This siled was prepared for the training seminar on Artificial Intelligence for International Organizations. Introducing AI technologies into International Development fields for achieving SDGs would be great opportunities to accelerate development. . This material is just explaining basic of AI and some examples of AI application in this field.

1. The Basics of
IF Inc. , CEO
Kobe Institute of Computing
Visiting Professor
小塩篤史 Atsushi Koshio
koshio@i-f.co.jp

2. Biography Prof. Atsushi Koshio
IF Inc. CEO
Kobe Institute of Computing Visiting Professor
Based on the backgrounds including future studies, data science, artificial intelligence, technology
management and so on, I propose a solution that utilizes field knowledge and state-of-the-art research
against future issues. IF Inc. conduct artificial intelligence and new business development that contributes
to solving future society's problems, and at the Graduate School of Project Design, I have a role on
training people with flexible ideas for the future. Currently, I also operate I cube Co., Ltd. that promotes
intellectual IoT, HyperCube Co., Ltd. aiming to prevent dementia by game, etc., and is engaged in AI
related consulting at Asian Development Bank. Past works include the development of a decision support
system for hospitals, the development of cloud-based electronic medical record systems, projects for
medical information data analysis, development of regional medical information systems, development of
image analysis systems for dentistry with AI, etc. In the field of research, I am involved in constructing
methodology for future studies, developing innovation emergent methods, developing innovative digital
marketing methods, collective intelligence systems, and carried out collaborating research on
management improvement by AI and simulation and innovative digital marketing with MIT Sloan School
of Management
■Current Position：
・IF Inc. , CEO
・Kobe Institute of Computing, Visiting Professor
・I cube Co., Ltd, CEO
・HyperCUBE Co., Ltd, CIO
・Every Plan Co., Ltd Lead Advisor
・Social Healthcare Design Inc. Advisor
■Social Activity：
・Asian Development Bank, High Level Technology Lead Specialist on AI
・Member, CODATA Sub-Committee, Japan National Committee
・GESTISS Board member
・Future Healthcare Research Institute, Board Member
・Lecturer at Nagoya Institute of Technology, Shinshu University, Nippon Medical School,
Meiji Gakuin University and so on.
■Professional Experience：
・Graduate School of Project Design, Former Dear and Professor
・Massachusetts Institute of Technology, Sloan School of Management Visiting Scholar
・Nippon Medical School, Department of Health Policy and Management, Assistant Professor
・The University of Tokyo, Policy Alternatives Research Institute, Researcher
And others
■Writings：
・Vignettes and differential health reporting: results from the Japanese World Health
Survey. 2017 Cadernos de Saúde Pública
・Projecting Long Term Care Needs in Japan? Microsimulation modeling for super aged
society, 2013 International Microsimulation.
・Application of social simulation methodology for health policy & management –
investigating change of healthcare demand and future healthcare delivery system. 2012
Journal of Japanese association for Medical Informatics
■Education：
・The University of Tokyo, Graduate School of Frontier Science Ph.D course
・Massachusetts Institute of Technology, Sloan School of Management, Massachusetts
Institute of Technology, Sloan School of Management Visiting Studetn
■Trusted Records：
・Government Agency including MHLW, MEXT, METI and local goverments
・Development of artificial intelligence, business strategy and new business of leading
manufacturers, major logistics, major telecom operators etc.,

3. AI is everywhere
• Amazon,Netflix Recommendation
• Smart Phone
• Smart Speaker
• Self driving car
• Translation Machine
There will be more and more in our life.

4. Artificial Intelligence
Foundation
Application
Game
Human
interface
Image recognition
Voice
recognition
Neural net
Expert system
Sensory
processing
Natural language
understanding
Search
Reasoning
Knowledge expression
Machine learning
Genetic
algorithm
Multi agent
Information
retrieval
Planning
Robot
Data mining
Machine Learning changed structure of AI fields.
Application research & development now leads all field

