Empathy is the capacity to understand and share the feelings of another.
In the novel Do androids dream of electric sheep? Philip K. Dick imagined a world in which humans faced the challenge of spotting bots and used for this a simple empathy test. If you have seen the movie Blade Runner, based in the book, you know what a Voight-Kampf test is. If you have not, here’s what it is:
Replicants (just another word for androids and robots) don’t react naturally. Think about it: most of us would have had an emotional reaction to the situation narrated. Leon, instead, gets lost in details such as which desert this happened at or what a tortoise is.
Most chatbots we’ve used in the last couple of months feel like that: out of context, not responding to calls for help… even Poncho, one of the few bots that seem to have users and a strategy, and that is otherwise really awesome, does not get this right. See below what happened when I tried to change the city of my weather report, restart the process or get the help page of the bot:
Chatbots, let’s face it, are still quite stupid. No NLP or AI API is there yet, and good luck with trying to DIY. Some other people have written about how annoying chatbots are and you can get a feeling by yourself just by looking at some of the chatbots already out there. If you speak Spanish you could even see for yourself how Eva, our chatbot for teen pregnancy prevention, also fails at many things.
One thing we can do, though, is using empathy, user research and the right team members and consultants to build empathic bots.
Explain what a chatbot is, what your chatbot’s purpose is, and the basic mechanics of your chatbot. Depending on your public this will need more or less detail. Run user interviews as soon as possible to learn how much your users know about chatbots and their functioning. You’d be surprised about how many people know about Siri but have never heard about chatbots.
Have a way for users to access help, reach a human, exit and restart the conversation
We like doing this with commands like help, human, restart… other chatbots have permanently visible buttons. This is, however you decide to implement it, key in faking some kind of understanding – or at least in not blocking the users in stupid never ending loops just because they want to do something a little different that what you designed your bot for.
Do one thing and do it well
Choose one domain and redirect the users to that field any time they try to ask you about the weather, the existence of God or the number of flies in Taiwan. Eva will laugh at your comments, but ask you to select a high level topic from the main menu immediately after admitting she actually did not understand what you asked.
Use language (and emoji) wisely
Short, concise, simple… make it easy to read and understand, and use the words your users use and read on their real lives. This can get tricky with languages like English or Spanish, where regional differences are important to get right.
Analyse the hell out of your bot
Measure conversation length, read them out loud to spot things that sound weird, check the keywords people use the most… There are many tools for this, we like to use a combination of old school analytics (provided by our chat building frameworks) and good old reading whole conversations. This last one is a must-do, as they give you the most human look at your audience’s behaviour. It does not scale as good, though.
Fake it ’til you make it
Can’t make your chatbot learn automatically from conversations? Change your messages and flows using what you learn from your analytics. For an extra wow effect fix broken flows as they are used for the first time and notify users in chat of the fix.
Don’t be scared to be needy
Push push push. Don’t wait for users to go there and find you for a chat. Specially if you have something to say. One of the features we want to add to Eva is a reminder to have responsible sex and buy condoms sent every second Friday. You can also use broadcast messages for research, to announce fixes…
We are at the beginning of the chatbot revolution, so a lot of the lessons are still to be learnt. What seems clear is that chatbots will be, for now, a task for the social scientists. It sounds like 2017 will be the year the inquisitive psychologist, the occurrent script writer and the nerdy domain expert join the code ninjas, devops rockstars and growth hacking gurus of tech teams. People that will make chatbots feel like they can understand and, why not, dream.