Instructions for Humans

Instructions for Humans - A Cargo Cult for Artificial Intelligence

A new work by Pete Ashton explores human interactions with complex systems, centred around and lending it’s name to a wider Instructions for Humans exhibition exploring Internet art in an age of mass surveillance at BOM (Birmingham Open Media) from 13 September to 16 December 2017.

Funded by Arts Council England, the work will be available for touring from 2018.

The phenomena of cargo cults was observed on South Pacific islands after WWII as islanders used ritual to try and make sense of the advanced military technology that had suddenly entered their lives. Instructions for Humans asks: what might a cargo cult for artificial intelligence look like?

The work uses recent developments in artificial intelligence and machine learning tools to explore the systems that guide society and our relationship to them. The work will address the following issues:

  • How machine gesture informs the human creative process and in turn how human gesture might inform mechanical representations.
  • What it means for a computer to “see”.
  • How society can be influenced by opinions derived from the perceptions of machines.
  • How interrogating mechanical systems can help us to question the biases of our own sense-based cognition.

The Work

Instructions for Humans will comprise digital, sculptural, and performance artworks informed by machine learning programmes. Centred in a gallery it consists of:

The Black Box - Representing the proprietary, closed systems such as Google and Facebook, which can be only partially understood through interrogating them, a monolithic black box periodically issues instructions on pieces of paper which fall to the floor. Visitors are invited to take, and attempt to follow, an instruction from the pile.

These instructions, which may or may not make sense, are generated from an artificial neural network trained on the documents and texts that define and control the city of Birmingham, from dry government policy documents to sensationalist news feeds.

The Instruction Station (working title) is a workspace where Ashton will be resident throughout the exhibition developing new artworks daily, often in collaboration with visitors and other artists. These works will attempt to make sense of our relationship with opaque algorithmic systems, using the tools our ancestors employed to talk to their gods to ask what would a cargo cult for artificial intelligence look like?


A significant outcome of Instructions for Humans will be two new performance works developed with artists Emily Warner and Aleks Wojtulewicz to premiere in November.

  • Emily will attempt to materialise the data streams through movement and gesture, building a feedback loop between her body and the system.
  • Aleks will have his muscles connected to electrodes controlled by the artificial intelligence system turning him into a puppet of the machine.

A programme of evening and weekend workshops and excursions exploring the digital and physical systems of the city will take place throughout November and December. Covering surveillance, data security, city planning and more, these “data survival” events will explore work from artists, academics, activists and privacy campaigners and involve experts in these fields.

There will also be workshops and talks looking at practical uses of machine learning technology with an emphasis on artistic production.


Work produced in the gallery will be documented and archived on the website

This work is the culmination of three years’ research by Pete Ashton. Lists of relevant books, articles, talks and artworks will be made available on the website.

Instructions from The Black Box will be collated in a publication along with guidance on using them to inspire creative practice.

Why Machine Learning?

Machine learning systems, commonly called “algorithms” and “artificial intelligence”, are complex statistical programmes that use vast quantities of data to predict a likely outcome. A simple example is predictive text or autocorrect which notices the phrases you commonly type into your phone and suggests them to you. A more controversial case is predictive policing which uses historical crime data to suggest where police resources would best be deployed. In all cases, the algorithm is “trained” on a corpus of data and all its results are constrained by the quantity and quality of this information.

A significant part of machine learning in recent years has involved image analysis and computer vision, for use in areas such as the development of self-driving cars and next-generation surveillance. This has caught the attention of both visual artists and privacy campaigners and forms the underlying structure of this body of work.


The archive website is

Development blog posts:

A secondary blog,, is collecting examples of humans performing for the benefit of machines. Further examples are welcome.

Research and development materials and code will be archived at this Github repository.

Suggested reading materials can be found here.

Main image stolen from this page



+44 7775 690106