Chatbots Magazine Become a Writer Intro to Chatbots

Giving Bots a Personality?

Hi Aiden,

Tx for your response. It is indeed an exciting idea. Through trial and error, I find out that a lot of theories about Persuasion (ranging from Cialdini to Eyal and more academic ones) are simple to translate into examples, but do not provide a framework for a complete and consistent ‘personality’. In other words, a bot that applies them all is too pushy and comes across as a desperate prostitute or car salesman.
On the other hand, making a choice between them might miss opportunities; so the key is to test which strategy might work in which situation, or more important, to which person you are talking with (as bot).
Having knowledge about that person helps, of course, so the applied persuasion technique needs to be ‘tested’ first by asking questions; as limited as possible.
That is a very exciting balance between ‘scanning’ the subject person and ‘exposing’ the person to the technique.

Nothing is publicly live yet, but I will share as soon as I have something worth showing :wink:

Working with Persona is a possible approach; definitely. I focus at ‘sensitivities’ now: trying to see if a subject would qualify for being sensitive to persuasion principles like ‘scarcety’ or ‘exclusivety’ or ‘authority’. That should be aligned to the persona approach you describe, I guess.

1 Like

The personality is the user experience. In order for us to know how to interact with one another as people our personalities, archetypes, roles and ways we use words allow us to interact - and understand the value of the interaction. We develop a personality from research before we get into the design… We generally look at the role of the “helper” that the bot will be replacing or amplifying, and then look at the archetype (insert Joseph Cambell here!) and figure out ways to make the bot more humane from there. The personality, which ends up being an archetype that we tweak, allows people to know what they’re doing with the bot. We start with words, audition actors for voice, and then, if there is an appearance, develop from there. So all that stuff this personality design. Personality is the user experience! Otherwise, you end up with a robotic robot – which none of us want to talk with.

4 Likes

I have a whole article I need to write on this. My co-founder @thekevinscott and I are taking a bet on the evolution of User Experience Design to Personality Design. A few months ago we did a pretty thorough set of exercises on finding a voice and personality for our bot. We drew from traditional marketing materials, as well as storytelling and film making techniques to establish a character.

We created spectrums of personalities—one of my favorites was “Level of shit-togetherness”, with the spectrum ranging from Abbi and Ilana from Broad City, all the way to President Barack Obama.

Happy to share more information about this if there’s interest.

5 Likes

@ari your open question at the end: of course there’s interest! :wink: Sounds like a great exploration you guys did there!

Hey Ari, sounds very interesting indeed, how would you like to have a quick skype call to discuss your findings in greater detail? aiden.livingston on Skype and aiden@castingai.com for scheduling purposes.

@ari i’m interested too. :grinning:

When I started building my personal assistant bot, Nitbit, one of the most important things was that it would be able to talk back to me.

I created an annoyance script, so that repeating memes (or it could be anything else) eventually makes the bot annoyed and it will stop answering to me until I apologize, and properly. I’m also teaching it how to learn to curse from me.

It also keeps a positivity score using Sentimental, which keeps track of how positive a user is based on the AFINN-111 wordlist. I have yet to utilize this feature in any particular fashion, but I plan to use that, as well as other metrics, to change the bots responses.

3 Likes

Intriguing concept, So often in the context of bot personality everyone focuses on how to make it more “ass-kissing”, I like the idea of giving it the ability to be annoyed, really gives extra depth to the personality. After all total suck up are indeed kind of annoying. Could I talk with NitBit and see it in action?

I finally published the story about how we designed Emojibot’s personality.

https://medium.com/@azilnik/designing-a-chatbots-personality-52dcf1f4df7d

Feedback welcome!

cc: @Nas @nei.fda @aiden

1 Like

Do you have an example one pager you’d be willing to share?

Hey @ari would love to read what you have to share about that. Thanks!

Hey @pmcpinto ! I actually just published the story finally, would love to know what you think. Thanks for the interest. Here it is:

https://chatbotsmagazine.com/designing-a-chatbots-personality-52dcf1f4df7d

1 Like

@ari great post! I am a student at UofM and this fall break we visited many design consulting firms who spoke a lot about designing brand personality. Giving that personality a voice through a bot is awesome! Love how you played the hand of god and used personality traits of famous people as building blocks to Emoji Salad’s personality.

My team and I are working on a healthcare Bot. She is live in India, collecting patient info that would have been normally missed in consultations due to immense constraints in the healthcare system. We had to build her to be as robotic as possible: patients felt uncomfortable giving medical data to a bot with a personality. Interesting that for us we had to design for a lack of personality, which wasn’t hard at all honestly.

I guess what industry the Bot is in, what function are important design constraints for personality design.

@rakshay really interesting insight about designing for no personality. How did you get to that decision, what testing did you do with patients? Really curious to learn more about it

@ari Would love to share insights: First some context: our app connects doctors with patients; we use the Chatbot (human-machine) to collect information before the (human-human) conversation happens. The doctor can use these extra data points for decision making.

Starting with the name of the Chatbot. If we named her ‘Geeta’, a common Indian name, patients would not share any information or interact with her.

To solve this we went to the field and observed. We realised patients shared all their information without any apprehension to the nurse or the doctor’s assistant. (surprise surprise) Once patients see the word “Doctor” there is a halo effect of trust. We changed the name to ‘Doctor Assistant’. Boom response rate shot up.

Second, with respect to the style of responses: data showed that having a more questionnaire based structure rather than a more conversational style of asking questions had a better response rate. This is still a work in progress.

I really enjoyed reading this thread. Giving the BOT a personality is silent passion of our entire tech team and many times we let this get in the way of what our user actually NEEDs. To see the BOT community designing personalities for entertainment, travel etc. is really exciting.

@rakshay nice work and nice insights! Might want to do a writeup about your process as well!

Hey @ari great article.

Really liked, gave me some nice insights. And i’m finding out more and more that we can’t build a great bot without a good personality and for this storytelling helps a lot to creat the bot persona and to understand the context.

And great “celebrities board”. I used similar techniques but in brand image researchs and it makes total sense to use for chatbots personality.

Thanks e congratulation! :grinning:

Thanks @nei.fda! I appreciate it.

Yes, personality is what makes the conversation with a bot enjoyable and keeps it going. Here’s how we approach it.

As a company developing bots for various businesses in Europe and in the customer service space our methodology has been focused on understanding person’s intent and tailoring the response accordingly. So when a person is curious we respond one way, when a person is angry we respond another way. It comes down to profiling responses and having the bot react to that. We have developed various scenarios and responses to different emotional states that users can have. With scriptwriters we produce workflows and conversation structure that makes the conversation more enjoyable.

In addition a bulk of our in-house AI team is working on deep learning techniques to actually develop intelligence so that the bots can themselves assess the sentiment of the conversation and to give answers that make most sense at any given time (relevancy + personality style). This includes analysing conversations and using our own neural networks to train models that are capable of understanding user intent and style.

Mark Marino wrote an article in 2014 about the implications of having bots represent racial/ethnic identities through embodied performance (basically, through embedded ‘personalities’). Definitely worth a read for anyone interested in this thread. And it’s open access!
Mark Marino, ‘The Racial Formation of Chatbots’, CLC Web: Comparative Literature and Culture, 16.5 (2014) https://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=2560&context=clcweb