Chatbot instead of museum guide | heise online

So far, chatbots have mainly been used by companies to answer customer inquiries or to manage complaints. But since the speech generators are now also based on methods of artificial intelligence, further fields of application are opening up. Chim is currently being trained as a dialogue-oriented companion for visits to the Städel Museum in Frankfurt.

“Who was the young woman in the painting? Did the painter know her personally? Were the hairstyles of the women of his time really that elaborate?” These are questions that Chim is fed as part of a research project. The Städel Museum acts as an educational partner and offers the project manager Linon Medien KG the environment in which the chatbot is prepared for its later use. Linon has specialized in supporting museums in the development of technical applications and acquired the research project from the Federal Ministry of Science and Research. As a project partner, the agency has that German Research Center for Artificial Intelligence (DFKI) brought on board.

The virtual assistant should learn to understand and answer the questions that come to mind when looking at the paintings. That is new. So far, digital museum guides have tended to unwind the knowledge of art experts. On a specially set up public website Anyone can currently view 14 selected works and enter their questions. “We have currently received around 1,800 useful questions that can be used,” says Oliver Gustke from Linon. In addition, this chatbot receives information from the audio guides and catalogs of the museum as well as from scientific treatises on the works for its learning process.

More from Technology Review

More from Technology Review

The Städel Museum has the advantage that the digital collection is already well tagged. According to Stefan Schaffer from DFKI, the chatbot automatically generates answers to questions from the available data, which in addition to the title, artist, material and dimensions contain extensive information on the history of the object. He uses annotated content that has been tagged and categorized. “With the closed questions, it will probably work quite well,” suspects Schaffer. It becomes more difficult with open questions, such as “Why is the dog yellow?”

The Center for Art and Media in Karlsruhe (ZKM) also wants to offer a digital companion in the first half of 2021. This is also the result of a two-year research and development project. This was financed by the “Digital Paths to the Museum” program run by the state of Baden-Württemberg. The ZKM chatbot will initially only provide answers to questions in the sense of FAQs, i.e. information on current exhibitions, opening times and basic information on individual works of art.

He should offer his services on the website, on his own smartphone and on steles in the exhibition rooms. Dominika Szope from the ZKM dampens expectations that are too high: “We underestimated the project in terms of time and personnel capacities,” she says, “but this is our child that we are going to raise now.” Their vision: in four to five years, the chatbot should be able to discuss media art.

The ZKM has for the development of its own machine for the visitor dialogue 2018 worldwide over 20 such digital companions found. The first one dates back to 2004 and still leads through the Heinz Nixdorf Museum Forum in Paderborn. There are also such applications in the Museo de Arte Moderno de Buenos Aires or in the Brooklyn Museum in New York City. But their conversation skills are developed to very different degrees and only a few from the list use artificial intelligence methods. It usually remains with service information, with audio guides that are played with the exhibits or augmented reality extensions that can be called up in the exhibition rooms.

Messenger services such as Facebook or WhatsApp are often used for interaction, or they are developments based on the Watson AI platform from IBM. In order to avoid any problems with data protection, the ZKM and the DFKI work together with a Berlin provider for open source solutions.

The ZKM also wants to run its smartphone variant via the messenger signal. Schaffer from DFKI emphasizes that DFKI’s own platform is used for programming and that the data with which the chatbot learns is on servers hosted in Germany. Chim should run via an app that will be tested as a prototype in the field from autumn 2021, and then be able to write and speak.

It remains to be seen whether the visitor dialogue in the Städel Museum in autumn will work out as the project partners would like. Chim should not only provide answers, but also ask questions and draw the visitors’ attention to one or the other detail in a very talkative way. During the field test, visitors are also asked about their experiences with Chim. “That definitely needs to be improved,” says Schaffer, who is currently not assuming that the chatbot will continue to offer its services in the Städel directly after the project ends in January 2022.


To home page