Robotreporter research into generative AI systems
Objective
The goal of the project is to build a system that generate news articles. The system will be demonstrated based on stock market data but can potentially be reused on other data. The system will be developed using ‘transparency by design’ principles and an evaluation will take place if people experience it as reliable and transparent.
Results
- A working prototype that is publicly available (www.robotreporter.nl) where people can use generative AI.
- Source code of the system, to allow other researchers to inspect the system and reuse code.
- Analysis results of the perception of the AI system by end users.
Relevance of the project
Since the source code will be published, the project can function as a blueprint for new AI systems and as an example system in education.
"People will trust technology only after they have used this themselves. Therefore we need more open AI systems that are designed with transparency in mind"
Sieuwert van Otterloo, director ICT institute
Duration
01 September 2020 - 01 February 2021
Approach
Existing new articles on stock market data are analysed to determine what observations and phrases are used in existing news articles. Based on these examples, patterns are defined for generating new observations from new data. These patterns are combined in new articles. These articles are then evaluated with end users, so that the algorithms and overall presentation of the system can be evaluated.
HU researchers involved in the research
Collaboration with knowledge partners
ICT Institute in Utrecht, a IT-consultancy office with a lot of practical experience about AI and privacy. Besides, David Paternostro is supervisor of internStan Meyberg who is also involved in this project.
More information
Blog: Robotreporter - transparant use of AI (Dutch)
Website: www.robotreporter.nl
