(AI-first design)
Welcome to the techno-forest, where the most precious resources are data and attention. You are probably already augmented with either the speed and stamina of horse legs or the precision of biomechatronics. If not, a lot of the people you're going to be interacting with are.
Machine learning and other techniques give you new ways to direct your attention and your users'. But that involves balancing when it's best to surface the thing someone almost certainly wants and when you help them explore new corners of the world. As we'll discuss, AI-first design reaffirms how important problem statements, goals, and feedback loops are. It gives you more--or at least different--arrows in your quiver. And it shows how laser focus can make you blind to the humanity all around you if you're not careful.
2. WELCOME TO THE WRITTEN VERSION OF THIS PRESENTATION!
HERE’S THE ABSTRACT FOR THIS TALK
▸ Welcome to the techno-forest, where the most precious resources are data
and attention. You are probably already augmented with either the speed and
stamina of horse legs or the precision of biomechatronics. If not, a lot of the
people you're going to be interacting with are.
▸ Machine learning and other techniques give you new ways to direct your
attention and your users'. But that involves balancing when it's best to surface
the thing someone almost certainly wants and when you help them explore
new corners of the world. As we'll discuss, AI-first design reaffirms how
important problem statements, goals, and feedback loops are. It gives you
more--or at least different--arrows in your quiver. And it shows how laser focus
can make you blind to the humanity all around you if you're not careful.
3. KEY ATTRIBUTES OF CYBORGS AND CENTAURS (OTHER THAN GREAT ABS)
PRECISION, POWER, SPEED, STAMINA
4. (YOU MAY NOT ALWAYS RECOGNIZE CENTAURS/CYBORGS, I WON’T REALLY BE
TALKING ABOUT PROSTHETICS)
TWO PERSPECTIVES, WE’LL GO BACK AND FORTH
1. Designers inevitably turn users into cyborgs and centaurs
2. People do a lot more than just use your products/services,
so they come to you as centaurs/cyborgs already
6. KASPAROV’S CENTAUR THINKING
NOT HUMAN VS. MACHINE
▸ But human PLUS machine
▸ Human
▸ Intuition
▸ Creativity
▸ Empathy
▸ Machine
▸ Brute-force memory and
calculation
http://bit.ly/CentaurHumanComputer
7. WORDS THAT PEOPLE LIKE TO USE IN THEIR DEFINITIONS ARE ACQUIRED LATER
“EXPLAIN AI TO ME LIKE I’M A 5 YEAR-OLD”
http://bit.ly/AI5yearold
8. USING ONLY WORDS THAT 5 YEAR OLDS ACTUALLY KNOW
“EXPLAIN A.I. TO ME LIKE I’M A 5 YEAR-OLD”
▸ AI is when you make a computer like a little brain.
▸ You help it to learn by giving it a lot of words and pictures
and numbers.
▸ If the computer hears you answer a lot of questions, later
on it can quickly answer your questions.
▸ But it only knows what you show it and tell it, so it’s not as
smart as you are.
10. DATA DATA DATA DATA DATA DATA (PS: PLEASE DON’T BUILD TERMINATORS)
11. WHEN ARE YOU DOING PEOPLE FAVORS AND WHEN ARE YOU NOT?
PEOPLE GENERALLY NEED AND/OR WANT CONTROL
▸ Control requires attention
http://bit.ly/ChrisButlerDesignThinkingForAI
13. NIGEL CROSS’ VIEW OF DESIGN (HUGH DUBBERLY VISUALIZATION)
http://bit.ly/HughDubberlyModelsOfDesign
14. UNDERSTANDING PEOPLE ALSO INVOLVES “DATA” AND “ATTENTION”
TYPICAL STRATEGIES FOR UNDERSTANDING POTENTIAL USERS
▸ Shadowing
▸ Observe people using products while they do things
▸ Behavioral mapping
▸ Photograph people in a particular space over a few days
▸ Consumer journey
▸ Track all the interactions someone has with a product/service/space
▸ Camera journals
▸ Ask users to keep a visual diary of their activities and impressions with a product
▸ Extreme user interviews
▸ Talk to people who know ABSOLUTELY NOTHING about the product/service and what they experience
▸ Storytelling
▸ Ask people to tell you personal stories about their experiences around a product/problem/etc
▸ Unfocus groups
▸ Interview as diverse a set of people together as possible (ps: I HATE focus groups, but these can be more interesting)
15. THIS IS WHAT IDEO ACTUALLY DID FOR INNOVATING AROUND “SANDALS”
GET PEOPLE TALKING
▸ An artist
▸ A bodybuilder
▸ A podiatrist
▸ A shoe fetishist
16. CENTAURS CAN’T WEAR NIKE AIR MAGS
EMPATHIZING MEANS KNOWING USERS AND OTHERS
17. RICHLY TEXTURED CHARACTERS ARE SWELL, STOCK STEREOTYPES AREN’T
GET RID OF PERSONAS
▸ Or rather, we get rid of fictional
personas that do not do justice to
the complexity of human beings
▸ AI-first design means looking at
the data
▸ Which can find us more interesting
people to imagine and consider
19. IF I CAN GET ACROSS ONE THING…
PRIOR PROBABILITY
▸ Look around the audience.
