Divinatory Recommender Systems

Recommender algorithms these days decide on the news we see, the products we buy, the decisions we make. They are the backbone behind the Amazon’s “if you like this book, you will also like…” feature, Google context advertising, and Facebook’s newsfeed. And while they work well in expanding the already existing domain of knowledge, they fail when it comes to learning something completely different, new, and innovative.

The standard approach most guidance systems and recommender algorithms use is to sample what the majority of your surrounding likes and recommend you the things from that surrounding that are missing from your current set of options (see Special Agency’s AI project). If you like Leo Tolstoy but you don’t yet know Fyodr Dostoyevsky, this will be helpful. However, if you want to get out of that “filter bubble”, you need an approach that would take you towards innovation, not towards conformity with everyone else’s tastes. You might want to find out that contemporary author David Mitchell (“Cloud Atlas”, “Bone Clocks”) weaves epic-like narratives reminiscent of Tolstoy and Nabokov even if none of your friends have read those authors together.

The same goes for the guidance that is related to decision-making. The standard approach is to base oneself on the set of existing possibilities. This can work well in some cases and lead to catastrophic consequences in some others (as Nassim Taleb’s book “Black Swan” so vividly shows). In order for a “dramatic organic” change to occur, an innovative leap is needed – a movement that goes beyond the current set of possibilities and – yet – is connected to it. This is normally referred to as “serendipity”, although not in the way most recommender systems understand this term. This advance cannot happen in any random direction, there are certain principles that help bring it towards evolutionary development. This is where divinatory practices come into play.

Many people understand “divination” as some kind of obfuscated form of fortune-telling. However, at its basis divination is a practice of prediction. To “pre-dict” is to create narratives before they occur in real life. So it has less to do with knowing the future and more with actually designing it.

Divinatory practices have several unifying principles that help make that “future design” evolutionary, robust and adaptive at its core. Just like many scientific studies on robustness and adaptability have shown (Kitano 2004Paranyushkin 2012), those principles are:

1) Modularity (independent clusters that are connected on the global level);
2) Local and Global Feedback (system control, action and reaction on the local and global levels);
3) Redundancy (some functions are performed by several elements, repetition, cyclicity);

If one studies any divinatory system, such as Tarot, Astrology, Qabbala, I Ching and others, they have all those principles implemented at their basis. For example, in Tarot (see Nodus Labs research article on divinatory networks), every of the 22 cards in Major Arcana is linked to 4 other cards (principle of Feedback) and when those connections are mapped as a graph, certain clusters of cards that are connected more closely emerge (principle of Modularity). Moreover, Tarot cards follow two parallel narrative lines, and the cards between those two narratives are related as symbolic synonyms (principle of Redundancy).

Therefore, when one asks a question and puts it through a divinatory system, that system will always “pre-dict” the possible narratives that will be based on the principles of modularity, feedback, and redundancy. In other words, every divination is an advice on how to make one’s future more robust, adaptable, and – in the end – evolutionary.

When applied to single decisions or to simple recommendations, divinatory practice will enable one to make the choices that lie outside of the obvious, towards the choices that introduce more heterogeneity (modularity) into one’s worldview, but that still connect to where one’s coming from (feedback).