5. 1st AI Booms(Search/Inference)
2nd AI Booms(Knowledge)
3rd AI Booms(Machine Learning)
(1956-1960s)
(2012-)
(1980s)
■The rush of big AI projects
(Expert System)
2nd AI Winter(‘74-’80)1st AI Winter(‘74-’80)
■The birth of AI[‘56]
(Dartmouth Conference)
■Games with clear goals
can be solved but not
practical
■Deep learning surpassed
conventional methods in
image recognition[‘12}
■Far from the knowledge
level of human experts

6. Democracy of Machine Learning
Data Processing
DNN Big Data GPU
Cloud
Data Input
Deep Learning has been promoted by DNN, Big Data, and GPU
and has been commoditized thanks to cloud network.
AI Output
AI is no longer a luxury item

7. What’s AI？
• Artificial Intelligence (AI), sometimes called
machine intelligence, is intelligence
demonstrated by machines, in contrast to the
natural intelligence displayed by humans and
animals. Computer science defines AI research
as the study of "intelligent agents": any device
that perceives its environment and takes
actions that maximize its chance of successfully
achieving its goals. Colloquially, the term
"artificial intelligence" is used to describe
machines that mimic "cognitive" functions that
humans associate with other human minds,
such as "learning" and "problem solving

8. Learning
Learn over time
without human
InterventionDeep Learning
Reinforcement Learning
Cognition
Form conclusion
with imperfect data
matching
prediction
identify best action
Perception
Interpret meaning of data
including text, voice, images.
image recognition
speech recognition
Natural Language Processing
What can AI do？

9. Traditional Software
development
Machine Learning
Algorithm
Output
Imput
Algorithm
Output
Imput

10. Inference System
System to obtain new results
Based on knowledge
Example
Result
Data
A ⇒ Bson
B ⇒ Cson
Knowledge
Rule
son
⇒
son
⇒
Grand son
A Grand son
C
Constructing knowledge & Rule is
Key components for Inference System
Learning System
System to obtain knowledge
From Data
Example
Data
A ⇒ B
B ⇒ C
A ⇒ C
Knowledge son
⇒
son
⇒
Grand son
Automating knowledge creation,
sometimes Human can not understand
knowledge generated by machine

11. 11
What is breakthrough in deep learning?
Learn what to learn
Overwhelming improvement of accuracy +
reduction of operation cost
Raw data Feature extraction
●Machine Learning：Decide what to learn by human
●Deep Learning：Learn what to Learn
Cat Face
Learning
Feature extraction~LearningRaw data
Test
：Cat
：Dog
Test
：Cat
：Dog

12. What AI can do now
• Human intellectual activity
• “Perception" → “Cognition" → “Inference" →
“Decision"
• Today's AI can be automated with high accuracy on
“perception” and "cognition“ in rich data situtation
• Perception: see, hear, smell, feel change
• Cognition: match, predict, explore
• What if you have eyes or ears that can predict near
future move for 24 hours?

13. Flood Control GIS
Prediction of rain
by satellite images
IoT
Grasp of rainfall
in the basin
Monitoring water
level of the river
dom flow control
by AI
Optimizing
discharge volume
based on flood control
and
electricity generation
hydroelectricity
satelyte

14. Needs Seeds
Automation
Speech recognition
Face recognition
Computer Vision
Optimization
Diagnosis
Natural Language Processing
Example • Finance for vulnerable
• Air Quality Monitoring
• Smart Energy
• Water resource Management
and agriculture
• AI Education
• Health Screening
Needs Seeds matching for AI application planning

15. Advantages of Machine
• RAAJI is Chatbot for
education on
reproductive health in
Pakistan
• People can talk more
privacy sensitive
information to machines
• Working 24hours
• Machine treat people
equally
• Easily scalable

16. Some Cons
• Privacy issues
• Used for discrimination
• Malicious Use
• Wrong Prediction
Ethical and Risk conscious development is
necessary

17. Now is the best timing for
introducing AI
in International Development
• Data can be used for analysis is drastically increasing
in quality and quantity by digitalization, IoT and
Space development
• Algorism for data processing had been already
commoditized by big AI players
• GPU as infrastructure for data processing is built in
cloud and easy to develop Prototype of AI
There are many issues with high affinity with AI in SDGs

