Experimental Design
Erik Lintunen
An artist and researcher (and secretly an aspiring mathematician) whose work is grounded in critical theory, new media and philosophy. I like to learn new things and design experiences to communicate complex ideas.
I’ve developed and exhibited a range of collaborative works across spaces, places and frequencies: Barbican Centre, Central Saint Martins, Chelsea College of Arts, HOME Manchester, IKLECTIK, London College of Communication, RA Lates, Resonance FM/Extra, Royal College of Art, Somerset House, SPILL Festival of Performance, fringe events internationally, as well as online. I teach creative coding, dabble in theatre, sound and radio, and occasionally perform.
In 2004, I was the recipient of the Clitheroe 2000 Young Achiever Award. In 2018, I was awarded residencies as part of Barbican HOME Open Lab; my work was commissioned for Resonance’s Strands supported by Jerwood Charitable Foundation; and I was selected as a finalist for the WIRED Japan Creative Hack Award.
I’m interested in AI and data practices, specifically the material-semiotic instrumentation through which machine learning systems operate.
As the world is becoming increasingly reliant on sophisticated ICTs, the inherent perplexity of knowledge representation and power is more apparent than ever (i.e. sociopolitical implications of personalised feeds). However, the operational aspects of these systems are often opaque and discreet.
There is incredibly fascinating work happening at the intersection of computer science, interaction design and art in visualising neural networks and in devising tools to interface with their internal processes (e.g. distill.pub). This field of research has, somewhat unexpectedly, rekindled my appreciation for mathematics and moving forward I see my practice bridging the aforementioned with information experience design.
Temple of Apollo
Apotheosis
Most machine learning systems may be broken down into three main components: data representation (how information is fed into the network), objective function (a way to represent the problem) and optimisation method (the means to achieve our objective). Bias, in this context, is often associated with its negative connotations (e.g. racial bias) or in a more technical sense with underfitting (i.e. failure to capture underlying patterns in the data). The mainstream conversation rarely addresses (inductive) bias for its utility, as the double-edged sword: assumptions help learning algorithms distinguish signal from noise (further reading: priors). So how should we frame the conversation around bias in AI to constructively discuss its real-world applications and their implications?
Developed in response to conversations with various experts and nonexperts about interpretability in AI, the core idea behind the featured stories is to consider bias in the context of knowledge production, because, without specificity it is merely reduced to rhetoric—a scapegoat for a more expansive problem.
Medium:
code in cloudSize:
VW x VH
The fault, dear Brutus, is not in our stars,
But in ourselves, that we are underlings.
Cassius, Julius Caesar (Shakespeare)
My starting point was to design an experience based on the written texts. I built several prototypes utilising various platforms, for testing ways in which I could approach the construction of an interactive system. I wanted to develop something to tell the stories in an engaging and playful way—and considering the historical legacy of prediction and forecasting technologies and their connections to deities and prophecies, the idea to fit the writing within a fictional framework and to create an oracle narrator character seemed fitting.
Using custom-built bots and role-menus on Discord (messaging application), I devised an immersive and fragmented narrative, in the sense that the experience was choice-based with certain channels only visible having chosen specific options. A key part of my research was also to determine the system that would be used to develop the AI. For the chatbot, I tested a state-of-art GPT-2 language model, which showed promise after some fine-tuning; but ultimately, I decided that it would be more interesting to explore the possibilities of using the bot as an interface for accessing the narratives and, therefore, required something more easily controllable—something that would reliably provide meaningful information to the user.
As part of my research into ways in which to create an interactive narrative, I devised plans for a terminal-based multi-user dungeon. The core idea was that users could connect to a shared virtual machine in the cloud and browse through media through a restricted terminal environment. However, in the end it was best to include a graphical user interface and due to the flexibility building my own server would provide in bringing the text to life, I decided to pursue a more configurable system using JavaScript and Python. Moreover, though some technical details might have been trivial to implement using existing software (e.g. a game engine or UI library), I believe in the value of breaking out of existing frameworks and challenging the status quo. And the learning process has been a lot of fun!