© Blu Ltd 139

© Blu Ltd 140

 • Merriam Webster
1. A branch of computer science dealing with the simulation of intelligent
behavior in computers
2. The capability of a machine to imitate intelligent human behavior

• The actual definition of AI is something that even experts

have trouble with…

• More generically in our view:

• AI is a tool to achieve an objective
• The hard part is figuring out that objective or business use case

Source: [Link]

© Blu Ltd 141

 Machine Neural networks Deep learning
Learning Evolutionary algorithms AI tools can use
multiple AI methods.
Voice to text
Natural Language Keyword search e.g., Facial
Processing recognition apps use
Meaning, scripting, translating
Computer Vision and
AI deep learning

Computer Object detection

Vision Optical character recognition

Humanoid
Robotics
Industrial

© Blu Ltd 142

 Wide Data Set and Methods

Massive
computing

AI => “AUGMENTED” INTELLIGENCE

© Blu Ltd 143
 AI theory Computing Storage Data
evolution transport
since WW2 $ $
$

Quantum jump in innovation

On-going AI revolution!

© Blu Ltd 144

 • Machine Learning
– Training a machine to identify patterns and/or predict outcomes

– Often used to find relationships between variables

• Correlation

– Contains many subsets of AI methodologies (Neural Network with Deep

Learning, Evolutionary Algorithm etc.)

© Blu Ltd 145

 • Neural Networks
– Inspired by biological networks in the brain; shares insights from cognitive neuroscience
– Use layers of processing to derive an output
– Each node is connected to other nodes and each connection can be weighted differently

Source: [Link]
Source: [Link]
[Link]
© Blu Ltd 146
 There are many types
of Neural Network
architectures …

It quickly becomes
very complex !!!

Source: [Link]
complete-chart-of-neural-networks-explained-
3fb6f2367464
147
© Blu Ltd 147
 • Facial recognition systems create a ‘map’ of facial features from a
photo/video.
• Once deep learning algorithm understands your ‘face map’, it can be used
in different ways
– Identity verification: compare this face map to a database of faces + identities
to find a match
– Image tagging: e.g. Facebook auto-tags people in your photos
– Image manipulation: ‘beautification’ on selfie apps

© Blu Ltd 148

 Simplified explanation of facial recognition
for identity verification
• A face captured in photo/video
• Facial recognition software draws a
‘boundary box’ around the face to
isolate it

© Blu Ltd 149

 • Software analyzes different points on the
face (e.g. distance between eyes, distance
from forehead to chin).
• Identifies dozens of facial ‘points’ that are
important for telling one face apart from
other faces.
• The face is reduced to a ‘facial signature’
(mathematical formula)

Image source: [Link]/news/93731/facial-

© Blu Ltd recognition-technology-vision-box/ 150
 • Different software may look at different
points and a different number of points
• The face is reduced to a ‘facial signature’
(mathematical formula)
• Based of the facial signature many items
can also be calculates such as in this case:
– Gender
– Age
– Face Orientation

© Blu Ltd 151

 • Facial signature is compared to a database of faces (e.g. Facebook,
company or government database) to find a signature match.
• Matches are based on probability

© Blu Ltd Image source: [Link]/articles/how-accurate-is-facial-recognition 152

 • Airport & corporate security
• Smartphone companies & apps:
Unlock phone or approve an action
• Social media: Facebook tags
people with 98% accuracy
• Marketing & advertising: data on
who (age, gender) walks into a
store or spends time in a certain
area is useful info for marketers

Source: [Link]
[Link]
© Blu Ltd 153
© Blu Ltd Source: [Link] 154
china-beaten-nose-job
 Meitu’s Mission
• “to let everyone become beautiful easily”

Meitu’s BeautyCam AI Portrait Feature

• Low-lighting selfies
• Blurry selfies
• Skin texture correction

Source: [Link]

© Blu Ltd 155

 Meitu’s BeautyCam AI Portrait Feature
• Blurry Selfie correction
• Left to Right
– Original
– Regular Beautification
– AI Portrait

Source: [Link]

© Blu Ltd 156

 Meitu’s BeautyCam AI Portrait Feature
• Low-lighting selfie correction
• Left to Right
– Original
– Regular Beautification
– AI Portrait

Source: [Link]

© Blu Ltd 157

 Meitu’s BeautyCam AI Portrait Feature
• Skin texture correction
• Left to Right
– Original
– Regular Beautification
– AI Portrait

Source: [Link]

© Blu Ltd 158

 Class
1. Any thoughts on the Meitu use case? Discussion
2. Based on data Meitu collects, what are other potential
revenue streams Meitu could utilize?

