Museum Chatbots: Historical Examples and Practical Decisions

This article reviews early museum chatbot experiments from a 2020 perspective. The examples are historical; current service availability has not been verified.

Technology has transformed the way we interact with the world around us. Whether we’re shopping online or streaming content through our smartphones, the past two decades have seen a change like nothing humans have ever experienced before. And if this rapid innovation weren’t taking place quickly enough, the Coronavirus pandemic has given institutions and organisations around the globe a shot in the arm that has made digitisation a necessity rather than a choice.

One of the most exciting and controversial aspects of technological advancement is undoubtedly artificial intelligence.

Chatbots use different approaches: some follow scripted rules, some retrieve prepared information, and others generate responses. These approaches should not be treated as equivalent to human understanding.

But nowadays, AI is fast becoming ingrained in the processes, software and systems that run many parts of our daily lives. From helpdesks to How-To guides, the way we communicate with technology is changing.

But what happens when this technology enters the long-established infrastructure of the museum space? A prime example of this is the chatbot.

Different kinds of chatbot

A chatbot is an interface for exchanging messages or spoken responses with an automated service. Its behaviour depends on its design, information sources and limits; a conversational style does not establish the accuracy of an answer.

Some organisations use chatbots for bounded information and support tasks. Whether this reduces staff work or improves service needs to be evaluated for the actual use case.

Text and voice are interface choices, not a complete classification of how a chatbot works. A museum should distinguish the service it needs from the technology used to deliver it.

Early museum experiments

So what does all this have to do with museums? Is it possible for museums to deploy chatbots as part of their customer experience?

Even in their simplest forms, chatbots can be used to provide more information, something many museum visitors welcome. At the National Art Museum in Belarus, chatbots are used simply yet effectively. Through Facebook Messenger, visitors can ask the bot to provide more information on an item they are looking at.

But bots can go further than this. At the Heinz Nixdorf Museums Forum, a conversational agent called Max was introduced. But Max was not a real person; he was a bot, designed to engage with visitors through a screen. In this way, chat bots can be used to created a mascot of sorts for museums — a seemingly real helper ready to ask questions and enhance the customer experience.
One of the main benefits of chatbots is the 24/7 service they provide.

Bots are not necessarily intended to replace human interaction. A service can handle a defined set of routine questions and direct other requests to staff, but it needs an honest description of when and how a person can respond.

However, chatbots aren’t perfect. Although they can provide faster customer service, they often work from a limited database, so their complexity is restricted. The success of chatbots depends on how they are implemented within a business or institution.

Cooper Hewitt Museum

There are many examples of chatbots being put to effective and imaginative use in museum spaces, and one of the earliest comes from the Cooper Hewitt Museum in New York. Considered by some to be a pioneer of chatbot technologies in cultural spaces, the Cooper Hewitt Museum created the Object Phone back in 2013.

This was a service powered by Twilio – a communication service specialising in SMS, video and voice AI technologies. As the name suggests, the Object Phone allowed visitors to text or call to ask for more information on a museum object.

Cooper Hewitt’s Object Phone is a useful historical example because it included a route back to people. In the 2016 account, a question could be passed to staff, whose reply returned through the service. The practical lesson is to design the handoff, including what happens when nobody is available to answer immediately. The 2016 project account describes daily object messages and routing questions through a staff Slack channel.

Anne Frank’s House

Anne Frank’s House is undoubtedly one of the most well-known and sensitive cultural sites in the world, evoking curiosity and intense emotional responses in visitors. So any attempt to integrate chatbot technology into the museum had to be done carefully and with great sensitivity.

On 21 March 2017, the International Day for the Elimination of Racial Discrimination, the Anne Frank House launched a Facebook Messenger bot offering information about Anne Frank, the historical context and visiting the museum. The museum presented it as a way to extend its educational work beyond the physical site.

The 2017 launch announcement described an educational and visitor-information service developed with Facebook Netherlands. That intention does not establish current availability or accuracy for every question.

The House Museums of Milan

In some cases, chatbots are combined with other innovative technologies to create a unique visitor experience. This was the case with the House Museums of Milan – a group of four historical homes in Milan. When the team created an itinerary to encourage people to visit all four homes, they decided to introduce gamification alongside it to attract a younger audience.

This resulted in a chatbot game, delivered through Facebook Messenger. The bot invites younger visitors to explore the four houses and find hidden clues in order to solve a real historical mystery.

Choose a bounded service

Start with a narrow service question. A bot that helps people find opening hours and access information has different risks from one interpreting a collection or speaking as a historical person. Define what it should answer and where the approved information comes from. See also deciding whether a new format fits the audience.

Test answers and handoff

Test likely misunderstandings and questions outside the intended scope. Make it clear that the interface is automated, give people another way to find essential information and assign responsibility for updating its sources. Availability at all hours is useful only if the service remains understandable and dependable. An interpretive character should be clearly distinguished from an authoritative factual answer.

Maintain sources and protect data

Set responsibilities for approved information, review dates, access to conversation records and retention. Explain what information is collected and provide a realistic route to human contact. Read about responsible handling of conversation data.