Artificial Wisdom

Zen Teachings generated by GPT-3

- Artificial Wisdom was selected for exhibition at Insight Conference 2021 -

The intention was to examine the use of Artificial Intelligence in spiritual writing.

Alan Watts' Zen teachings appear to have a style which can be modelled and replicated analytically. This allows us to produce similarly moving teachings without a living author. It suggests that the state of being characterised by zen training may be in some way quantifiable. 

Buddhism describes a living state of enlightenment, or non-duality, from which decisions and actions flow in perfect harmony. If we are building AI to replicate our behaviours, we might seek to model this state.

alan-watts-hokusai

A portrait of Alan Watts combined with Hokusai's "Great Wave" using Deep Style Transfer

architecture-bird

"When the mind takes a photograph, it does not capture the motion of life."

"You are not a person, you are the space between your thoughts."

buttered-landscape-1

"When you open yourself to the universe, the universe opens to you"

Zen and other spiritual disciplines train certain personality traits of patience, wisdom and compassion. These characteristics are reflected in the speech of the practitioner. Using a 175 billion parameter model, the GPT-3 algorithm learns to copy the style and context of the input text by predicting the next most likely word. After a number of example quotations by Alan Watts were passed through, the model increasingly learned to approximate and replicate them.

The model traces the way Alan Watts' statements move through the language space, and in a way, we are modelling an aspect of his personality described by the form of his speech. His personality and ability to speak, in turn, has been sculpted by his experience in Zen training. 

earthworm-cosmos-1-1

"You can't learn without asking questions,

and you can't ask questions if you already know the answers."

Is there a link between states of consciousness and linguistic output? 

- If sacred and prophetic writings are transmissions from an external or deeply subconscious source, then the linguistic structure of these writings are of particular interest -

Using GPT-3 to generate new teachings in the style of the Tao Te Ching, the work asks if there is a link between the generated content of the GPT-3 model and the conscious state of the original author. The AI-generated writings are presented alongside original examples, asking the viewer to come to a qualitative judgement as to their similarities and differences. We typically use our intuition as a judge of character, and so a qualitative impression will determine the algorithms ability to emulate human wisdom.

The goal of this project is eventually to attempt quantitive metrics for wisdom in speech, such as markers for compassion, understanding and acuity. There are many potential applications to this, since they would effectively enable adding "filters" to a piece of text and seeing how conversations might go differently. 

In a society which will increasingly feature A.I. generated text, trained from our existing speech habits, we can install a positive feedback loop by which human and machine language may influence each other, favouring chosen desireable traits.

Selected Works

StyleGAN VideoProject type

Knowhere RadioProject type

INSIGHT Conference 2021Poster Presentation

Wherefore I AmSculpture

Tree of WisdomSculpture