An Odd but Elaborate Experiment
Part 3 of this week's series on algorithms/analytics/prediction/divination . . . .
If you’d like to refresh on this week’s plan, just click here. Meanwhile—
On a Lighter Note
There’s a lot to cover in this episode, so the Lighter Note will have to be short. Did you know there is a British sketch comedy troupe called “Tarot”? I didn’t—but there is. The audience draws Tarot cards to determine which of the “anarchic,” “bark-out-loud funny” skits will be performed in each show.
Once I started looking for Tarot humor, I found “Oracle Vs Tarot Comedy Skit” on YouTube. And laughed out loud through the whole thing! Really. (Fair warning: there’s some salty language.) The comments are funny too, and often insightful.
This is the story of a scholarly project, carried out by three graduate students at the University of California, Berkeley. It was designed to find out how people respond when they know their Tarot “readings” are being generated by Artificial Intelligence.
Exploring the “Magic” of Algorithmic Predictions with Technology-Mediated Tarot Card Readings
Joyce Lee, Sejal Popat, Soravis Prakkamakul—2019 MIMS Final Project
The conceptual structure and final outcome of this project are not easy to explain, so I encourage you to read the report for yourself. For a quick take, there’s an infographic that (sort of) explains the project approach.
In the meantime, here’s my best attempt at a summary:
The point of the whole thing was to find out whether people would see an algorithmic prediction differently if it were packaged into an experience that creates an impression of divinatory mystique. The “Background” section of the final report goes into detail about different types of divination, and offers some analysis I found quite interesting.
The authors don’t really explain why they chose Tarot as the divinatory medium for their experiment, but I’m guessing it was because of the visual aspect, as well as the immediacy of a reading event. The gist of their protocol involves having the participants go into an enclosed booth, answer nine questions posed by a “chatbot,” and receive three Tarot cards—representing Past, Present, and Future. They are interviewed afterwards about the experience.
Behind the scenes, a set of predetermined algorithms extrapolate data from a participant’s question responses, and apply multiple techniques in order to choose which cards they will receive. The participant also sees interpretive text related to the cards they receive.
The number of participants was so limited (five men and five women, similar age range and education) that nothing can be concluded on the basis of their interview responses. But as far as I can tell, no one expressed highly positive feelings about the experience.
The authors of this project appear to have been concerned primarily with how participants felt about the process. I found nothing in their analysis about the divinatory results.
However . . .
I think there’s something to be learned from all this, when it’s viewed from a Tarot perspective. And what I find most fascinating is the complicated scheme devised by experimenters to create the illusion—or at least impression—of a Tarot reading.
In actuality, the supposed “reading” was the reverse of what we normally think of. Instead of dealing or drawing the cards and then discussing them, the cards are chosen by the program on the basis of the discussion. That probably makes sense in terms of the project design—but I’m guessing it also reflects the general assumption that Tarot readers construct their interpretation of the cards by picking up cues from the querent.
Somewhat mysteriously, the questions used to categorize participants and decide on their cards were based on a model meant to facilitate “falling in love.” The questions were modified slightly to create a more technology-oriented slant.
Now we get to the logic of the card-choosing process. First, in order to determine the Past card—always from the Major Arcana—the program analyzed a participant’s responses to the question sequence, and created an ad hoc Meyer-Briggs assessment.
To refresh, Meyer-Briggs is a psychological testing system that sorts individuals into 16 personality types, based on 4 so-called dichotomies: extraversion/introversion, sensing/intuition, thinking/feeling, and judgment/perception.
As a framework for assigning Past cards:
each of the 16 possible personalities was mapped to a Major Arcana card, using descriptions from reference websites for both the Meyer-Briggs framework and the tarot card meanings.
“Reference websites” for Tarot were learntarot.com and tarot.com.
Okay then! Here’s a sample result:
A user classified with introversion, intuition, thinking, and perceiving traits (INTP) would receive the “The Magician” card for their past, as well as the following prediction:
“The Magician card is a reminder that you are a unique being, and have many gifts that others do not hold. You are inventive and creative, and the Magician embodies these qualities. In the past, you have been singularly focused on originality and efficient results. However, this can make you prone to over analyzing your own thoughts and theories, to a point where you can stagnate in self-doubt.”
If you’re curious about other Major Arcana assignments, the project report contains a complete table of correlations and messages for the Past card.
The Present card was mapped to the Minor Arcana, and built on an entirely different model—using keywords that subjects might have mentioned in their answers to the question set. Here’s the protocol:
Worth noting—the suit assignment of common apps and online resources. For example: Google goes to Swords, Netflix to Wands, Tinder to Cups, and Amazon to Pentacles.
So now we’ve reached the Future card. I was extra-curious to see how they approached this, but it’s a bit disappointing . . .
Like the “present” tarot card, the “future” prediction was also mapped to a Minor Arcana card. The suit and card number were identified using the same methods, but based on responses of the three questions focused on the future; measures were also taken to ensure that the user did not receive a future card in the same suit as the “past” card.
Here’s the list of chatbot questions, mapped to the three timeframes:
And for context, here’s the start of the question process:
I’m still in the process of formulating some thoughts about this whole endeavor, so I may have more comments when we get to the summing up of this week’s series. But briefly, I’m much less interested in how the “experiment” turned out than in how it was set up—and what that tells us about perceptions of Tarot at the intersection of academics and popular culture.
I want to leave off with one other point worth contemplation. The Tarot cards dealt out to participants were taken from a custom-designed deck, created like this:
The main tarot card imagery was sourced from public domain image libraries online, rendering a more traditional aesthetic. Each design also included four relevant icons as more modern visual accents, to contextualize the meaning and interpretation of each card. The imagery was then run through a texture style transfer algorithm to add a consistent and futuristic feel.
Here’s the result:
And no—there isn’t a Fool.
I’m going to leave it at that for today’s episode. Tomorrow takes a different, but equally thought-provoking turn.
Warmest regards, Cynthia