3 Lessons learnt from Building our first Museum Chatbot… 15 years ago!
By Stefania Boiano and Giuliano Gaia
Once upon a time, in 2002, when mobile phones were like this, we were working at the Digital Department of the Science Museum “Leonardo da Vinci” of Milan. Inspired by the place, we tried to bring Leonardo da Vinci back to life. Not with Dr. Frankenstein’s techniques, but by programming a virtual character able to chat online with the Museums website users as if “he” was Leonardo. Or at least, this was our hope.
We teamed up with a technical partner, a software agency specialized in Natural Language Interfaces, and put a couple of brilliant university students at work building the “Leonardo” knowledge base of the chatbot.
6 months later, the chatbot was ready to be published, and our enthusiasm was at *star level*. We even wrote a paper for the US conference “Museums and the Web” (which was later selected for their print proceedings). In that paper we outlined a series of future steps for museum chatbots, the last one being the integration of different chatbots so that a chatbot can “pass” a user question to another chatbot to get a better answer — something similar to what Facebook announced in the last F8 conference.
Problem is, despite our enthusiasm, our chatbot project was a failure. It was just too easy to fool; we aimed at building a genius like Leonardo and we ended up publishing something slightly more conversational than Lurch of the Addams Family. Users quickly lost interest in chatting with something really not up to the task
In fact we learnt three important lessons.
- Bring the focus on the task, not on the conversation. Our initial message was: “chat with our virtual character”. This was a mistake, because it brought the user attention on the weakest part of the chatbot: its (poor) conversational abilities. If we had said instead “solve this quiz with the help of our character”, the user attention would have been focused on the task and he/she would have been more forgiving with our chatbot’s faults.
- Don’t try to mimick a real person. By saying “Chat with Leonardo da Vinci”, we were setting user expectations impossibly high, while if you create a brand new character, he/she/it can have faults and idiosincracies that the user can forgive.
- Usage context is essential. Back in 2002, mobile chatting was not ubiquitous unless you used expensive SMSs. So the user was forced to chat while sitting at a desktop computer, and this made the interaction with the chatbot much less interesting. Today, in the smartphone/messenger era, the user can chat with your chatbot while at the same time exploring the galleries, opening a much richer range of possibilities.
We kept all of these lessons in mind in the last months, when we built our chatbot game for the House Museums of Milan, with a much more successful outcome. We will write more about the project in future articles, in the meantime we’d love to hear from you. Have you had any similar experience with chatbots, while producing or using them? Which were your mistakes? What did you learn?