7. • “What
is
happening?”
• “Why
is
it
happening?”
Not
much
problem
there.
iksinc@yahoo.com
Current BI tools are good at answering:
8. However,
there
are
inherent
dangers
in
manually
searching
for
rela0ons/pa3erns/trends
in
charts/dashboards.
Analysts'
biases
are
unavoidable;
A
study
at
Bayer
about
10
years
ago
found
that
70%
of
analysis
could
not
be
replicated
by
changing
the
analyst.
iksinc@yahoo.com
9. Gartner
defines
prescrip0ve
analy0cs
as:
“…the
applica0on
of
logic
and
mathema0cs
to
data
to
specify
a
preferred
course
of
ac0on.
While
all
types
of
analy0cs
ul0mately
support
be3er
decision
making,
prescrip0ve
analy0cs
outputs
a
decision
rather
than
a
report,
sta0s0c,
probability
or
es0mate
of
future
outcomes.”
AI
iksinc@yahoo.com
11. Predic?ve vs. Prescrip?ve Analy?cs
Predic0ve
analy0cs
is
simply
focused
on
the
outcome
–
good
for
a
sports
be3or
Prescrip0ve
analy0cs
is
what
the
coaching
staff
needs
iksinc@yahoo.com
13. What Characterizes Intelligence
• Ability
to
interact
with
real
world
• To
perceive,
understand,
and
act
• Searching
the
best
solu0on
• Reasoning
and
planning
• Modeling
the
environment
• Solving
new
problems,
planning,
and
making
decisions
• Ability
to
deal
with
uncertain0es
• Learning
and
adapta0on
iksinc@yahoo.com
14. What is AI?
• The
term
ar0ficial
intelligence
was
coined
by
John
McCarthy
circa
1956.
He
defined
it
as
“the
science
and
engineering
of
making
intelligent
machines”
Ar0ficial
intelligence
is
technology
that
appears
to
emulate
human
performance
typically
by
learning,
coming
to
its
own
conclusions,
appearing
to
understand
complex
content,
engaging
in
natural
dialogs
with
people,
enhancing
human
cogni0ve
performance
(also
known
as
cogni0ve
compu0ng)
or
replacing
people
on
execu0on
of
non-‐rou0ne
tasks.
Gartner
Defini0on
iksinc@yahoo.com
15. Weak AI
• Also
known
as
Narrow
AI
• a
descrip0ve
term
used
for
AI
that
can
demonstrate
human
like
intelligence,
but
only
for
a
specific
task
or
tasks.
Majority
of
today's
AI
systems
fall
in
this
category.
iksinc@yahoo.com
16. Ar?ficial General Intelligence (AGI)
• Also
known
as
Strong
AI
• a
term
used
to
describe
a
certain
mind-‐set
of
ar0ficial
intelligence
development.
Strong
AI’s
goal
is
to
develop
ar0ficial
intelligence
to
the
point
where
the
machine’s
intellectual
capability
is
func0onally
equal
to
a
human’s.
iksinc@yahoo.com
17. Ar?ficial Super Intelligence (ASI)
• A
term
used
for
AI
of
the
future.
It
will
be
be
superior
to
any
level
of
human
intelligence
and
will
(poten0ally),
if
allowed,
be
in
complete
control
of
its
own
decision
making.
“It
seems
probable
that
once
the
machine
thinking
method
had
started,
it
would
not
take
long
to
outstrip
our
feeble
powers...
They
would
be
able
to
converse
with
each
other
to
sharpen
their
wits.
At
some
stage
therefore,
we
should
have
to
expect
the
machines
to
take
control.”
Alan
Turing,
the
'godfather
of
AI'
from
Nick
Bostrom’s
latest
book:
‘Superintelligence:
Paths,
Dangers,
Strategies
iksinc@yahoo.com
21. Learning in AI
• Most
domina0ng
subfield
of
AI
today.
Machine
learning
is
concerned
with
making
computers
learn
to
make
predic0ons/
decisions
without
explicitly
programming
them.
Rather
a
large
number
of
examples
of
the
underlying
task
are
shown
to
op0mize
a
performance
criterion
to
achieve
learning.
• Two
major
styles
of
machine
learning:
Supervised
and
unsupervised
iksinc@yahoo.com
23. Supervised Learning Models
• Classifica0on
models
• Predict
whether
a
customer
is
likely
to
be
lost
to
compe0tor
• Tag
objects
in
a
given
image
• Determine
whether
an
incoming
email
is
spam
or
not
iksinc@yahoo.com
24. Supervised Learning Models
• Regression
models
• Predict
credit
card
balance
of
customers
• Predict
the
number
of
'likes'
for
a
pos0ng
• Predict
peak
load
for
a
u0lity
given
weather
informa0on
iksinc@yahoo.com
25. Unsupervised Learning
• Training
data
comes
without
labels
• The
goal
is
to
group
data
into
different
categories
based
on
similari0es
Grouped
Data
iksinc@yahoo.com
26. Unsupervised Learning Models
• Segment/
cluster
customers
into
different
groups
• Organize
a
collec0on
of
documents
based
on
their
content
• Make
Recommenda0ons
for
products
iksinc@yahoo.com
27. Deep Learning
• It’s
a
subfield
of
machine
learning
that
has
shown
remarkable
success
in
dealing
with
applica0ons
requiring
processing
of
pictures,
videos,
speech,
and
text.
