an engineer who created some of the generative image technology N’Diaye uses, told Digital Trends. “What’s funny to me is that [that sentence] could be about either machine learning systems or Eldritch gods. There is something alien and disturbing about having neural networks make increasingly important choices that parallels religion in some ways, I think. While these systems can be shockingly dumb, they can also exhibit intelligence and performance that clearly rivals human intelligence in specific domains — and they work in ways that I doubt we’ll ever fully understand.”
A.I., it seems, works in mysterious ways.