Systemic Oversimplification Limits the Potential for Human-AI Partnership

Abstract

The modern world is evolving rapidly, especially with respect to the development and proliferation of increasingly intelligent, artificial intelligence (AI) and AI-related technologies. Nevertheless, in many ways, what this class of technologies has offered as return on investment remains less impressive than what has been promised. In the present paper, we argue that the continued failure to realize the potential in modern AI and AI-related technologies is largely attributable to the oversimplified, yet pervasive ways that our global society treats the relationship between these technologies and humans. Oversimplified concepts, once conveyed, tend to perpetuate myths that in turn limit the impact of such technologies in human society. To counter these oversimplifications, we offer a theoretical construct, which we call the landscape of human-AI partnership. This construct characterizes individual capability for real-world task performance as a dynamic function of information certainty, available time to respond, and task complexity. With this, our goal is to encourage more nuanced discourse about novel ways to solve challenges to modern and future sociotechnical societies, but without defaulting to notions that remain rooted in today’s technologies-as-tools ways of thinking. The core of our argument is that society at large must recognize that intelligent technologies are evolving well beyond being mere tools for human use and are instead becoming capable of operating as interdependent teammates. This means that how we think about interactions between humans and AI needs to go beyond a “Human–or–AI” conversation about task assignments to more contextualized “Human–and–AI” way of thinking about how best to capitalize on the strengths hidden within emergent capabilities of unique human-AI partnerships that have yet to be fully realized.

Keywords

• Human-AI partnership
• human-autonomy teaming
• sociotechnical systems
• AI ecosystems
• function allocation
• task complexity