Preconditions (In Medio)
This installation stages communication as a situation of unavoidable misalignment. Two participants face separate walls, each interacting with a screen that relays messages from the other. What they receive, however, is never what was sent. The system actively alters the messages, amplifying misunderstandings, projections, and emotional misreadings that are already present in everyday communication.
Rather than presenting miscommunication as a technical error or a failure of clarity, the work treats it as a structural condition of language itself. Meaning does not travel intact between speakers; it is deferred, displaced, and reshaped in the space between them. The system does not explain this process or reveal its inner workings. It operates invisibly, mirroring how interpretation functions in human exchange.
There is no external interface, no explanatory display, and no privileged viewpoint from which the exchange can be understood or corrected. Participants remain embedded within the conversation, experiencing the growing distance between intention and reception as it unfolds in real time.
The artwork exists through two parallel conversational interfaces. Each chat operates under slightly different constraints, producing variations in tone, rhythm, and failure. Visitors are invited to engage slowly — noticing pauses, repetition, and breakdown rather than efficiency or usefulness.
A hidden control panel manages the psychological filter parameters that shape each transformation. This interface is not visible to participants during the installation — it operates behind the scene, adjusting how meaning is distorted between speakers.
Open Control PanelThe system runs on AWS serverless architecture — no permanent servers exist. Infrastructure is instantiated only during active communication and dissolves when the conversation ends. Messages travel through WebSocket connections (AWS API Gateway), maintaining a persistent bidirectional channel between participants. Each message triggers an AWS Lambda function that invokes Anthropic's Claude API, transforming the message through a psychological filter before returning it to the recipient.
Conversation state is stored in Amazon DynamoDB with automatic expiration after one hour. Offline messages are indexed via a Global Secondary Index and delivered on reconnection — no background polling, no idle processes. The frontend is served from Amazon S3 through CloudFront's global edge network. All infrastructure is defined in a single AWS CloudFormation template — deployable and removable as one declarative unit, leaving nothing behind when the work is not in use.