[Paper Review] Human-AI Collaboration in Data Science: Exploring Data Scientists' Perceptions of Automated AI

Dakuo Wang, Justin D. Weisz+7 more|arXiv (Cornell University)|Sep 5, 2019

Big Data and Business Intelligence58 references81 citations

TL;DR

The paper explores how data scientists perceive AutoAI and its role in future data science practice through semi-structured interviews with 20 IBM data scientists.

ABSTRACT

The rapid advancement of artificial intelligence (AI) is changing our lives in many ways. One application domain is data science. New techniques in automating the creation of AI, known as AutoAI or AutoML, aim to automate the work practices of data scientists. AutoAI systems are capable of autonomously ingesting and pre-processing data, engineering new features, and creating and scoring models based on a target objectives (e.g. accuracy or run-time efficiency). Though not yet widely adopted, we are interested in understanding how AutoAI will impact the practice of data science. We conducted interviews with 20 data scientists who work at a large, multinational technology company and practice data science in various business settings. Our goal is to understand their current work practices and how these practices might change with AutoAI. Reactions were mixed: while informants expressed concerns about the trend of automating their jobs, they also strongly felt it was inevitable. Despite these concerns, they remained optimistic about their future job security due to a view that the future of data science work will be a collaboration between humans and AI systems, in which both automation and human expertise are indispensable.

Motivation & Objective

• Understand current data science work practices in enterprise settings and how AutoAI could fit into or change them.
• Explore data scientists’ attitudes toward AutoAI, including perceived benefits, drawbacks, trust, and impact on job roles.
• Characterize patterns of human-AI collaboration in data science and how AutoAI could function as a collaborator.
• Assess whether AutoAI is seen as a tool or a first-class partner in the data science workflow.

Proposed method

• Conducted semi-structured interviews with 20 data scientists at IBM who actively practice data science in diverse business contexts.
• Shown an AutoAI demo (UI with progress, pipeline visualization, and leaderboard) to participants to ground their understanding.
• Performed open coding on ~80 pages of interview notes and ~400 pages of transcripts to identify themes.
• Described participants’ work practices as a ‘scatter-gather’ cycle of individual analysis and collaborative idea generation.
• Reviewed related HCI/CSCW literature to position AutoAI as a potential first-class collaborator rather than a mere toolbox addition.

Experimental results

Research questions

• RQ1 How do data scientists currently practice data science in enterprise settings, and where could AutoAI fit in?
• RQ2 What are data scientists’ perceptions of AutoAI's benefits, drawbacks, and trust, and how might AutoAI alter collaboration and workflows?
• RQ3 In what ways could AutoAI support or hinder the data science process across data preparation, modeling, and deployment?
• RQ4 How do domain knowledge and cross-disciplinary collaboration influence the integration of AutoAI into data science work?
• RQ5 What future roles might human data scientists and AutoAI assume in a collaborative workflow?

Key findings

• Data scientists report mixed feelings toward AutoAI, with concerns about losing depth but optimism about collaboration and efficiency gains.
• AutoAI is viewed as an inevitable future of data science, potentially transforming workflows and job roles.
• A scatter-gather collaboration pattern emerges, where AutoAI can enhance individual productivity during scatter phases and support team deliberations during gather phases.
• AutoAI is characterized as a potential first-class collaborator capable of cross-team coordination and transparency, not just a tool.
• Data scientists see AutoAI as complementary to human expertise, while managers are attracted to potential cost savings from automation.
• Concerns include the risk of deploying lower-quality models if automation reduces interpretability and human oversight.