Project

Back to overview

From Tools to Teammates: Human-AI Teaming Success Factors in High-risk Industries

English title From Tools to Teammates: Human-AI Teaming Success Factors in High-risk Industries
Applicant Bienefeld Nadine
Number 187331
Funding scheme NRP 77 Digital Transformation
Research institution Arbeits- und Organisationspsychologie D-MTEC ETH Zürich
Institution of higher education ETH Zurich - ETHZ
Main discipline Applied psychology
Start/End 01.09.2020 - 31.08.2024
Approved amount 544'714.00
Show all

All Disciplines (3)

Discipline
Applied psychology
Information Technology
Internal Medicine

Keywords (4)

Team mental models; Human-AI teams; Adaptive coordination; Artificial intelligence and machine learning

Lay Summary (German)

Lead
Immer mehr Teams nutzen KI zu Entscheidungsfindungs- und Automatisierungszwecken. Was denken Teammitglieder über diese digitalen «Kollegen»? Wie funktioniert die Zusammenarbeit mit KI und wie kann man diese optimieren damit die richtigen Entscheide getroffen und Teamziele erreicht werden können?
Lay summary
Der Einsatz von KIs im Teamkontext ist noch weitgehend unerforscht. Aus der Teamliteratur wissen wir, dass geteilte mentale Modelle (ein Verständnis über was von wem, wann, wie gemacht wird) essentiell sind für Entscheidungsfindung, Koordination und Zielerreichung in Teams. Ob und wie sich diese mentalen Modelle auch auf KIs beziehen und was dies für den Teamerfolg bedeutet ist noch unklar; insbesondere aufgrund der bestehenden Komplexität und «black-box «Problematik selbst-lernender KIs. Diesen Fragen gehen wir anhand von vier Studien in realen Teams aus Medizin und Aviatik nach. In Studie 1 untersuchen wir mentale Modelle, Teamprozesse sowie Teamleistung in medizinischen Teams, die bereits intensiv mit KI zusammenarbeiten und ziehen Vergleiche zu herkömmlichen Teams. In Studie 2 übertragen wir die Resultate aus Studie 1 und nutzen sie in der Entwicklung einer neuen KI für medizinische Teams. In Studien 3a (medizinische Teams) und 3b (cockpit crews) erarbeiten und testen wir ein «Interventionstraining zur Optimierung von Teamarbeit mit KI» und leiten daraus ein allgemein gültiges «online-tool zur Optimierung von Teamarbeit mit KI» ab, das von Teams aus sämtlichen Industrien genutzt werden kann.
Direct link to Lay Summary Last update: 10.02.2020

Responsible applicant and co-applicants

Employees

Collaboration

Group / person Country
Types of collaboration
University Hospital Zurich (USZ) Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
- Industry/business/other use-inspired collaboration
NASA Ames Human Systems Integration Division United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
International Emergency Medicine Conference SGNOR-SSMUS Talk given at a conference «Shall I trust him?” How medical providers cooperate with their artificial teammates. 27.05.2021 Bern, Switzerland Bienefeld Nadine;


Knowledge transfer events

Active participation

Title Type of contribution Date Place Persons involved
Grand Round series Workshop 01.11.2020 Universitätsspital Zürich, Switzerland Bienefeld Nadine;


Communication with the public

Communication Title Media Place Year
New media (web, blogs, podcasts, news feeds etc.) Increasing uncertainty to innovate ETH podcast International 2021
New media (web, blogs, podcasts, news feeds etc.) Who do I trust? The human or the machine? NRP77 Digital Transformation Newsletter Rhaeto-Romanic Switzerland German-speaking Switzerland Western Switzerland Italian-speaking Switzerland 2021

Associated projects

Number Title Start Funding scheme
183010 Welcome to the team Dr. Watson: New Roles and Responsibilities in AI-Supported Healthcare Teams 01.09.2018 Digital Lives
167195 ICU-Cockpit: IT platform for multimodal patient monitoring and therapy support in intensive care and emergency medicine 01.02.2017 NRP 75 Big Data

Abstract

Given the rapid deployment of AI in various work contexts, increasing numbers of professionals will work in human-AI teams. The collaboration in these types of teams is different from ordinary teams that use other kinds of technologies because self-learning AI applications initiate actions, make decisions, and thus turn from tools to teammates. These AI-teammates, however, still have limited capabilities to communicate and coordinate in dynamic team contexts. Humans, therefore, need adequate mental models of AI to correctly interpret, adapt, and perform well in unison with AI. In study 1, we examine human-AI teams in a sample of doctors and nurses to define (A) what are people’s mental models about AI?; (B) how do shared mental models in human-AI teams differ from those in human-only teams and what is the impact on performance? and (C) how can human-AI teams adapt and coordinate well in dynamic contexts in response to high versus low AI reliability?. In study 2, we apply these findings to the interface design of the ICU cockpit, an AI-application aiding doctors and nurses in their decision-making and will facilitate its adoption into clinical practice. In study 3, we develop a training intervention to help human-AI teams face the challenges outlined above and test its efficacy in a sample of doctors and nurses as well as airline pilots. Finally, we develop “Team Mental Modeler”, an app-based tool that can be used by human-AI (and ordinary) teams from various industries to assess their shared mental models for learning and debriefing purposes. We will widely disseminate our findings within the scientific and practice-oriented target communities as well as with the general public.
-