Do Androids Play Games? Where Does Gamification Fit in a World of Robots and AI, Dr. Michael Wu

twitter: @mich8elwu
linkedin.com/in/MichaelWuPhD
©2018 PROS, Inc. All rights reserved. Confidential and Proprietary.
Do Androids Play Games?
Where Does Gamification Fit in a World of Robots + AI?
Michael Wu, PhD (@mich8elwu)
chief AI strategist @ PROS
2018.11.27

twitter: @mich8elwu
Michael Wu, PhD (@mich8elwu)
chief AI strategist @ PROS
2018.11.27
@mich8elwu

twitter: @mich8elwu
The Concept of AI isn’t New
supersmart
fast ~light speed
remember every detail
work 24/7
always learning
get smarter everyday
never get tired
never complaint
…
etc…

twitter: @mich8elwu
A Very Brief Evolution of
Business Analytics
(Business Intelligence)

twitter: @mich8elwu
An Example
R
Rresistance
buy
sell
50%
sell
50%
buy
descriptive predictive prescriptive
R
R
R
sell

twitter: @mich8elwu
temporal predictive analytics
 trend line
historical data x
Predictive Analytics—Estimate
future datafuture prediction
present
𝑓 𝑥
model

twitter: @mich8elwu
𝑓 𝑥
model
text data
x = tweet, blog,
comment, news
articles, sms…
sentiment
classifier
𝑓 𝑥
(+) positive
(0) neutral
(−) negative
social media
interactivity
x = retweet, like,
reply, share, +1…
influence
algorithm
𝑓 𝑥
influence
score
data that you have data that you don’t have
temporal predictive analytics
 trend line
general predictive analytics
aren’t limited to time domain
 social media examples
• influencer scoring
• sentiment analytics
 other examples?
future predictionhistorical data x

twitter: @mich8elwu
R
Rresistance
descriptive
simplest examples of:
 predictive analytics: trend line
 not limited to forecasts in
temporal domain
 sentiment + influence

twitter: @mich8elwu
Prescriptive Analytics—Optimize
GPS navigation systems
 prescriptive analytics: GPS
temporal domain
general prescriptive analytics
 not limited to prescription of
routes in geo-spatial domain
 can prescribe biz strategies

twitter: @mich8elwu
Prescriptive Analytics—Optimize
GPS navigation systems
do you know other examples?
temporal domain

twitter: @mich8elwu

twitter: @mich8elwu
business intelligence (BI):
passive decision support
human still makes the decision
because traditional analytics
have limited accuracy

twitter: @mich8elwu
Analytics Accuracy
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
analytics results
accuracy of
model
qualitydata
quality
CPU
cycles
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
1950’s Vacuum Tube Computers
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
we already have a theoretical
understanding of simple neural
networks (2-3 layers deep)
we are limited to
simplistic models b/c
of computing power
most data (even if
available) are not
digitized (inaccessible)
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
1950’s Transistor Computers
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
in the next few decades, our
theoretical understanding will
improve linearly, but our data
storage and computing power
will increase exponentially
bytes
kb
Mb
Tb
Pb
Zb
Eb
Gb

twitter: @mich8elwu
1970’s IC + Mainframes
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
in the next few decades, our
theoretical understanding will
improve linearly, but our data
storage and computing power
will increase exponentially

twitter: @mich8elwu
1980’s Microprocessor + PC
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
we almost have enough
CPU power to use the
most advanced neural
network models then
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
1990’s Client-Server + Internet
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
although we can use the most
advanced model, we still
can’t use it with all the
available data due to
practical limitations
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
2000’s Distributed Computing
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
we (in theory) have infinite
CPU power, so this constraint
becomes an economic limit,
and we are now only limited
by data volume
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
2010’s Big Data + GPU
𝑓 𝑥linear
model
complexity
logistic
2nd order
2-3 layer
neural net
deep neural net
(~10s layers)
we can use the most
advanced model and
not worry too much
about data or cpu
constraints
bytes
kb
Mb
Gb
Tb
Pb
Zb
Eb

twitter: @mich8elwu
what happens when humans are always in
agreement with machines predictions?
do humans still need to have the final say?

