November 29, 2021
Fraunhofer Heinrich Hertz Institute’s (HHI) explainable AI (XAI) demo is part of the exhibition “Artificial Intelligence. Machine Learning Human Dreams ” that recently opened at the “Deutsches Hygiene-Museum Dresden” (German Hygiene Museum Dresden). The institute's “Visual Question Answering (VQA) Demo” visualizes how decision-making processes take place within neural networks. For this purpose, the demo uses a so-called heatmap. The heatmap highlights which pixels in an image or words in a text played the decisive role in the AI’s decision-making process.
The exhibition will run until August 28, 2022. It provides an overview of the areas that already use AI today and reflects on the influence AI technology has on our lives. Fraunhofer HHI's VQA demo explains the black box problem of AI applications to visitors. At the same time, it offers a suitable solution using XAI technology developed at the institute.
The black box is one of the biggest challenges in AI research. The term black box describes that an AI’s decision-making process takes place in neuronal networks and is invisible to us. However, in order to deploy AI applications in safety-critical areas such as transportation or healthcare, all decisions must be traceable.
The VQA demo illustrates how XAI works using a method for image and text recognition. A trained AI solves an image recognition task. For this purpose, users ask a question to the AI about a selected image. Besides answering the question, the AI also marks the areas in the image as well as the words in the question that were crucial to its decision-making using a so-called heatmap. These markings allow viewers to easily identify which parts of the image and question led to the AI's decision. In addition, the demo indicates how certain or probable it regards the answer.