Making AI applications comprehensible is important, in particular when used for decision-making. This is where the AI application planned by Erfurt University of Applied Sciences and ADICOM Software KG comes in. Until now, AI applications have only been able to form hypotheses, but not explain them. With the aid of speech synthesis, the project team wants to make it possible for the hypothetical data relationships of AI to be explained comprehensibly and transparently.
Hyper KI - comprehensible hypothesis-forming AI
Why are you a strong team?
We love our work, get along brilliantly and complement each other with artificial intelligence skills from science and business. Our team consists of an SME distinguished as one of the best IT service providers in Germany, an internationally renowned chief scientific officer with a fountain of innovative ideas and a university professor who shares her fascination of artificial intelligence with students, carries out practical method research and has knowledge of many concepts as a consultant.
Explain your idea in three sentences.
In order to explain its hypotheses and thus enable comprehensible human-AI collaboration, AI learning systems need language. For this, we use the ‘identification by enumeration’ method (Gold, 1967) as a basis and design hypothesis spaces (HS) that can change during the learning process. The aim is now to make AI capable of explaining how it finds hypotheses, what further information it needs and why a current HS needs to be modified during learning.
What makes your idea special?
The idea is based on fundamental theories, considers innovative trends and combines approaches from various scientific disciplines in a new way. It requires thinking ahead and applying symbolic AI, inductive inference and different language modelling. It uses ‘stop words’ as semantic signposts in sentences and natural language communication. It sees AI as the designer of the learning process and explicitly considers negative examples.
What are the next steps?
The experimental environment we have just created in Python helps to further develop the ‘identification by enumeration’ process for language learning and research HS. The results of these experiments lead us to think in new ways with far-reaching insights and a refined project concept. The next steps are the systematic evaluation of the findings, focussed experiments on specific research issues and the targeted expansion of the experimental environment.
Contact
- Contact person: Prof Dr-Ing Oksana Arnold
- E-Mail: oksana.arnold@fh-erfurt.de
- Website: www.ai.fh-erfurt.de/arnold
The Policy Lab of the Federal Ministry of Labour and Social Affairs