twitter: @mich8elwu
artificial
intelligence (AI):
automation of decision
+ proper execution of
all subsequent actions
decision
action
data

twitter: @mich8elwu
decision
action
automated
loan origination
algorithmic
trading

twitter: @mich8elwu
What Makes AI Intelligent?
deep
learning machine
learning AI

twitter: @mich8elwu
deep
learning machine
learning AI
V8
engine
car

twitter: @mich8elwu
big data machine learning model
regression, classification, clustering
feedforward neural networks, recurrent
neural network, reinforcement learning,
deep learning, LSTM, generative
adversarial learning, convolutional
neural networks, boosting, bagging,
random forest, decision trees, gradient
boosted decision trees, adaBoost,
Kalman filter, latent Dirchlet allocation,
Dirichlet process, latent semantics
analysis, principle component analysis,
linear discriminant analysis, k-means
algorithm, spherical k-means,
agglomerative hierarchical clustering,
Gaussian mixture model, Gaussian
process, collaborative filtering, etc. …

twitter: @mich8elwu
outcome
big data
feedback
machine learning
analytics results
actions
model
decisions
survey

twitter: @mich8elwu
outcome
big data
feedback
machine learning
analytics results
actions
model
decisions
survey
real-time
automatically tracked
semi-automated

twitter: @mich8elwu
model
outcome
big data
feedback
machine learning
analytics results
actions
model
decisions

twitter: @mich8elwu
model
outcome
big data
feedback
machine learning
analytics results
actions
model
decisions
LEARNING
LOOP

twitter: @mich8elwu
Lee Sedol

twitter: @mich8elwu
Gamification is Really a Big Data Discipline
requires a lot of data:
for feedback/reinforcement
generates a lot of data:
from the driven behaviors
fairly reward players
compare player performance
etc.
drive desired
player behaviorstrack all
player behaviors
understand player behaviors,
intrinsic motivation, cheating
etc.
data
33

twitter: @mich8elwu
Components of Gamification System
user
behavior
behavior
data
rules
engine
gamification
platform
gamification
management
system
behavior reporting,
dashboards + analytics
points, badges,
leaderboards, ranks,
etc.
34
feedback
mechanism

twitter: @mich8elwu
the definition of AI evolves
 AI is whatever machine can do
now that can’t be done before
AI = machine mimicry of certain
aspects of human behaviors
with 2 important characteristics:
 automation: the ability to
automate decisions and/or
subsequent actions
 learning: the ability to learn and
improve its performance with
usage
A Definition of AI

twitter: @mich8elwu
CHANGE
SAME
? ? ? will we adopt AI-based technology?
technology adoption can be
analyzed as a behavior change

twitter: @mich8elwu
- Fogg’s behavior model (FBM):
 3 factors underlying human behavior
 temporal convergence of 3 factors
The Behavior Model for Gamification
37
motivation ability triggeraction
wants can told to
BJ Fogg. 2009. A behavior model for persuasive design. In
Proceedings of the 4th International Conference on
Persuasive Technology pp7

twitter: @mich8elwu
- Fogg’s behavior model (FBM):
 3 factors underlying human behavior
 temporal convergence of 3 factors
 ability = access to required resources
at the moment when you need to
perform the behavior
The Behavior Model for Gamification
38
action
motivation
ability
trigger
activation
threshold
(simplicity)

twitter: @mich8elwu
CHANGE
SAME
? ? ? will we adopt AI-based technology?
absolutely… because simplicity
drives behavior changes!

twitter: @mich8elwu
What happens when Ordinary
“Things” Starts to Learn and Think?

twitter: @mich8elwu
feedback over network
compute distributed over network
the internet
Key Tech Enablers of AI + Internet of Things (IoT)
big data
machine
learning
compute
power
human
decisions
+ actions
Vint Cerf
where did all the big
data come from?
where did
the internet
come from?