© Blu Ltd 159

 Meitu and Dior launched the ‘AI Make Up
Artist’ feature on the Meitu app
• Augmented reality (AR) ‘virtual try on’
for makeup
• Teaches people how to apply makeup
step by step (in real life)
– AR markers on users face to guide them

Source: [Link]/posts/michel-chi-pin-lin-05bb0186_meitu-dior-
© Blu Ltd meitueve-activity-6844647887712993280-vtOJ 160
 • Deep learning for facial skin analysis
– Machine captures and analyzes 98% of
the face in 12 seconds
– Used in department stores, beauty
salons, etc.

• Uses skin research data from clinical

research
Source: MeituEve Panoramic AI Skin Analysis Machine
– Partnership with Shanghai Skin Disease [Link]/watch?v=6L8q1CTNW3E
Hospital, others

• Diagnoses many skin issues and

recommends skincare actions
• Benefit: data monetization opportunity
for Meitu
– Their photo app collects lots of face data
© Blu Ltd 161
 “The Meitu AI Open Platform is an AI service platform launched by
Meitu to focus on core fields that include human facial technology,
human body technology, image recognition, image processing, image
generation, and others. This provides users with market-proven
professional AI algorithm services and solutions.”

Website: [Link]

© Blu Ltd 162

 • Deepfakes
– Deepfake = Deep Learning + Fake
– Manipulation of video/images using AI
– FakeApp – app that allows you to create your own
videos with standard GPUs
– Videos adjusted frame by frame, not just a
superimposed picture

• What makes it scary?

– Videos can be manipulated relatively easily…

© Blu Ltd 163

 • How does it work?

© Blu Ltd Source: [Link] 164

 50-year-old Japanese man tricks social media followers into thinking
he is a young female biker

© Blu Ltd Source: [Link]/news/world-asia-56447357 165

 Can you tell that this is a
fake video? Much harder
to spot fakes now!

How can this risk be

handled?

More importantly, can

this tech create
marketing and
advertising buzz?
© Blu Ltd Video: [Link] 166
 1. Did Tom Cruise and Paris Hilton look real? Class
Discussion
2. Did they sound real?
3. If you were not following celebrity news, would you believe
that Tom and Paris were actually dating?

© Blu Ltd 167

 Generated faces for…
Marketing
Advertising
Media/Art
Video game characters
$19.95/month

Benefits
Lower cost
Allows individuals & small
firms to compete with big
companies
[Link]
© Blu Ltd 168
© Blu Ltd 169
 A South Korean news channel
used a deepfake news anchor

‘Content creation’
in marketing
YouTube
Instagram
Twitch
News, blogs
What else?
Source: [Link]/news/business-56278411

© Blu Ltd 170

 • If good, why?
• If bad, why?

© Blu Ltd 171

© Blu Ltd 172
© Blu Ltd 173
 Individual
1. What are you looking to get out of the course? Exercise
2. On a scale of 1-10 (10 – being advanced), how knowledgeable
are you about AI?
3. Which areas of AI are you interested in?
4. Are there other questions you want to ask me?

For online students, please email your responses to

kevinpereira@[Link]

© Blu Ltd 174

1. What are you looking to get out of the course?
• To get a general knowledge about the application of AI in marketing, it’s even better if you could
expand it into other fields.
• I want to gain more trendy insights about how AI technology used in Marketing, especially in
Sports industry
• I would like to learn more case studies of how AI apply to marketing in more practical way, less
technical
• Clear structure about AI hot topics - include machine learning, language programing, etc. and what
is the impact of business
• Success stories as well as things that failed due to AI/How should I handle the AI governance at
work and explain AI and its impact in simple language to business counterparts
• How can we use AI to improve efficiency and ROI?
• How AI will impact education in the future?
• AI applications in my daily life
• Knowledge structure of AI
• Which are the basic AI tools which can help us in regular life?
• A thorough understanding of existing AI technologies businesses are using nowadays, how exactly
those AI technologies can solve business challenges/questions
• Predict the trend of AI's role in marketing and predictive consumer insights in the next 5 to 10 years
• Deeper understanding of the application of AI in marketing, and how enterprises can generate
more profits with AI

© Blu Ltd 175

2. On a scale of 1-10 (10 – being advanced), how
knowledgeable are you about AI?
25%

20%

15%

10%

5%

0%
1 2 3 4 5 6 7 8 9 10
© Blu Ltd 176
3. Which areas of AI are you interested in?
• How can I use AI to make me more efficient at my job/future job?
• Generative AI and Computer Vision
• AI applications in many different industries
• Industrialization / use of AI for marketing and customer insights / effects of future
education pattern
• Prospective use of AI for the entrepreneurs
• Natural Language Processing (NLP) and LLM applications in industry, any startup
opportunities
• AI in CRM analysis, customer segmentation, and predictive analysis on customer
preferences
• Machine Learning/Deep Learning and Robotics
• RegTech and how to address the potential ethical issues brought by AI?
• I would also like to know if you can share some knowledge combined between AI
and Web 3.0. When everything is going to be decentralized in Web 3.0, how deep
of personalization the AI could achieve. what kind of negative impact you could
imagine that it could bring?

