Creative Review | Rachael Steven

We are now firmly in the era of the chatbot. Retailers from H&M to Lidl have used bots to market new collections and choose from a wide range of products (in Lidl’s case, its wine range). Channel 4 launched a witty bot to promote AI drama Humans and beauty brands from Sephora to Smashbox have built bots to help customers find makeup to suit their complexion. Taco Bell has even set up a bot that allows customers to order tacos through Slack.

Museums have also been experimenting with AI: the Cooper Hewitt Museum in the US was one of the first to do so with Object Phone – a service that allowed users to ‘call’ or ‘text’ objects in its collection and receive a response. Last year, the Anne Frank House in Amsterdam launched a chatbot to provide information about Anne Frank and World War Two.

Most branded bots offer fairly limited interaction – customers are asked a series of multiple choice questions and shown products based on their selection – and success has been mixed. Facebook announced it would be “refocusing” its AI efforts after it was revealed that its Messenger bots had a 70% failure rate (meaning they could only answer 30% of queries correctly without human intervention), leading some brands to pull out of using Messenger bots altogether.

But as the technology continues to develop – and bots get to grips with the intricacies of human speech – there is potential for them to be used in much more exciting ways. For museums that could mean educating audiences about historical movements or great artists as well as answering common visitor queries.

The Cartier Foundation for Contemporary Art (or Fondation Cartier as it is known in France) is the latest cultural institution to design a chatbot, created using Google’s Dialogflow technology.

The chatbot exists on the Fondation Cartier website and will soon be launching on Facebook Messenger. It can answer questions about opening times and facilities, provide information about artists and artworks in the Fondation’s collection and suggest exhibitions and events for people to visit.

The chatbot was developed by BASE Design and is just one part of a much wider project. BASE has spent over a year working with the Fondation to create a new website that better showcases its vast collection of art (it is one of the biggest cultural institutions in Paris, with around 1500 artworks by 350 artists) and makes it easier for visitors to explore its content.

The new website includes integrated Deezer and YouTube modules – allowing visitors to play audio clips and videos on the main website. BASE also created a new information architecture and structure to aid navigation and help communications teams produce richer content such as themed newsletters. The new website features a custom typeface and, with its ample white space, is designed to reflect the Fondation’s architecture (a vast glass-walled building designed by architect Jean Nouvel). You can read a detailed post by BASE on the making and design of the new website here.

Jacques Letesson, Managing Director at Base, describes the chatbot at “the cherry on top of the cake” (the cake being the Fondation website).

With the Fondation keen to reach a wider audience – particularly as it now hosts events abroad in cities as far as Shanghai – it was looking for a new way to engage with audiences online, particularly those who might not be able to visit the physical site in Paris.

The Fondation has almost 500,000 followers on Facebook – a substantial following for a Parisien cultural venue – so engaging with people on the social media platform was also a key objective. “But we wanted to avoid the typical way of connecting social media and a website – usually, people try to create a bridge between [the two], but we wanted people to directly access the Fondation’s content from their social platform,” says Letesson.

BASE started out researching various bots and different AI technologies before opting to use Dialogflow. (It is also using Facebook’s technology to create the soon-to-launch Messenger bot).

Fondation Cartier’s new websiteFondation Cartier’s new website

“[Building a chatbot] was a completely new process for us,” he says. “We did a lot of research to discover how corporate companies were using chatbots but we found that what most of them were doing was very common – it was ordering shoes and ordering pizzas which is very basic. 

“When you’re building a bot for a cultural institution you need to think about what people might query. You need to think ‘what is the offer of the Fondation? What is the user expecting?’ It was a bit tricky to make sense of that. We read a lot of books, we read articles, we tried different platforms – one by Google and another open source technology – and in the end, we went for Google’s.”

BASE then decided on a function for the chatbot – to provide information about artists and events as well as answering visitor queries. The next step was to design conversations – working out what questions people might ask and how they might ask them and plotting how conversations would flow. This was no easy feat – particularly as the bot had to function in both French and English.

“Designing a conversation is far more difficult than designing a web page or even coding a web page because it’s about human interaction and framing human interaction within a specific context,” says Letesson.

The bot sits on the homepage of the website and will soon be launching on Facebook Messenger

The team experimented with the bot’s tone of voice – introducing humour and wit and varying degrees of ‘human-ness’ – and in the end settled for something that is polite but still machine-like.