twitter: @mich8elwu
big data
feedback over network
compute distributed over network
the internet
Key Tech Enablers of AI + Internet of Things (IoT)
big data
machine
learning
compute
power
human
decisions
+ actions
Vint Cerf Kevin Ashton
the internet of things (IoT)
cheap internet bandwidth
cheap open source hardware:
Arduino or Raspberry Pi

twitter: @mich8elwu
Growth of IoT
source: Mario Morales - IDC

twitter: @mich8elwu
Growth of IoT
50
40
30
20
10
0
billionsofdevices
‘90 ‘92 ‘94 ‘96 ‘98 ‘00 ‘02 ‘04 ‘06 ‘08 ‘10 ‘12 ‘14 ‘16 ‘18 ‘20
1990 world
population
2018 world
population

twitter: @mich8elwu
Architecture of IoT Devices
sensors
local
compute
/storage
network
/internet
cloud
compute
/storage
things internet
• sensor + big data
• light weigh local (edge/fog)
compute/storage
• local network (via wifi mesh
network) + internet
• intensive cloud compute/storage
enables collective learning
if I’m not
home

twitter: @mich8elwu
Top Industries Using IoT

twitter: @mich8elwu
AI + IoT is not w/o Risk

twitter: @mich8elwu
AI + IoT is not without risk

twitter: @mich8elwu
to ensure our own survival
we need to be better humans
AI+IoT is still at its infancy
unlike a human child
• learning is constant, complete
(learns everything) + fast
• never forget
• they’re smarter + will be better
than us in everything we do
they learn from us, humans,
our every decisions + behaviors

twitter: @mich8elwu
We All Have a Mission as Gamification Practitioners
we need to gamify ourselves as
a human race to exhibit better
human behaviors (e.g. integrity,
empathy, compassion, etc.)
so our digital creation (AI and
robots) will have good data to
learn from, just as our children will
have good parents to model after

twitter: @mich8elwu
thank you, q&a,
+ follow me
twitter: @mich8elwu

twitter: @mich8elwu
PROS Headquarters
Houston, Texas
3100 Main Street, Suite
900
Houston, TX 77002, USA
+1-800-555-3548
PROS San Francisco
101 Montgomery Street,
Suite 400
San Francisco, CA 94104,
USA
+1-415-283-3000
PROS Chicago
5215 Old Orchard Road,
Suite 505
Skokie, IL 60077, USA
+1-847-583-8450
PROS Austin
3600 Parmer Lane,
Suite 205
Austin, TX 78727, USA
+1-713-335-5829
PROS Toulouse
Le Galilée
185 rue Galilée
31670 Labège
France
+33 (0) 811 70 78 78
PROS München
Leopoldstrasse 23
80802 München
Germany
+49 (0) 89 24442 3097
PROS Paris
10 Boulevard Haussmann
6th Floor
75009 Paris
France
+33 811 70 78 78
PROS London
4th Floor, East Wing
Communications House
South Street
Staines-Upon-Thames
TW18 4PR
United Kingdom
+44 (0) 1784 777 010
PROS Frankfurt
Frankfurt Herriot’s
2nd Floor, Herriotstraße 1
60528 Frankfurt, Germany
+49 (0) 69 677 330 15
PROS Sydney
The Ark
Level 32
101 Miller Street
North Sydney
NSW 2060
Australia
+61 2 8912 2199
PROS Dublin
Ormond Building
31-36 Ormond Quay
Upper
Dublin 7
Ireland
+1-800-555-3548
PROS Sofia
1 Alabin Str., TELUS Tower
16th floor
Macedonia square
Sofia 1000, Bulgaria
+359 2 958 05 95
Thank you

Do Androids Play Games? Where Does Gamification Fit in a World of Robots and AI, Dr. Michael Wu

More Related Content

Similar to Do Androids Play Games? Where Does Gamification Fit in a World of Robots and AI, Dr. Michael Wu (20)

More from Gamification Europe (6)

Recently uploaded (20)

Do Androids Play Games? Where Does Gamification Fit in a World of Robots and AI, Dr. Michael Wu