2. #MachineLearningForDesigners
A Practical Definition of
ML
Become buzzword compliant
Interaction Intelligence
Start your engines
The (Near) Future of
Design
Futureproof your UX skills
Part I
Part II
Part III
4. #MachineLearningForDesigners
Artificial Intelligence
Any showing of “intelligence” made by a computer
could be called AI.
"AI is whatever hasn't been done yet." – Douglas Hofstadter
The scope of AI is disputed: as machines become increasingly
capable the definition of "intelligence" changes, a phenomenon
known as the AI effect.
Part I | A Practical Definition of Machine Learning
8. #MachineLearningForDesigners
It’s a cat if it has whiskers
and it is furry
and it is small
What is a cat? version 2
Part I | A Practical Definition of Machine Learning
9. #MachineLearningForDesigners
It’s a cat if it has whiskers
and it is furry
and it is small
and doesn’t climb trees
What is a cat? version 3
Part I | A Practical Definition of Machine Learning
10. #MachineLearningForDesigners
Is this a cat?
Rules to detect a cat:
1. It has whiskers
2. It is furry
3. It is small
Traditional programming... Machine learning
Let me guess how i can distinguish
a cat :)
Machine Learning
Part I | A Practical Definition of Machine Learning
11. #MachineLearningForDesigners
Part II | Intelligent Interactions
Part II
Intelligent
Interactions
Making sense of data
Designing for the machine
New fundamentals
Transparency and control
12. #MachineLearningForDesigners
UnstructuredStructured
● It fits better in a table
● Closed-option variables. No free-text data
● It’s “less human”
● Easier for machines
● Organic, no predefined format
● Freeform text fields (like reviews)
● It’s “more human”
● Harder for machines
Structured vs unstructured data
Part II | Intelligent Interactions: Making sense of data
13. #MachineLearningForDesigners
Machines “prefer” structured data.
It’s easier to use and you can create better experiences with it:
Unstructured vs structured data: problem
Part II | Intelligent Interactions: Making sense of data
Typically text, images and
videos.
15. #MachineLearningForDesigners
Feedback Loops
User
“Oh! You don’t
like this song”
Recommend
different style song
Recommend
similar song
Listen to
song
Like
song
Dislike
song
Skip
song
Recommend
similar song
Listen to
similar song
Machine Recommend
different song
Part II | Intelligent Interactions: Design for the machine
16. #MachineLearningForDesigners
What I say I like vs What I actually watch
Implicit vs Explicit Feedback
Part II | Intelligent Interactions: Design for the machine
19. #MachineLearningForDesigners
• What data does my algorithm need to do its job?
• How do users provide feedback?
• Why would users want to give it feedback?
• What failures might I anticipate?
New Considerations
24. #MachineLearningForDesigners
Bedrooms
Bathrooms
Sq ft / Sq m
Location
Parking
Amenities
Last sale price
Comparable sales
etc...
Estimated price (range)
Data Algorithm(s) Output Feedback Loop
Regression engine Actual sales price
User corrections
House price forecasting
Part II | Intelligent Interactions: New fundamentals of design
26. #MachineLearningForDesigners
Past book purchases
Other purchases
Comparison purchases
etc..
Book suggestions
Data Algorithm(s) Output Feedback Loop
Recommendation
engine
Actual Conversion
Wishlist
Dislikes
Shopping recommendations
Part II | Intelligent Interactions: New fundamentals of design
33. #MachineLearningForDesigners
• What intelligence can I add to the experience?
• Can I personalize the experience?
• What needs might I anticipate for my user?
• What data do I need (or can I use)?
• What can we glean from unstructured data?
New Considerations
40. #MachineLearningForDesigners
• Where might there be need for transparency?
• Where might there be issues of trust?
• Does my design raise concern over privacy?
• Is there need for user privacy controls?
• Should I communicate uncertainty (degree of
confidence) in my advice?
New Considerations
42. #MachineLearningForDesigners
To a Designer a recommendation is... To a Data Scientist a recommendation is...
