8. Human-AI Collaboration
Enough With "Human-AI Collaboration"
The key idea of the video is that the field of human-computer interaction should prioritize putting humans first and view AI as a tool rather than a collaborator, in order to avoid exploiting AI laborers and empower users.
Interaction with AI systems is often described as deeper and more profound than with simpler systems, leading to a widespread agentistic turn in HCI research.
The AI industry heavily relies on cheap labor from poor countries in the Global South for data annotation, with workers being asked to review thousands of videos a day at an accelerated speed.
The promise of upward mobility and gaining technical skills in the technology sector for workers from the Global South is often false, as AI-driven tools are being used to automate job skills and replicate the labor of experts without their direct involvement, echoing historical exploitation of working individuals.
AI systems can be built without relying on manually annotated data by using synthetic datasets, learning from playing digital games, or accessing freely available human data, but each approach has limitations.
Human-labelled data is necessary in situations where there is no automatic feedback to accurately describe the desired behavior of a system.
The idea is that AI should be viewed as a tool or instrument, not as a collaborator, as it has unique capabilities that surpass traditional tools like hammers or scalpels.
The metaphor of AI as collaborator was developed to support and empower users, but it can be improved to empower users without disenfranchising AI laborers.
Stop using phrases like "human-AI collaboration" or "human-AI partnership" and instead prioritize putting humans first in the field of human-computer interaction.
Key insights
📚 The number of papers on "human-AI collaboration" has been steadily increasing, indicating a growing interest in this area of research.
💰 The AI industry heavily relies on cheap labor from poor countries in the Global South, where workers are paid less than $30 a week for grueling tasks like data annotation.
🤖 AI-driven tools that automate job skills can be associated with exploitative practices, highlighting the need to address the distancing of knowledge labor through AI.
🌐 Generating synthetic datasets and utilizing freely accessible data from the Internet are promising approaches for building AI systems without relying on large amounts of human data.
📚 There will always be situations where human-labelled data constitutes the best description of the desired behaviour of the system.
🖋️ Language models and image generation models have capabilities that surpass traditional tools, leading to the creation of new terms like "supertool" to describe AI's unique abilities in writing stories, poetry, and generating beautiful images.
🔄 The metaphor of AI as collaborator is a response to the fear of automation, aiming to empower users and create a future where humans and AI work together, building on each other's strengths.
💡 Terminology matters in AI, as it reflects our values and priorities, and putting humans first is essential in the development and use of AI technologies.
Enough With "Human-AI Collaboration" Advait Sarkar CHI 2023: The ACM CHI Conference on Human Factors in Computing Systems Session: AltCHI: H in HCI Describing our interaction with Artificial Intelligence (AI) systems as 'collaboration' is well-intentioned, but flawed. Not only is it misleading, but it also takes away the credit of AI 'labour' from the humans behind it, and erases and obscures an often exploitative arrangement between AI producers and consumers. The AI 'collaboration' metaphor is merely the latest episode in a long history of labour appropriation and credit reassignment that disenfranchises labourers in the Global South. I propose that viewing AI as a tool or an instrument, rather than a collaborator, is more accurate, and ultimately fairer. Web:: https://programs.sigchi.org/chi/2023/... Pre-recorded presentation videos for Alt.CHI at CHI 2023