Pia's personality as movement 💃


#1

How do you give Pia a personality… Well there are two approaches we are pursuing to do this. First off, we have been looking closely at the dialogue flows of Pia and it communicates with you. Style, tone and language use are all critical and something that we aim to tailor based on who we are speaking to.

Second up, we have been working on giving Pia an animated form. I am not talking about an avatar or something pretending to be a human in design and style. Pia will take a fare more fluid but omnipresent form that represents the fact that it is technology and not human. Sounds suitably vague right :face_with_raised_eyebrow:

Why don’t I try to show you… Below are four different forms that we are building for when a user is chatting to Pia or interacting on content throughout the site. Pia will alter its form and speed depending on the type of response it plans to come back with or the content it is trying to load. The way that we can highlight personality is via the animation characteristics. A subtle difference that probably 95% of you will never directly notice but subconsciously helps to build a stronger personality of Pia.

  1. :point_down: Shown when Pia is loading content, relatively fast motion and to the point:
    PIA_AN01_loading
  2. :point_down: Shown when Pia is trying to solve a problem or calculating something, slower with an increased distortion to show the task is hard to solve.
    PIA_AN01_solving
  3. :point_down: Shown when Pia is responding to a chat interaction, simple and smooth
  4. :point_down: Shown when Pia is suggesting a page, website, video or piece of content, rises at the end of the animation to indicate a rhetorical question or response… How about this?

We are really looking forward to sharing with you these movements in practice in the next couple of weeks. It should be fun to try it all out and please do share any of your comments below…


No more stock photos please
#2

:star_struck: Ooooh! In my opinion, these look great. I can imagine it’s hard to mimic feelings with shapes and blobs, but I can definitely relate to these. Where abouts in the chat are these going to be @James?


#3

Hey, good question (thanks for setting me up, next round of coffees on me :stuck_out_tongue_winking_eye:). They will either be used within the chat interface itself as Pia is responding to a user. Alternatively, they will also be used within the main site in the usual places where you would expect to see loading symbols etc :+1:


#4

giphy They look great @James. A lot better than this annoying guy!


#5

Smooth! :kissing_smiling_eyes:


#6

Looking epic @James :raised_hands:


#7

Am liking this! It capture the dynamic we seek…


#8

Pleased you like it @ian! Looking forward to sharing more with you soon :grin:


#10

Check out number 3 (Pia responding to a chat interaction) in action! :speech_balloon: Chat looking flawless on mobile! :iphone:
ezgif-3-09351c296a


#11

Hi, nice indeed. I think that the chat should probably NOT start with open ended questions. As of now, if I really type what’s on my mind, pia is obviously not prepared and the same for most users who would want to test the bot, I guess. It can get a bit frustrating after the 3rd time pia does not understand what I mean. Probably better giving some limited options from the beginning to frame the discussion.


#12

Hi @hervelequer, thanks for your comment. This is just an example design.

One of the core elements of the beta test is to train Pia too. Therefore, we actively WANT Pia to not know the answers to things as that then helps us to teach Pia the answers to questions it gets asked most frequently. We will build some decision tree based dialogue flows over time but actually it is far more beneficial to train AI without constraints as the questions we get asked help to form the future roadmap of more structured decision tree flows. I hope that makes sense?


#13

Thanks @James, it makes sense. Let’s see how it goes but I would the suggest that pia does not get to repeat “I don’t understand” too many times in a row, quite frustrating tbh. Maybe you could introduce a decision tree after the 3rd fail, to get discussion going. Especially at the start when pia still has lots to learn… what do you think?


#14

Thanks @hervelequer and understood. Let me work through some options and I hear your frustrations.


#15

Ah don’t get me wrong, I’m personally not frustrated at all. But I can imagine the frustration / disappointment of a first time user immediatly hitting a wall when engaging with pia. I guess that many people’s approach would be similar to mine. Like, challenging pia’s IQ and asking trick questions out of curiosity. If pia does not understand, it should act as human as possible, rewording everytime and finally coming up with plausible options to get moving…


#16

Thanks @hervelequer, we are already looking in to some options for this. Added to the roadmap and looking forward to testing out with everyone soon

Cheers

James