“If you humanise a bot too much, people try to interact with it as they would a human being…. If it’s a chatbot people should know it’s a chatbot – it shouldn’t try to be a human,” says Letesson. “At the same time, if you don’t frame the conversation and say to the user, ‘I’m sorry, I can’t help you with that now because I’m still learning’ then people will always try and find those dark spots. There is a balance that needs to be struck,” he explains. “We also had to be sure we weren’t overdoing it .. it’s a difficult thing to get right and we had to remember that this is a prestigious French cultural institution.”

BASE then had to work closely with Fondation Cartier to define ‘entities’ – in other words, the ‘things’ that users might ask about. “When you say ‘I would like two pizzas’, a chatbot knows that pizza is the entity and two is how many you want. So we had to define a number of entities – such as exhibition names, artist names, events, brochure, books, and so on – and at the same time work out what the user would like to know about those things.” Entities had to be correctly tagged and categorised on the Fondation’s website to enable the chatbot to pull up relevant content for users. With thousands of pieces of content, this was a complex and time consuming process.

The bot can pull up information about exhibitions and events as well as artworks from the Fondation’s collection

One of the hardest aspects of designing a chatbot is trying to predict the type of things people might ask. Museums and galleries have a diverse audience – much more diverse than your average development team – so user testing at an early stage was essential. “If you have five developers and those five developers start to think about what a user might ask, in two days, you are going to go round in circles and come back to the same conclusions,” says Letesson. “We knew we had to invite users from BASE and the Fondation to test [the bot] as soon as possible.”

This testing provided valuable insights for the development team:  “Users were asking a lot of questions and of course the chatbot failed to answer most of them. It was always giving back the answer ‘I’m sorry, I’m new’. It was frustrating for the user but very enlightening for the engineers,” explains Letesson.

The next step was to map out conversations: “We visualised conversations with notes using yellow for rich cards [visual content offered up by the bot] and red for user input,” he explains. This is a common approach in UX and allowed the team to map out how users might jump from one question to the next – “for example, if a user is given information about an exhibition and she says ‘I want tickets for that’, then the bot must understand that she means tickets for that exhibition,” says Letesson. “Plotting all that out really helped us frame the conversation. It was then a matter of testing, failing and building new prototypes.” 

Teaching a bot to understand both French and English presented a further challenge – particularly as Letesson says Google’s technology is more advanced in English than it is in French. “In English, you can interact with the bot from scratch without having to teach it basic time or place related questions. It has a lot of knowledge already – but we still have people working several days per week reading user input and gathering a critical mass of input to make sure what we have is working correctly before we go crazy and start to do other stuff with it,” he explains.

BASE is now focusing on developing the bot and adding new features as AI technology develops. (He likens the current bot to a toddler with a basic grasp of language – one that has to learn phrases and speech patterns over time.) The bot will launch on Facebook Messenger once it has reached ‘maturity’ – a process that Letesson predicts will take from three to six months.

“The chatbot is the tip of the iceberg in terms of what we want to do with it – because it’s very experimental technology, the code is evolving every week,” he adds.

The Fondation Cartier website

The hope is that the bot will eventually offer a seamless experience for users – allowing them to learn more about artists and artistic movements, find out about new events and even buy tickets for exhibitions all from within social media or through the chatbot function on the Fondation’s website. BASE also hopes to save museum staff time by teaching the bot to answer a range of commonly asked questions – for example whether visitors are permitted to eat in the garden of the Fondation (something people often call the museum to ask).

Chatbots might still be hit and miss but as they become more sophisticated, Letesson believes more cultural organisations will seek to use them. “It’s about meeting people where they’re spending the most time,” he says.

One of the Fondation’s main objectives is to foster dialogue between artists and visitors – in the physical space this is done through hiring knowledgeable staff who are on hand to talk to visitors about the works in an exhibition – and Letesson sees the chatbot as an extension of this.

“It’s about creating a dialogue. At the beginning it’s a machine-oriented dialogue, but when you finally visit and you walk through exhibitions there will be people there who can walk you through them … people who really know what they’re talking about,” he explains. “[The bot] is a way to trigger that same feeling of being able to [interact] with an institution one-to-one.”

Letesson admits there is no guarantee that the chatbot will be successful but says the project has been viewed more as an experiment – a playful addition to a robust website that will hopefully serve the museum for years to come. “Even if the chatbot doesn’t work out the website will still be as we envisioned,” he adds.

The post Building a chatbot for Fondation Cartier appeared first on Creative Review.