© Blu Ltd 177

3. Which areas of AI are you interested in?
• How can AI be used for non-profit organizations?
• AI applied in Finance/investments/fintech and medtech
• I would like to focus on the natural language processing technology, which could
help me build a smart local knowledge banks system in order to improve the
learning efficiency.
• AI applications in the FMCG area
• AI applications in the chemicals area
• AI in data security and privacy protection
• How can I use AI technology if I want to apply it in the video game industry?
• Self-driving cars
• AI decisions/solutions vs Human ethics
• Virtual simulation in manufacture and other areas in Industry
• Usage, future utilization, and valuation
• How it can replace translators/interpreters
• How do chatbots work?

© Blu Ltd 178

4. Are there any other questions you want to ask me?
• What's your advice to a marketer given that this role may be eliminated in future
• What should I do to not be replaced by AI in my future job?
• What are the potential job opportunities will be generated in terms of AI and how will it influence our life?
• People talks about AI replacing human's job. Any new job created by AI? How do you see it?
• How do you manage to transform your career from finance to AI/tech? What you did in order to achieve this
transformation
• How AI can help marketing and generate more revenue?
• In order to better adapt to the future life which AI is taking a more and more important place, what kind of
education is yet important for kids. And for adults, how to use AI to facilitate our life rather than being replaced
by it.
• With more mature AI tools implemented, what’s the most impact to our future life, positively and negatively.
• How to trust and value AI in respect of the truth, correctness, future blueprint the sellers are advertising
• We often hear AI will change the way we work and we live, but we just not know how to do it. So can professor
teach us more about how to use AI tools.
• AI decisions/solutions vs Human ethics
• What is your take on some companies' current rejection to generative AI for the sake of creativity/art? How
businesses/humans/creative service providers (art, music, etc.) can use generative AI to enhance creativity instead
of being threatened?
• Why Geoffrey Hinton was so regretful about the whole thing?/How far from now will AI has its own emotion and
thinking?/How to detect fake images like Pentagon explosion?
• If human just treats AI as a tool, so why they try to stop the development of Chatgpt 5.0? Is it AI will become really
smart than people and dominate human being finally as show in the movie?

© Blu Ltd 179

4. Are there any other questions you want to ask me?
• Other than Microsoft open AI, what other companies are capable of developing comparable generative AI?
• What would be changed in manufacturing? What would happen in technical and consumer electrical products?
• People say Nvidia's GPU already become the AI tax. Do you think the computing power cost for AI training,
inference will go down?
• Will those horrible science fiction films come true?/Which is the best description do you think between human
and AI? Get along with, grow together, concur.
• The use of artificial intelligence in business applications involves a significant amount of data sharing. Is there a
concern about information leakage?
• Any content or online courses you recommended for better understanding of AI and AI related research,
research directions
• What's the practice or problems you get in your consulting business?
• How to better communicate with technology team as a founder according to your experience?
• How soon does AI become essential tools in world? Now many people are talking about AI, but just talking
• What activities can we do to apply AI as an individual or a start-up?
• I want to know how to process the original data? The quality of data? How to evaluate? Or the selecting criteria?
In another words, how to reduce the false data or informations?
• I would like to learn more about your experience as a consultant. Has there been any cases about digital
transformation and change management related to AI innovation?
• Could you please show us how to train your own AI program?
• How to deliver the AI basics to those who are not familiar with (say like the investors who are not familiar with
computer science)
• How to consider the overfitting problem in training the model
• What’s the standard to get an A or above for this course?
• Can you make the exam easier?  (NICE TRY!)
© Blu Ltd 180
© Blu Ltd 181
 • Training a machine to collect text and interpret its meaning
• Challenges with language interpretation
– Same word has multiple meanings
• Context matters
– Different languages don’t have 1-to-1 mappings for translation
• Inuit words for snow and ice

© Blu Ltd 182

 • NLP involves finding the meaning and context of words,
sentences, and paragraphs
• Techniques used, grouped under syntax & semantic analysis
• Syntax analysis: determining grammar rules for words and
cluster them according to similarity
• Semantic analysis: determining word meanings & context to
generate human language
– Challenging because human language rules are complex

© Blu Ltd 183

 NLP is an umbrella term that covers two related techniques:

1. Natural Language Understanding (NLU) - figures out the

meaning behind text and speech by converting human text &
speech to a structured format that computers understand.

