Chatbots Magazine Become a Writer Intro to Chatbots

Designing a Chatbot's Personality

(Ari Zilnik) #1

Some people here were interested in me writing this story, so I figured I’d share the work @thekevinscott and I did.

Feedback welcome!

(Nas) #2

Hail to @ari!
It took you some time to write this one, but it was worth the wait. Great article that is funny, interesting and has real cliffhangers. You must have a great personality yourself :wink: Thanks for sharing these experiences with the community.

The last bit triggers me: I’m also trying to humanize output by using context. Humor is often only funny in its context and remarks are often only relevant in their context, but I feel like I’m drowning in the process often. Contexts actually overlap, or are multi-interpretable (is that a word in English?), or they are too vague and do not guide output, tone-of-voice or other aspects in which personality can ‘show’.

Yet, I do have the experience already that (the added value of) personality shows best exactly in those circumstances: when contexts that can be explained in different ways are picked up in a specific and recognizable way: that is how personality shines through the words, in a more poetic way of saying.

Anyway, keep up the great work and good luck with the next steps!

How do you manage Conversation States?
(Ari Zilnik) #3

Thanks for reading, @Nas!

Yeah I was actually thinking about what you just wrote! Humor is often used as a crutch for a bad experience. For example, the first “version” of the personality we were testing was really mean and rude because I thought it would be funny. And it was, with my friends. But then I tested it with a wider network of people, and they thought the bot was unnecessarily mean. We wanted to be honest, and sometimes it’s okay to be boring instead of funny if that means that the conversation is more clear and understandable.

What do you think?

(Nas) #4

Yup, same experience @ari . I also thought insults and black humor spitted out by code were incredibly funny and ‘edgy’, which was not…let’s say… widely shared when I started trying things with a more ‘average’ audience. At all.

Clarity is no 1, maybe even before (perceived) honesty, in my experience. Only when people start ‘trusting’ the bot (…and how do we measure that? I have no idea yet), intelligent humor seems to be appropriate. With intelligent humor I mean remarks and wording that reflects the context, rather than what you described as a ‘crutch for bad experience’ and what some websites show on their ‘funny’ 404-page.

My bots with personality designs so far try to (after a process of describing personality traits) deploy its personality by:
1 - at first show personality by use of wording, but always let clarity prevail
2 - meanwhile determine the trust level, so ‘Intelligent humor’ can be used when this variable reaches a certain threshold (still experimenting with this one)
3 - try to stay an interesting bot by alternating humor, empathy (or the explicit lack thereof, which can be funny…), confirmation and initiative

The latter factor (initiative) is the most exciting one for me right now: when and how can the bot take a new initiative in the dialogue, without it feeling awkward, or like the bot doesn’t understand the user. Like in a boring dialogue between 2 people when both know: “now we need to talk about something else or go home”.

Meanwhile, some of my developers start hating my personality due to increasing complexity for every sentence the bot is spitting out. But hey, my personality is great, my mother told me. :wink:

(0) #5
As designers, personality is at the forefront of user experience.

Really great read @Ari !!

@Nas have u play’d Fall Out? I think the conversational model that they use would allow for the ‘trust level’ to be determined ( could easily test how ‘dark’ a sense of humor & adjust bot personality accordingly ).

(Nas) #6

He, @beta, thanks for the suggestion. Nope, I’m not much of a gamer (life is too exciting atm) but I’m definitely going to try, just for the experience of it!

(scottganz) #7

I wouldn’t say necessarily that humor is a crutch or a detriment. Humor’s original evolutionary purpose is to deal with things that are wrong or amiss in a way that relieves the tension that the issue would otherwise create. Therefore, a great way to integrate humor into a bot is to make jokes about the fact that the bot is itself a bot. This way, the humor is self-deprecating. It comes at the expense of the bot and not the user. This makes the bot more honest and builds a bit of sympathy. But like all things, it has to be used gently or it will become tiresome.