▸ Raise your hand if you think that
there are no left-handed people in
this audience.
▸ Stand up if you think 90% of the
audience is left-handed.
▸ (You probably are calculating this,
you’re just going on intuition, fueled
by past experience)
20. YOU START WITH LESS INFORMATION BUT YOU GET MORE AND MORE
PRIORS LET YOU UPDATE YOUR EXPECTATIONS
▸ Wikipedia tells us that about 10% of the population is left-
handed.
▸ If I started making bets with you about how many people here
were left-handed, wouldn’t you want to incorporate that
information?
▸ If I told you that there were actually a lot of right-handed fascists
roving the area, wouldn’t you use that information?
▸ We adjust our expectations as we get more information. The
probabilities shift.
22. (MY PHD WAS IN LINGUISTICS)
GOOGLE KNOWS A LOT MORE ABOUT ME: SO THEY SHOW ME
ALMOST NOTHING ABOUT BIOLOGICAL MORPHOLOGY
23. (FULL DISCLOSURE: I HAVEN’T SEEN THIS FILM)
“WJATS THE AREA OF POLAND”
▸ Bing corrects my spelling and says
“120,728 sq miles (312,686 km^2)”
▸ But also (for me, a Californian)
“About equal to the size of Nevada”
▸ Accuracy + intuitiveness
▸ Machine + human
▸ “Optimal” is dependent upon
perspective
24. ADJUSTING PROBABILITIES WITH NEW INFORMATION = FEEDBACK LOOPS
FEEDBACK LOOPS = PERSONALIZATION
▸ If your system doesn’t learn, it’s not an AI
system
▸ If it does learn, that means it’s not picking up
PERSONA SOPHIE, it’s picking up lots of
different archetypes
▸ A lot of these systems are about figuring out
how to
▸ Get people interested (acquisition)
▸ Give people the right stuff (upsell)
▸ Stop people from leaving (churn reduction)
▸ BZZZZZZZTTTTTTTT! NO!
25. (PS: IT’S NOT ALWAYS APPROPRIATE TO HUG YOUR USERS)
AI SYSTEMS USUALLY MEAN RELATIONSHIPS
▸ Once you are in the realm of
collecting and using data in a
personalization kind of way,
you are now dealing with
them as individuals
▸ Which means, more than
ever, you are in the area of
relationship creation and
management
27. LET’S TAKE THIS METAPHORICALLY
▸ Pee: We consume a lot of information,
we know give off a lot of information,
too.
▸ How do we help users manage
where it all goes?
▸ Hay: Human torsos/mouths are not
great for grazing.
▸ We don’t always want what we
want.
▸ Freak: Traditionally, centaurs are
considered noble or wild, rarely
ashamed.
▸ But this can’t always be the case.
We are all social animals. How do
others see us is a lot of how we
see ourselves.
28. NOT ALL CENTAURS AND CYBORGS ARE RELUCTANT
Automate work, make room for whimsy
(http://bit.ly/CentaurSketches)
You can identify as a cyborg
(http://bit.ly/IdentityAndJuliana)
31. REPRISE
THE PARADOX OF AUTOMATION
▸ In critical systems, the times when humans need control require MORE skill (not less)
▸ So you want to show wrong things and random things for machine learning AND
human practice
32. WHAT DO YOU ASK WHO DO TO, IN WHAT WAY, WHEN?
AUGMENTING HUMANS BY DIRECTING THEIR ATTENTION
▸ Machine learning and other techniques give you new ways
to direct your attention and your users'.
35. WHAT DO YOU ASK WHO DO TO, IN WHAT WAY, WHEN?
AUGMENTING HUMANS BY DIRECTING THEIR ATTENTION
▸ Machine learning and other techniques give you new ways
to direct your attention and your users'.
▸ But that involves balancing when it's best to surface the
thing someone almost certainly wants and when you help
them explore new corners of the world.
36. RECOMMENDATIONS AND HOW THEY COME ABOUT
▸ “Silence of the Lambs” would be right among comedies if people who loved Mel Brooks parodies really did love
cannibalism
▸ It could be right if Netflix were trying to learn something it doesn’t know or isn’t sure about
▸ A children’s show as a match for a show about a heroic serial killer is probably just a bad string matching thing
37. WE HAVE COME TO ASSOCIATE FILTER BUBBLES WITH RECENT FACEBOOK,
BUT THEY’VE BEEN AROUND FOR A WHILE
FILTER BUBBLES: BOOKS ON AMAZON, 2008
http://bit.ly/BooksAsFilterBubbles
38. YOU CAN HELP BUILD A BETTER WORLD
HOW AI/MACHINE LEARNING WORKS
▸ By default, the data we use represents
biases that are present in the world
▸ So by default, an AI system will
reproduce the status quo
▸ “To design is to devise courses of action
aimed at changing existing situations into
preferred ones” (Herbert Simon)
▸ Because feedback loops move faster now,
we have greater responsibilities than
merely acquire/upsell/keep
▸ See also
▸ http://bit.ly/EthicsEverybodyElse