18. Materials

20. Future of Employment
47% of present work can be computerized

21. AI R&D timetable
1947
Alan Turing proposes
the concept of AI
1979
Kunihiko Fukushima
introduces theory of
Neural Networks
2006
Breakthrough in Deep
Learning at the
University of Toronto
2018
Now Deep Learning
is commonly used
infrastructure for
ICT
1946
ENIAC, one of the
world’s first computer
50’s 60’s
First AI Boom
The age of Reasoning
Prototype AI developed
80’s 90’s
Second AI Boom
The age of
Knowledge
Representation
Systems able to
reproduce human
decision-making
2010 to now
Third AI Boom
Machine learning and
deep learning
More accurate
pattern recognition
by less cost

22. 22
What is breakthrough in deep learning?
Learn what to learn
Overwhelming improvement of accuracy +
reduction of operation cost
Raw data Feature extraction
●Machine Learning：Decide what to learn by human
●Deep Learning：Learn what to Learn
ネコの顔
斜め線
Learning
Feature extraction~LearningRaw data
Test
：Cat
：Dog
Test
：Cat
：Dog
職人芸になり属人的でスケールしないし、精度も低い。
人工知能史上最大の
ブレークスルー

23. 23
What’s Deep Learning？？
Only
Entering
Pixel data
Final
Decision
より俯瞰的な状況把握

24. Quantitative
Data
Log Data of
Digital Device
Image
Voice
Natural
Language
Input
Deep Learning
Other Machine
Learning
Algorithms
Natural
Language
Processing
Process
Optimization
Prediction
Perception
Matching
Generation
Output
Implicit knowledge
Five sense
Quantum computing? Reasoning
Creation?Next?
SuitableforanalysiswithAIExpansion of possibility of treating and using dataTraditional
Statistics

25. 25

26. 26
Ethical Issues on AI
Three Laws of Robotics
1.A robot may not injure a human being or, through inaction,
allow a human being to come to harm.
2.A robot must obey the orders given it by human beings
except where such orders would conflict with the First Law.
3.A robot must protect its own existence as long as such
protection does not conflict with the First or Second Laws

27. What’s important points for
applying AI technologies in
practical setting?

28. Purpose Design
Manager
Business Design
Sales
System Design
System engineer
System
Development
Programmer
Answering Why,
What, How of
system development
You need a
Knowledge on
project design
Project Design
EntrepreneurWHY
WHAT
HOW

29. Issue
Purpose
Requirements
for Achievement Current Hurdles
Specific Service
Elimination of
hurdles
Required
Technology
and Resource
Value
Profit Structure
Channel
Customer
Data Holders Data
Data Resource
Cost StructureBusiness Model
Overview of AI business design
Resource
finding:
knowledge on
technologies and
yourself
Issue finding :Desk
research, field work
Vision
making :Imagination
and comunication
Requirement
definition :
Logical thinking
and
exploration
problem
definition :
Logical thinking
and
exploration
Service Design :
empathy, creativity
and trials
problem
solving :
Logical thinking
Value statement:
deep insight for
customer
User research:
quantitative and
qualitative research
Stake holder
analysis: system
thinking and
incentive design
Marketing
Communication and
promotion
Business Development:
finding advantage on
operation
Marketing strategy:
promotion and price
design
Accounting :
Costing and Financing

30. Sample of Potential of AI for international development
Smart energy
optimizing energy loss with AI
predicting demand and supply
for preventing Blackout
Energy
Inclusive finance
chatbot with Natural Language
Processing can include people
who can’t read.
Finance
AI health monitor
Monitoring Air Quality by AI
screening Health data
including images and
questioner
Health care
Smart agri
prediction of weather and
Estimating crop yield.
Agriculture
AI teacher
adaptive learning with AI
more people can access to
essential education
Education

