It takes a village to raise a bot

A presentation at Frontend Conference in August 2018 in Zürich, Switzerland by Desiree Garcia

Slide 1

Slide 1

e g a l l i v a s e k a t It t o b a e s i a r to Desiree García @thedezzie

Slide 2

Slide 2

Slide 3

Slide 3

Conversational systems and emerging AI

Slide 4

Slide 4

"Just add water"

Slide 5

Slide 5

Slide 6

Slide 6

"One size fits all"

Slide 7

Slide 7

Slide 8

Slide 8

Slide 9

Slide 9

Engineers couldn't do it alone.

Slide 10

Slide 10

Why do people want conversational systems?

Slide 11

Slide 11

“Don’t play what’s there, play what’s not there.”

  • Miles Davis

Slide 12

Slide 12

Machines still have a hard time with language http://bit.ly/ai-language-problems

Slide 13

Slide 13

Simple conversational systems that are great

Slide 14

Slide 14

Slide 15

Slide 15

Complex conversational systems that are great

Slide 16

Slide 16

Slide 17

Slide 17

Slide 18

Slide 18

Slide 19

Slide 19

Time + Money

Slide 20

Slide 20

Is it convincing?

Slide 21

Slide 21

Is it convincing effective?

Slide 22

Slide 22

Content first

Slide 23

Slide 23

Content last

Slide 24

Slide 24

Intents Entities Dialog builder #greeting Conversation starts Condition #greeting Watson responses (0) No responses added Edit . Order pizza flow and some … Order pizza flow and some … Condition Condition No conditions set No conditions set Watson responses 20 We got ya down for a $topping pizza. Anything else Watson responses 20 We got ya down for a $topping pizza. Introduction

Slide 25

Slide 25

@location . @restaurant @cuisine

Slide 26

Slide 26

"In just minutes a day"

Slide 27

Slide 27

Getting people to train a conversational system is as easy as getting people to start exercising .

Slide 28

Slide 28

. http://www.r2d3.us/

Slide 29

Slide 29

Initial training ingestion employee manual Initial training improvement apprenticeship ongoing training refinement mastery

Slide 30

Slide 30

Content matters, because training matters

Slide 31

Slide 31

Deploy your conversational system app Assess its performance using human judgment Optimize its operating thresholds Retrain machine learning model with golden examples from humans 5. Lather, rinse, repeat

Slide 32

Slide 32

Deploy your conversational system app Assess its performance using human judgment Optimize its operating thresholds Retrain machine learning model with golden examples from humans 5. Lather, rinse, repeat

Slide 33

Slide 33

"Shut it down, shut it down!"

Slide 34

Slide 34

In AI, everyone is learning as they go along

Slide 35

Slide 35

They built teams.

Slide 36

Slide 36

developer domain expert sme content strategist ux designer

Slide 37

Slide 37

developer domain expert

Slide 38

Slide 38

Intents Entities Dialog builder #greeting Conversation starts Condition #greeting Watson responses (0) No responses added Edit . Order pizza flow and some … Order pizza flow and some … Condition Condition No conditions set No conditions set Watson responses 20 We got ya down for a $topping pizza. Anything else Watson responses 20 We got ya down for a $topping pizza. Introduction

Slide 39

Slide 39

@location . . #recommend @restaurant @cuisine

Slide 40

Slide 40

Empowering the creators

Slide 41

Slide 41

developer business domain expert data scientist

Slide 42

Slide 42

domain expert

Slide 43

Slide 43

"Ther e's w ater i n my room !" e k i l u o y d l "Wou e m o s u o y bring us to ice?"

Slide 44

Slide 44

They're closest to the content

Slide 45

Slide 45

moar content engineer data scientist domain expert business

Slide 46

Slide 46

How's your relationship to content?

Slide 47

Slide 47

moar content engineer data scientist domain expert business

Slide 48

Slide 48

"It's just content"

Slide 49

Slide 49

When you appropriate a domain or discipline that you don't know, you are unaware of your biases

Slide 50

Slide 50

Slide 51

Slide 51

"It's just content"

Slide 52

Slide 52

...what if you don't need AI?

Slide 53

Slide 53

Slide 54

Slide 54

Yay!

Slide 55

Slide 55

Inspiration for this talk

Slide 56

Slide 56

Slide 57

Slide 57

Would you put a baby in charge of customer support?

Slide 58

Slide 58

How about letting a toddler recommend medical treatment?

Slide 59

Slide 59

Or getting financial advice from a teenager?

Slide 60

Slide 60

It takes a village to raise a chatbot

Slide 61

Slide 61

Thank you! These slides live here: bit.ly/raising-chatbots An visual introduction to machine learning: r2d3.us A review of chatbot platforms bit.ly/chatbot-platforms An excellent guide to conversational design Conversational Design by Erika Hall (A Book Apart) Sparkle and Spin - A Book About Words, by Ann Rand and Paul Rand Twitter: @thedezzie Email: dezzie@automattic.com