Fostering effective hybrid human-LLM reasoning and decision making

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction

Background:

• Modern Large Language Models (LLMs) demonstrate impressive capabilities but also exhibit non-trivial errors.
• Despite significant research into LLM limitations, the dynamics and implications of human-LLM collaboration are underexplored.

Purpose of the Study:

• To argue for prioritizing research on human-LLM interaction.
• To examine biases hindering effective human-machine collaboration.
• To discuss goals for enhancing human-LLM reasoning and decision-making.

Main Methods:

• Perspective piece examining existing literature and proposing future research directions.
• Analysis of biases impacting human-LLM collaboration.
• Discussion of potential solutions and future research goals.

Main Results:

• Current research inadequately addresses the potentials and risks of human-LLM collaboration.
• Biases can impede effective collaboration between humans and LLMs.

Conclusions:

• Future LLM research should prioritize enhancing human-LLM interaction.
• Achieving mutual understanding and complementary team performance are key goals for effective human-LLM reasoning.