• Deep
learning
is
characterized
by:
• Extremely
large
amount
of
data
for
training
• Neural
networks
with
exceedingly
large
number
of
layers
• Training
0me
running
into
weeks
in
many
instances
• End
to
end
learning
(No
human
designed
rules/features
are
used)
iksinc@yahoo.com
28. Examples
of
Deep
Learning:
Object
Detec0on
and
Labeling
iksinc@yahoo.com
29. Examples
of
Deep
Learning:
Automa0c
Descrip0on
Genera0on
of
Images
iksinc@yahoo.com
30. Example
of
Deep
Learning:
Predic0ng
Heart
A3acks
iksinc@yahoo.com
31. Natural Language Processing &
Speech Recogni?on
• NLP
&
speech
recogni0on
are
those
subfields
of
AI
that
make
it
possible
for
machines
to
communicate
with
humans
by
understanding
wri3en
or
spoken
text
• The
text
could
be
structured
or
unstructured.
• These
two
subfields
of
AI
are
finding
many
applica0ons
in
the
industry
to
build
new
UIs
that
are
proving
more
effec0ve.
•
Alexa,
Cortana,
Siri
are
all
examples
of
these
AI
technologies.
• IBM
Watson
is
another
example
of
using
NLP
to
assist
in
evidence
based
medicine
iksinc@yahoo.com
35. Named-‐En?ty Recogni?on
• It’s
a
subfield
of
NLP;
Named
En0ty
Recogni0on
(NER)
labels
sequences
of
words
in
a
text
which
are
the
names
of
things,
such
as
person
and
company
names,
or
gene
and
protein
names.
• Helpful
for
automa0c
informa0on
extrac0on
to
build
rela0onships
between
different
en00es.
Think
of
Jeopardy.
iksinc@yahoo.com
36. Op?miza?on and Planning
• Lots
of
overlap
with
opera0ons
research
• AI
centric
op0miza0on
methods:
• Gene0c
algorithms
• Based
on
natural
selec0on
in
a
popula0on
• Simulated
annealing
• Based
on
crystal
forma0on
in
solids
through
cooling
iksinc@yahoo.com
37. Op?miza?on and Planning
• Ant
colony
op0miza0on
• Based
on
how
ants
leave
markers
for
other
ants
• Par0cle
swarm
op0miza0on
• Based
on
behavior
of
the
flock
of
birds,
pool
of
fishes
etc.
iksinc@yahoo.com
42. ORION,
an
acronym
that
stands
for
On-‐Road
Integrated
Op0miza0on
and
Naviga0on,
is
perhaps
the
largest
commercial
analy0cs
project
ever
undertaken.
It’s
required
well
over
a
decade
to
build
and
roll
out,
and
more
than
$250
million
of
investment
by
UPS.
Savings
in
driver
produc0vity
and
fuel
economy:
$300
-‐
$400
million
a
year,
100
million
fewer
miles
driven
and
a
resul0ng
cut
in
carbon
emissions
of
100,000
metric
tons
a
year.
iksinc@yahoo.com
43. • Vetride
system
is
an
another
example
of
prescrip0ve
analy0cs
developed
by
a
local
company
• The
system
is
being
used
at
64
VA
sites
all
over
USA
and
is
installed
approximately
in
1200
vehicles
• Provides
veterans
transporta0on
at
demand
op0mizing
a
number
of
parameters
with
many
constraints
iksinc@yahoo.com
46. Some Poten?al Sugges?ons
• Candidate
task
characteris0cs
for
ini0al
AI
projects:
• Complexity
level
:
low
to
medium
• High
volume
and
repe00ve
• No
legal
or
ethical
risks
• Fair
level
of
user
interac0on,
internal
or
external
iksinc@yahoo.com
Whit
Andrews,
Gartner
VP
Michael
Azoff,
Ovum
Principal
Analyst
47. Three Phase Process
• Prepara0on
phase
• Structure
preparedness
• Organiza0onal
preparedness
• Knowledge
preparedness
• Buy-‐in
and
value
crea0on
• Create
awareness
and
value
proposi0on
• Brainstorm
project
ideas
• Demo
value
• Organiza0on
wide
adop0on
iksinc@yahoo.com
49. Summary
• AI
is
set
to
play
a
big
role
in
businesses
across
a
wide
spectrum
• Tune
out
the
hype
and
focus
on
how
you
can
ini0ate
a
low
risk,
low
complexity
project
to
get
started.
• Be
op0mis0c,
that
is
an
AI
trait
iksinc@yahoo.com