The Curse Conversation with DMFSTCTN

IMPAKT Festival 2022 The Curse of Smooth Operations

With just two weeks left until the start of the festival, it’s time for the third instalment of The Curse Conversations: a series of interviews in which we delve deeper into the themes of the festival. We talk with artist duo DMSTFCTN about GOD MODE (ep. 1), which they will perform on Saturday 5 November with live music by HERO IMAGE 

GOD MODE (ep. 1) is an interactive audiovisual performance exploring the use of simulation in artificial intelligence training. Throughout the performance, the audience can interact with the simulation and modify the appearance of the environment u​sing their phone. Why is this interaction important for you? And what are you hoping the audience takes away from their interactions?

Real-time 3D simulations similar to the one in GOD MODE (ep. 1) are used to generate training datasets for computer vision systems. Given the increasing presence, potential and power of such systems, as the current state of the art, we are interested in modeling and mimicking them to understand them, together with others.

There is a kind of immediacy and satisfaction that comes from a simple interaction generating a complex visual change on screen, and introducing this into the performance allowed us to demonstrate a technical system while also carrying the story forward. Specifically, in GOD MODE (ep. 1) we turned a common AI training technique used in similar real-time simulations into a simple game mechanic whereby anyone can change the appearance of the simulated environment by pressing a button on their phone. The audience is first introduced to the concept passively through the story, as the AI protagonist explains and complains about the constant changes in its training environment, and then actively through cooperative gameplay, which helps the AI progress through the training.

In your previous works, there is a strong focus on visualizing money flows and tax havens. What inspired this shift in focus to the seemingly mundane setting of a 3D-simulation of a supermarket? And what do you think of the increasing presence of such digital processes in our everyday lives?

In GOD MODE (ep. 1) we wanted the simulated environment – a cashierless supermarket – to feel mundane. The supermarket is a familiar interface to a large datafied, technological system, and being able to relate to it makes for an ideal setting to explore and test the claims that technologies of AI and simulation bring about the organisation of the world. This is a process and focus that we carry over from previous works.

What interests us here is less the application of AI than the process to train AI, epistemologically and technically, with questions around the relationship between language, perspective and perception, as well as questions around how (procedural) content generation, interaction and narration can interlink within highly adaptable simulations running on real-time rendering software.

The name suggests that this will not be the only episode of this project. What is your intention with the name and is there going to be a next episode?

GOD MODE (ep. 1) is the first output from the ongoing series GOD MODE. The name directly refers to the common video game cheat code “god mode”, which is generally used to obtain some form of player-character invincibility. There are many historical examples of AI agents discovering ways to cheat training simulations – and the AI protagonist of GOD MODE (ep. 1) finds its own way.

The series comprises works developed and presented in real-time rendering software, or game engines, and we are currently working on two more outputs – a multiplayer game extracted from the first episode, to be released later this year, and a second episode further developing the story and the character introduced so far.

Music plays an important role in the performance. Can you tell us a bit more about the composition that will be performed live by HERO IMAGE?

The soundtrack consists partially of contemporary royalty-free background music and of notification sounds found in consumer devices, simulation architecture and video game streaming. After seeing early prototypes of the simulation, HERO IMAGE and us worked these ideas into a fairly traditional musical accompaniment rather than just using found or diagetic sounds. There are moments in the performance which are quite joyous and whimsical at times, and for those simple piano and ringtone compositions were created. These sonic layers are distinct, and haphazardly slide around atop one another, following the narrative cues given by the accidental emotions of the AI protagonist.

What part or programme of the IMPAKT Festival are you looking forward to most?

We are really curious about the film work being shown as part of IMPAKT’s exhibition – the film Emily by Danica Dakić in particular, as it explores the relationship between the process of learning human language and the potential for self-expression. Our current work also focuses on the limitations of natural language processing for AI and aims to present AI as less “a mind that consists exclusively of congealed thoughts” – in the words of Federico Campagna – and more a mind capable of self-perception and self-expression.

Website by HOAX Amsterdam