31. PETER MARX
(GENERAL ELECTRIC)
described how General is
using machine learning for a
number of different purposes,
including using drones for
inspecting power lines and in
manufacturing.
ELECTRICITY SUPPLY IMPROVING HEALTH
PROF. STUART RUSSELL (UC
BERKELEY)
described how AI is being used
for monitoring verification of
the nuclear test treaty to
distinguish between natural
seismic tremors and shocks
triggered by nuclear tests.
NUCLEAR TEST
PAUL BUNJE (XPRIZE),
described the Global Fishing
Watch which uses simple
machine learning to utilize the
data in fishing vessels and
applies machine learning
algorithms to identify where
the vessels have been and type
of activities they are engaged
in.
GLOBAL FISHING WATCH
According to ERIC HORVITZ
(MICROSOFT), predictive
modelling of cholera outbreaks
can now be developed in advance
based on powerful algorithms
that can be used to distribute
fresh water, or supply vaccines.
AI algorithms can continuous
tracking of clinician’s
movements in hospital,
without revealing who they are.
Sensors in hallways close to
hand-hygiene dispensers with
a deep learning recognition
system monitor clinicians’
hygiene practices, with a
performance better than many
of the state-of-the-art systems,
reducing hospital-acquired
infection ratesーProf. Fei-Fei Li
(Google & Stanford University).
MONITORING HYGIENE
Chris Fabian (UNICEF
Innovation)
described the use of simple AI
in mid-upper arm bands
(MUACs) to monitor the
nutritional status of children,
which are valuable tools for
famine response. Companies
are developing similar tools
using face recognition.
MONITORING NUTRITION
Andrew Zolli (Planet Labs)
described how Planet Lab is
using satellite imagery to
monitor planning, agricultural
indicators in Kenya, a flood
event in Sri Lanka, the growth
of Dar es Salaam city and an
Internally Displaced Persons
(IDP) camp in Uganda.
USING SATELLITES
Labs are embedded within
food containers to control
temperature, food supply,
humidity etc. and use AI black
box optimization to evolve a
process that gets to the best
ingredients to obtain the very
best basil, quantity and
quality-wise, according to
Antonie Blondeau (Sentient
Technologies).
MIT’S OPEN AGRICULTURE

33. Practical Application:
Automation of Dental Field
• There are several software to support
recognition, but no automation tools for
supporting diagnostic imaging in dental field

34. Our target: Cepharometrics
• Cepahrometrics is our fisrt
target. This method is very
traditional and have long
history. It still have been used
in clinical setting globally
• This method has been used
only in orthodontics, but it can
be used by all dentists by
automating
• Plot feature points determined
from the image and make
skeleton diagnosis from the
positional relationship
34

35. Automation of Cepharometorics
using Deep Learning
35
• Feature point extraction by image analysis using Deep
Neural Network

36. Method
• Data for machine learning
• Cepharogram 200
images(female and male, age
20-80)
• 387*480 pixel (1/5 of original
images)
36
• Input data
• A local patch image (70 * 70
pixel) randomly generated from
each image, and a distance
between the target point and the
patch image
• Number of data 60,000
• Cepharo 200 imapse*patch
images generated from each
image 300
• Model
• Input: patch images 70*70 = 4,900
• Intermediate Layer: First layer 1000 , Second Layer
200 . Third Layer 1000
• Output Layer: 2(Distance between predicted point
and point of interest x and y)
Distance
from point of
interest
Pixel Data

37. The way to determined points

38. Evaluation
• Evaluation Indicator
• Prediction of point of Sella (Red mark)
• 𝑟𝑘 = ∆𝑥 𝑘
2 + ∆𝑦 𝑘
2
• 𝑓 𝑟𝑘 = ቊ
1(𝑟𝑘 ≤ 8)
0(𝑟𝑘 > 8)
• N as number of predicted points
• 𝑃𝑧=4.0𝑚𝑚 =
σ 𝑘=1
𝑛
𝑓(𝑟 𝑘)
𝑛
× 100 %
• Method(1):Legacy Machine Learning
• 𝑃𝑧=4.0𝑚𝑚 = 86.5 %
• Method(2):Deep Neural Network
• 𝑃𝑧=4.0𝑚𝑚 = 93.5 %
38
Sella (Red mark) is the most difficult point to detect. The result of other points were
better than Sella.

39. Structure of the system
39
Interaction with
server by sending
images to server and
receive predicted
points
Capturing Image data
by CT
Automation of cepharometrics
through Web application
Generating 2D images
By simulation from
3 D images
Uploading
Image
Sending image with
predicted points
View through Web
And dentists can
adjust points