TLDR: A research paper introduces ‘8bit-GPT,’ a chatbot simulated on a legacy Macintosh operating system. By intentionally creating an inefficient and defamiliarized interaction, the project aims to make users reflect on their over-reliance on modern AI, the nature of human-AI interaction, and the impact of anthropomorphic language. A user study confirmed that this ‘slow technology’ approach successfully prompted critical reflection on technology’s embeddedness and anthropomorphism.
In an era dominated by sleek, hyper-efficient artificial intelligence, a new research project called 8bit-GPT takes a deliberate step back in time, simulating a language model on an obsolete Macintosh Operating System. This innovative work aims to challenge our growing over-reliance on modern chatbots and encourage deeper reflection on the nature of human-AI interaction.
The proliferation of assistive chatbots has led to widespread dependence for decision-making and information-seeking. While convenient, this reliance has shown adverse effects on information retention and can foster superficial emotional attachments. 8bit-GPT, developed by Hala Sheta from the University of Waterloo, seeks to counteract these trends by introducing friction and defamiliarization into the chatbot experience.
Designing for Reflection, Not Efficiency
The core philosophy behind 8bit-GPT draws on reflective design principles such as ‘slow technology’ and ‘counterfunctionality’. Instead of prioritizing speed and seamlessness, the project intentionally creates an inefficient and somewhat awkward interaction. The goal is to foreground the AI as a tool, making its presence undeniable and prompting users to think critically about their engagement.
Slow technology emphasizes designing for reflection and mindfulness over sheer performance. By amplifying the temporal aspects of interaction, often invisible in modern tech, 8bit-GPT makes the user aware of the tool’s form and function. Counterfunctional design extends this by subverting familiar functionalities, creating artifacts that are ‘strangely familiar’. For 8bit-GPT, this means inhibiting efficient interaction and unlimited context windows, while adding a nostalgic interface and a distinctive, often ‘silly’ manner of speech.
Bringing a Legacy System to Life
The technical setup for 8bit-GPT is as unique as its concept. The language model is simulated on a legacy Macintosh Operating System, specifically System 7.5, using the Basilisk II open-source emulator. This emulator was chosen for its support of internet access and local file-sharing, crucial for the communication between the AI and the user interface.
The physical artwork installation itself is a journey into the past. The Mac OS emulator is projected onto the screen of an authentic Macintosh Plus, surrounded by period-appropriate props like its user manual, original keyboard, Apple Mouse II, and labeled floppy disks. Even though System 7.5 supports color, a black and white display is preferred to further emphasize the distance from contemporary technology. The smallest screen dimension (640×480) is also chosen for this effect.
On the backend, a Python script manages the interaction, reading user input from the emulator, formatting requests for an inference server, and writing the model’s output back. The model used is Llama-2-13b-chat, an older, lightweight model fine-tuned to output casual ‘Redditspeak’. A C program on the emulator side handles the user interface, sending single-line inputs and displaying multi-line responses, often with intentional delays or ‘Robot dozed off…’ messages to highlight the system’s presence.
Also Read:
- Bridging the AI Literacy Gap: How Community College Instructors Envision Teaching AI to All Students
- AI ‘Deathbots’ Offer Digital Afterlives, Raising Ethical Questions on Grief and Memory
User Experiences: Friction and Reflection
A user study with 15 participants from the University of Waterloo revealed the success of 8bit-GPT’s objectives. Despite an average System Usability Scale (SUS) index of 57.33, which suggests fair usability, participants generally found the intentional friction successful in prompting reflection.
Key themes emerged from the study:
- Interface Friction: Participants noted the clunky keyboard, small blurry screen, and inefficient system, which prevented them from entering a ‘flow’ state. This constant awareness of the tool led to reflections on technological advancement.
- Conversational Asymmetry: The constraint of single-line inputs versus multi-line, sometimes poorly formatted, outputs caused frustration for some but also encouraged others to be more precise in their phrasing.
- Notion of Place: The bulky, dated setup made users reflect on a time when technology had a physical presence, contrasting it with the invisible, embedded nature of modern devices.
- The Turing Test and Anthropomorphism: Users often tested the AI’s intelligence and sentience, which prompted reflections on the nature of AI and the impact of anthropomorphic language. Many realized they personify AI for linguistic convenience but understood its fundamental mathematical nature.
In conclusion, 8bit-GPT successfully uses defamiliarization and intentional inefficiency to make users critically examine their relationship with AI and technology as a whole. This research offers valuable insights into designing technology that encourages mindfulness and reflection rather than just efficiency. You can find more details about this fascinating project at the research paper.
