RIXA - Explaining Artificial Intelligence in Natural Language
Becker, Maximilian
1; Schwall, Finn 2; Vishwesh, Vishwesh 2; Wu, Sihan 1; Birnstill, P. 2; Beyerer, Jürgen 1,2
1 Institut für Anthropomatik und Robotik (IAR), Karlsruher Institut für Technologie (KIT)
2 Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB)
1; Schwall, Finn 2; Vishwesh, Vishwesh 2; Wu, Sihan 1; Birnstill, P. 2; Beyerer, Jürgen 1,21 Institut für Anthropomatik und Robotik (IAR), Karlsruher Institut für Technologie (KIT)
2 Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB)
Abstract:
Natural language is the instinctive form of communication humans use among each other. Recently large language
models have drastically improved and made natural language
interfaces viable for all kinds of applications. We argue that
the use of natural language is a great tool to make explainable
artificial intelligence (XAI) accessible to end users. We present
our concept and work in progress implementation of a new kind
of XAI dashboard that uses a natural language chat. We specify 5
design goals for the dashboard and show the current state of our
implementation. The natural language chat is the main form of
interaction for our new dashboard. Through it the user should be
able to control all important aspects of our dashboard. We also
define success metrics we want to use to evaluate our work. Most
importantly we want to conduct user studies because we deem
them to be the best method of evaluation for end-user-centered
applications
models have drastically improved and made natural language
interfaces viable for all kinds of applications. We argue that
the use of natural language is a great tool to make explainable
artificial intelligence (XAI) accessible to end users. We present
our concept and work in progress implementation of a new kind
of XAI dashboard that uses a natural language chat. We specify 5
design goals for the dashboard and show the current state of our
implementation. The natural language chat is the main form of
interaction for our new dashboard. Through it the user should be
able to control all important aspects of our dashboard. We also
define success metrics we want to use to evaluate our work. Most
importantly we want to conduct user studies because we deem
them to be the best method of evaluation for end-user-centered
applications
| Zugehörige Institution(en) am KIT | Institut für Anthropomatik und Robotik (IAR) |
| Publikationstyp | Forschungsbericht/Preprint |
| Publikationsjahr | 2023 |
| Sprache | Englisch |
| Identifikator | KITopen-ID: 1000167428 |
| HGF-Programm | 46.23.04 (POF IV, LK 01) Engineering Security for Production Systems |
| Bemerkung zur Veröffentlichung | IEEE International Conference on Data Mining (ICDM), Shanghai, 1-4 Dezember 2023 |
| Relationen in KITopen |