Through our recent Design in AI survey we found that on AI projects there is frequently a lack of alignment between technical and non-technical team members. During this talk, we will share the results of our report and then talk about specific methods to build alignment. You will learn how two of our favorite workshops, Empathy Mapping for the Machine and Confusion Mapping, can build stronger teams and better products. You will walk away with a better idea of the nuances required in product and design practice for AI systems.
by Chris Butler (Director of AI, Philosophie)
at NYAI #18: AI & UX on Tues, 27 Feb 2018 at Capital One Labs
nyai.co
FULL ENJOY 🔝 8264348440 🔝 Call Girls in Diplomatic Enclave | Delhi
NYAI #18: Team Alignment for Human-Centered AI (Chris Butler - Director of AI, Philosphie)
1. NYAI Team Alignment for Human-Centered AI
chrisbutler@philosophie.is
@chrizbot
https://goo.gl/aqpnq8
2. NYAI Team Alignment for Human-Centered AI
● How do AI teams work together today?
● How could they work better together?
● What are techniques that can better build
alignment for AI projects?
8. NYAI Team Alignment for Human-Centered AI
High Impact:
Data availability
Data cleanliness
UX/UI design
9. NYAI Team Alignment for Human-Centered AI
Controversial:
False positive/negative
Latest research
Worst case outcomes
10. NYAI Team Alignment for Human-Centered AI
Unknown:
Trust
Bias
Non-deterministic behavior
11. NYAI Team Alignment for Human-Centered AI
DATA
Cleaner data, more widely available
data, free data, and more data are
needed to build AI projects.
CASE STUDIES
Examples and resources for successful
projects is invaluable to those in this
new and unknown space.
COMMUNICATION
Methods for sharing from the
technical side to non-technical like
education, case studies, and
interpretable information.
12. NYAI Team Alignment for Human-Centered AI
- Survey Participant
“Engineers aren't allowed
to use AI terms.”
14. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• Stolen from design thinking.
• Path towards personas.
• Collect, as a team, everyone’s understanding and assumptions.
• For the machine, it is about how it interacts with humans, not
personifying the machine.
15. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• Creating a shared model of what it needs to do.
• Create consensus with viewpoints from all team functions.
• Answer important questions about what it needs to do in the
context of your product.
17. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• How does it interact with the world?
• Actions they could take, things it should generate, or
recommendations should make.
• Example: “automatically assign a field tech to a job when it is
ready” which helps the dispatcher by automatically doing
things that are necessary to get the job ready for assignment.
18. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• What does it need to understand about the world?
• Video, images, information, history, and other context they
need to do they work they need to do.
• Example: “tech’s truck inventories” which will allow the system
to make recommendations on who is available to take jobs (and
qualified, etc.).
19. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• How does it build trust?
• Communication with the humans they work with to
expose why they are making the decisions they do.
• Example: “shows the reasons for assigning a tech” which builds
trust with the dispatcher on why the system is doing
something.
20. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• What is the starting point for understanding?
• Heuristics that are learned from people doing the job today
that are less fuzzy or regulations that it needs to adhere to.
• Example: “don’t assign techs that aren’t certified” to be within
regulation requirements.
21. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• What do we think is ‘good’ and ‘bad’ performance of it?
• Outcomes that are desired or should be avoided when
training or providing rewards to the machine.
• Example: “bad: return trips” which identifies a key error of
wasteful return trips by the field tech.
23. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
Draw the grid on a board (robot face optional)
Announce each section and ideate privately on post its
Place ideas on the board while deduping
Repeat for each section (x5)
Dot vote most important concepts across all sections
25. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• You have a prioritized set of needs for the machine you would
want to prototype or build.
• Which data to collect/clean and outcomes to measure.
• Use this consensus as momentum to start building.
27. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• From data science and statistics.
• Visualization of the performance of
an algorithm.
• False positives and false negatives
help understand how it fails.
29. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• Non-deterministic systems fail sometimes and we need to know
what that means for the people using them.
• Accuracy is not a reliable metric for the real performance.
• Help discover potential worst case scenarios.
• Start on the discussion about what needs to be
tested/simulated.
31. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
Draw the grid on a board
Privately ideate the false-positive and false-negative sections
Optional: also ideate true positive and true negative
As a group, place on board while deduping
Repeat for each section (x2)
Stack rank severity and impact of problems in false-positive and
false-negative quadrants
32. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• You now have a set of important worst case situations to
consider when implementing your AI/ML.
• Discussion around why certain worst case outcomes are the
most important to the end users.
35. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• Stolen from service design.
• Each person assumes a particular role and behavior.
• Can use costumes or props.
• Board games are another way too (http://iotservicekit.com/).
36. PAPIS.IO 2017 EMPATHY MAPPING FOR THE MACHINE
• Interactions between people and machines are important to
understand and expose issues.
• Feedback is visceral because you are experiencing it.
39. NYAI Team Alignment for Human-Centered AI
Create roles for humans and machines
If you have personas or empathy maps this is a good starting point
Give out roles to each person and let them study - no sharing yet!
Give prompt and start roleplaying
Retro as a team
40. NYAI Team Alignment for Human-Centered AI
• Empathy (or at least compassion) for the people that will be
interacting with your services.
• List of notes on what worked, what didn’t, and what you need
to improve.