Advice or suggestions given to users A system of ordering things according to each user preferences
Part III | The (near) future of design: Engines of Intelligent Experiences
We need a common language
46. #MachineLearningForDesigners
• Befriend a data scientist
• Educate and evangelize
• Keep the conversation going
BBVA Data Analytics Blog
BBVA G+ Communities:
Machine Learning Technologies
Join the community:
jonah.burlingame@springstudio.com
@mind_arc
alejandro.vidal@bbvadata.com
@doblepensador
#MachineLearningForDesigners
47. #MachineLearningForDesigners
Machine Learning for Humans by BBVA Data & Analytics
Machine Learning for Designers by Patrick Hebron
Experience Design in the Machine Learning Era by Fabien
Girardin
When User Experience Designers Partner with Data
Scientists by Fabien Girardin and Neal Lathia
The Step-By-Step PM Guide to Building Machine Learning
Based Products by Yael Gavish
Specially: #5: Machine Learning is Very Much a UX
Problem
Human-Centered Machine Learning By Josh Lovejoy and
Jess Holbrook
Applications Of Machine Learning For Designers by Lassi
Liikkanen
Data Jujitsu by DJ Patil
Power to the People: How One Unknown Group of
Researchers Holds the Key to Using AI to Solve Real
Human Problems by Greg Borenstein
The ethics of good design: A principle for the connected
age by Aaron Weyenberg
Are notifications a dark pattern? by Andrew Wilshere
A Strategists guide to artificial intelligence by Anand Rao
Further reading ;)
Editor's Notes
[JONAH]
Right level of technical depth so you realize ML is not magic but not so technical that you feel stupid for not understanding.
Want real feedback. Paper evaluation.
Introduce Alex Vidal Mata
[JONAH]
[JONAH]
[JONAH]
Douglas Hofstadter :"AI is whatever hasn't been done yet.
[JONAH]
There are a lot of intelligent behaviours across all the industries. Sometimes we’re so used to them that we don’t take attention to them and we think they are not intelligent.
Note that intelligence definition is very tight with each problem to solve. We’ll talk more about this later.
[JONAH]
AI is the ‘umbrella term’ for a larger group of technologies. Machine Learning is but one science within AI. Similar to how UX is an umbrella term for many disciplines. Today we are going to focus on ML specifically.
[ALEX]
ML is the answer of AI to one critical question… What is a cat?
Imagine that you want create a product like isthisacat.com when any user could upload a picture of his pet and know if his pet is a cat or not.
For that amazing Series A funded startup you need one of these:
A huge workforce to process all the pictures and manage your huge web traffic
An AI algorithm that classify cats
The latter sounds easier and cheaper, let’s try it. We have to program the algorithm in the next slide
Our first minimum viable algorithm is this one. We’ll classify the image as a cat if it has whisker and it is furry.
Perfect. Let’s try it with a few examples. Aw, so cute.
But… ups. We didn’t take into account lions. Ok, not problem. Next iteration’ll be better.
[ALEX]
Oops...
[ALEX]
As you can see, even with this seemingly simple example this method of programming with rules has limitations - it simply doesn’t scale.
[ALEX]
ML is the answer of AI to one critical question… What is a cat?
Imagine that you want create a product like isthisacat.com when any user could upload a picture of his pet and know if his pet is a cat or not.
For that amazing Series A funded startup you need one of these:
A huge workforce to process all the pictures and manage your huge web traffic
An AI algorithm that classify cats
The latter sounds easier and cheaper, let’s try it. We have to program the algorithm in the next slide
[JONAH]
ALEX
ALEX
ALEX
ALEX
ALEx
ALEX
ALEX
Talk about the first years of Material Design Guidelines.
[ALEX]
ALEX
[ALEX]
ALEX
(COment here this change allow us in banking idutry to server better our 90% of our clients with personalized advisory)
ALEX
ALEX
[ALEX]
[ALEX]
[ALEX]
[ALEX]
And here?
Maybe we can think about a continiuum between two extreme cases (work in progress)
You can add in your interface options to “forget” things about your users. For example here Amazon allow me to remove from my browsing history any items that I don’t want them to keep
And here?
Maybe we can think about a continiuum between two extreme cases (work in progress)
And here?
Maybe we can think about a continiuum between two extreme cases (work in progress)
And here?
Maybe we can think about a continiuum between two extreme cases (work in progress)
[JONAH]
ALEX
I’m going to share with you something that happens three times in the exact same way in three different projects.
Do you remember the airbnb’s insight example? It looks like
For a DS a recommendation system is a very specific kind of algorithm that tryes to rank for each user the most prefered items (books/movies/songs…) for him.
If look the Airbnb’s example they’re recommending you to change your price becuase probably your price is more than they expected.
Try to look for the differences. Left side a machine tries to arrange all the books that amazon has in a way that the books you’d love will be on the first positions
As I told before touchable devices changed everything in the practice of design. But now with ML we’ve new opportunities