Exploration of how to apply game design principles to invite more voices into the design and use of machine learning / data science systems. Picture credit: https://unsplash.com/photos/gFFhJPuERII.
Three examples of building for play in data science.
1. Three examples of building for play in data science.
Sam Pottinger
UC Berkeley
2. Tools change the way we think.
What something makes easy (“at hand”) versus what it makes hard to see the contours of a problem. What is emphasized gets attention. What is
hidden is ignored. Deciding how we see and interact with data is a position of power.
3. Games provide a good analogy for data visualization.
Often tools in data science present a “right” way to see the data and a
“correct” narrative to be constructed.
Our most common tools (papers and presentations) present the strongest
form of this: there is a pathway from which one cannot deviate.
Alternatives sometimes cannot be considered without the author’s
intervention.
Creating “safe” spaces for exploration that open up into increasing levels of
complexity is something games are quite good at. They are thinking about
how to make media flexible. Of all forms of HCI, they think about the
player as a partner in constructing the experience.
4. Let’s play!
I have three small demos to share today that explore some different concepts:
● Spaces which push users forward without giving an explicit narrative.
● Spaces which allow us to explore a problem without prescribing the
solution.
● Spaces which empower the user to play with the framing a problem
and the pathway taken to reach conclusions.
5. 1: Play with discovery
This is one of the more constrained
visualizations we will consider but it tries to
open up through a narrative without explicit
messaging.
This is something games explore as well: how
do we encourage users through an experience
dictating all player actions?
We will do this one together as a group.
https://gleap.org/static/special/podcast_viz/i
ndex.html
Game: Sight lines, mechanics
recontextualization.
Data: Multi-dimensionality
6. 2: Play with solutions
Often data science asks us to come up with
answers: models which optimize something, a
recommendation for an action to be taken, a
conclusion for what drove a certain event.
However, what happens when input is needed?
When it’s better to not dictate solution?
This next demo asks the user to find the best
outcome with a computer providing context on
what different proposed solutions.
https://foodsimsf.com
I’ll let you play this one before we come back
together as a group. How could a computer
solve this on its own? What is lost by a
computer doing that autonomously?
Game: Gameplay loop
Data: Participatory design
7. 3: Play with framing
A data scientist often has to determine how
best to frame a question and, in the process
making decisions, about what phenomena to
focus on and which to exclude. Could we
enable users to do the same?
We will switch gears and you will explore
this one for about 5-10 minutes before we
share observations.
https://incomegaps.com
This explores ideas around building
competency and introducing mechanics over
time as the user as more context. In this case,
this allows the user to explore the different
ways that a question can be framed.
Game: Progressive disclosure.
Data: Assumptions manipulation.
8. These slides are available under the CC-BY-4.0 License. All works presented are mine (A Samuel Pottinger). The
following images were also used under the following licenses:
● gFFhJPuERII by Brandon Romanchuk under the Unsplash License.
● Lkx4GfCYdQI by Charles Deluvio under the Unsplash License.