Last time, I talked to you about how clearly I saw the need for a political chatbot — mainly as a halfway point between canvassers talking to people, and content on a website.
A good chatbot can function to handle people’s common questions about your party expediently. However it still sounds like something in the realm of science fiction, and I’ve been having trouble getting people to take it seriously.
In the end, I decided to the easiest way to make my case was to build a demonstrator — so today, we meet Tane, my unofficial Green Party chatbot.
With this in mind, I’ve set up an unofficial Green Party Facebook page from my account, and built a chatbot to handle questions.
If you remember I fleshed out a lot of design and testing considerations for a chatbot in a previous story. That first attempt at a chatbot was for one which supports my blog, and had a cocky personality which could direct you to content. I own that blog, and my own “brand”, so I could pretty much do what I like.
I’m a Green Party member, but I’m not senior within the party at all. Hence in design, I had to try an limit myself — I needed to try and work with preexisting material from the website as much as possible, and likewise make sure that what I built was respectful to that.
Even so, one of my first things on my ‘to do’ list was to make sure the Facebook page and the chatbot are explicitly clear that this is a technology demonstrator, and not officially sanctioned by The Green Party.
Starting with the personality
I wanted this chatbot to look and feel very different to my previous attempt. I actually ran a competition in my team over March around designing a chatbot using Chatfuel, and some of what they built, especially exploring featuers in Chatfuel were so good, I’ve just had to steal them! A lesson if ever there was that learning can be a two-way process.
As I’ve talked before about chatbots and AI, I felt it was important to start by working out the personality of my chatbot. I like the idea of modelling the chatbot with a persona to help me when shaping their responses — it’s almost a character sheet.
So first of all … gender. Well, most assistants are female — think Cortana, Siri etc. This has been commented on such as here, the idea that an assisant is subservient, so make it female. So it felt pretty obvious for the Green party we’d have to flip that, and make my chatbot male.
I decided to give my chatbot several attributes which aligned with Green values. He would be Maori, and named Tane, and have a special response for Maori greetings. [This is because the Green party wants to encourage the learning of Te Reo Maori in schools]
He would also be gay, with an immigrant boyfriend, also embracing the core Green party values of diversity and equality. And he could talk a little about these things.
Tane was roughly modelled on some of the people I’ve met at Green meetings. So he feels like he represents people you’d meet within the party.
I found myself borrowing the language patterns of one candidate in particular, who I found to be an soft-spoken, but excellent communicator.
Giving him a face
When I’ve run exercises on chatbots, one of the most important things you can do to give them a sense of character is to give them a profile picture from the off. This really helps to give them a bit of an identity over being a shadowy silhouette. Any response you write, it helps to look at that profile picture and ask “does this feel consistent”.
I thought about searching through Google to find an appropriate face, however given as I’d be eventually (hopefully) showing this to a Green MP. This was fraught with issues if I’d used someone’s face without their permission. In the end I found a picture of a single tree, which seemed to embody and embrace the core Green values.
Passive engagement only
Tane is designed as a tool to engage with people interested in the politics of The Green Party, but he’s designed to be passive, not a pest.
He’s designed to respond to questions from the public about what The Green Party stands for. But the user has to initiate the conversation. He’s not going to try and befriend you on Facebook, and contact you without your consent to go “hey new human friend, I’d just like to talk to you about the upcoming election”.
To me, this is an important distinction. I want to use chatbots as a channel to service visitors with questions, not as another form of spam.
Unfortunately chatbots have a bad rep because there are too many people using them to force advertising down your throat. I’m really not very keen on this at all. But I also think they have an amazing power when used responsibly to help service people faster, not just for nuisance sales.
Collect together policies
The core of any political party really are their policies, what they would do. So this was a great place to start in building up material for Tane.
As I’ve said, for the Green Party (and other parties in New Zealand) policies are somewhat spread out all about the website. There are multiple policies which all try and take action in a single area — for example “improving the environment” as a goal, there are policies about using the rail system more for freight, greater use of renewable energy, tax incentives for companies using environmentally friendly transport.
My approach to handling this information was to have a response to a general topic which was a statement of intent of their values in this area. Underneath this would be a series of selectable material, which you could scroll through — this would link to the website and take you to more information.
I like to call this the “activist engagement” — somewhat like how people such as protesters or evangelists ideally engage with you. They tell you a high level explanation of why this stuff matters, then give you the option of a relevant pamphlet for more details.
A good example is the response I’ve got in place around equality, which is a catch-all. It should pick up any responses about age, sexuality, race and use this response,
Below this is a range of items (including picture) which a user can scroll through. Chatfuel calls this a gallery, and it’s one of the tricks I learned from my students. It allows a user to select more details on the specific topic that most fits them …
Although women’s policy is here, there’s also a separate response block for Women specifically, as there are a few special topics there. If you have a question with triggers about specific women’s rights, domestic violence, or equal pay, you’ll be taken to this section (which collects a specific umbrella of women’s issues together).
Did women really need it’s own separate equality section? Should each type of equality have it’s own response? You could argue for and against those, but it just felt the right way to handle the material from the website. But it’s one of those decisions which can be changed and refined as people use the chatbot. What I have learned from my first chatbot is that they are definitely something you refine as people use them!
So on and on I continued building, until I was left with this section nicely filled out with all the core policies …
It’s not just policies
Right now, ticking down to the general election in the UK, you could make a pretty decent approximation of a Teresa May chatbot by just repeating the phrase “strong and stable leadership” ad-nauseum. Indeed her public speaking seems to extend to very little else!
As I’ve said before, to make an engaging chatbot, you need to engage on more than just your core topics. People will insist on trying to be sociable.
I worked general questions which might be asked about what Tane’s favourite drink and food into core Green Party policies on clean water and support for school dinners.
However, this still left a lot of sociable elements, as well as some core items like “what are your values”, “take me to your leader” and “how to respond to that f-bomb”.
I added several additional layers of response which you can see below … these are organised into social interactions that Chatfuel calls “blocks” …
These are core blocks about the party …
“Help me Tane” — if you say this phrase, or anything including the word help, Tane will give you more detailed instructions on how to better ask questions of him.
“Speak to a human” — assumes you’re not getting the information you need from Tane, so will allow you to contact the Green Party directly.
“Leaders” — gives information on the two co-leaders of the party.
“Values” / “Campaign” — gives the details about the values of the party together with taking you to the core policies of the party.
“Volunteer” — tells you how you can help out the Party.
Reasons To Vote Blocks
“Green” (Party) — This includes pre-determined responses that talk at a high level why you should vote for The Green Party, how a vote for The Green Party isn’t a wasted vote (due to proportional representation).
“National” (Party) — Why National’s track record is something we want to break.
“Labour” (Party) — Why although The Green Party is likely to form a coalition with Labour, you should still vote for The Green Party over Labour if you can.
As ever, this includes a response to any kind of swearing, and if asked about his personal life / is he in love / is he in a relationship, Tane will talk about his boyfriend and commitment to diversity.
All this can make it seem that you have a pretty rugged design … then along comes testing. We’ll see what that uncovered next time!
If you’d like to have a go with Tane, he can be found at my Unofficial Green Party page here “@greenTane” on Facebook.
Want to create your own chatbot?
I’m running a workshop on building a chatbot in November at Agile Testing Days. I’m just one of many amazing reasons to attend — why not find out more about what’s on offer here?
Talking Politics 2 — Meet Tane, my friend electric was originally published in The Test Sheep on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source: ministry of testing
Talking Politics 2 — Meet Tane, my friend electric