2. Natural Language Generation (NLG): text and speech

generated by computers

© Blu Ltd 184

 • NLP applications often involve optical character recognition
(OCR) and deep learning
• OCR allows computers to recognize text in scanned documents
or handwritten text
• NLP models convert text & speech to formats that computers
understand (vectors)
• Deep learning models are then used to:
– process text
– classify words
– cluster similar words
– associate words & phrases with meanings

Source: [Link]/natural-language-processing-in-
banking-current-uses-7fbbaee837de
© Blu Ltd 185
 • Collecting the input
– Convert Voice to Text– 95 % accuracy in English
– Converting sound to Text
– Multi-language
• Key word search
– Sorting algorithm
• Meaning interpretation
– Using semantic sets and context to derive meaning

© Blu Ltd 186

 • Meaning of:
• word
• sentence
• Paragraph
• Text
• context
• Note that Siri, Alexa etc. are just voice/sound interfaces and are not in it of
themselves, AI
• An NLP use case: Translation
[Link]

© Blu Ltd 187

 • Spanish to English
• Translation for specific language pair (Spanish to English)

© Blu Ltd 188

 • Some language translations make no
sense (e.g. idioms)

• French: “avaler des couleuvres”

• English translation: “to swallow snakes”

• Real Meaning: “to accept something

unpleasant” or “to have something
forced on you”

Source: [Link]/expressions/avaler-des-couleuvres/
© Blu Ltd 189
 • Semantic Meaning
– Identifies contextual meaning

Source: [Link]

© Blu Ltd 190

• Word Frequency – one analysis method

Source: [Link]
© Blu Ltd 191
 Sentiment - an attitude, thought, or judgment prompted by feeling (Merriam-Webster)

“Today was a crazy day. It started with missing the bus in the morning and just as I was about to
then catch a second bus, I slipped and fell into a puddle. After somehow getting to work on the
MTR, which was ridiculously crowded, I got to the office where I found out that my favorite
coffee machine was broken. After getting to my seat, I was called into my boss's office to find
out that I was about to be fired. With my box of possessions in hand, I headed home, where I
picked up a handwritten letter in my mailbox. I thought it would be some good news, but it
turned out to be from girlfriend who wrote that she was breaking up with me.”

• Is this guy having a good or bad day?

• What does the AI think?

Microsoft Text Analytics:

[Link]

© Blu Ltd 192

 Sentiment - an attitude, thought, or judgment prompted by feeling (Merriam-Webster)

Song from LEGO move ([Link]

“Everything is awesome, everything is cool when you're part of a team

Everything is awesome, when you're living out a dream
Everything is better when we stick together
Some have said you and I are gonna win forever?
Lets party forever
We're the same unlike you, you're like me we're all working in harmony
Everything is awesome, everything is cool when you're part of a team
Everything is awesome, when you're living out a dream”

• What does the AI think?

Microsoft Text Analytics:

[Link]

© Blu Ltd 193

 • Listen to all online dialogue about your company
• Understand what customers, media, competitors are saying
about you?
– Are customers happy?
– Are there any emerging problems?
– Can you predict future customer needs?

• Measure how healthy your online brand is

– Brand awareness & reputation
– Online presence

• Useful if your customers use social media

© Blu Ltd 194

 • Sentiment analysis tools ‘read’ social media posts and understand the
intent behind it
– Is the person happy or angry with the company?
– Is the person serious or joking? (this is hard to figure out)
• The combined results show how customers feel about the company
across different areas
– Product quality, customer service
– Are overall opinions positive or negative?
• Data sources
– Social media posts and comments (different platforms)
– Blogs & forums
– News (sometimes)
• Companies can monitor overall sentiment trends & fix
new problems without reading each social media post
Source: [Link]/news/10-sentiment-analysis-
© Blu Ltd tools-to-measure-brand-health/575334/ 195
• Online mentions
• Sentiment score
• Keywords
• Languages
• Countries
• Social media
sources

© Blu Ltd Source: [Link] 196

 Social media & web data to predict macro events
Separating signal from
noise
• Sentiment analysis: what is
the general mood?
Insight
• Anomaly detection:
anything strange
happening?

Investment
Action

AI Techniques
• Natural Language
Processing understands
text, meaning, and context
© Blu Ltd
[Link] • Neural networks used for
anomaly detection

© Blu Ltd 197

 • Chatbots
– Used to simulate conversations, primarily through text or sound

Many commoditized chat bots to Rapid specialization of offerings

choose from - Chinese and Cantonese Chatbot
- Microsoft ‘Bot Platform’ - Industry or topic semantic
- ChatScript - Local support
- Pandorabots => See [Link] … and many upcomers
- Facebook ‘Bots for Messenger’
- [Link]
- Imperson
- Google Duplex -
[Link]
ch?v=D5VN56jQMWM

© Blu Ltd 198

 Class
1. What are possible uses of chatbots? Discussion
2. Why could chatbots be a) better and b) worse than humans?
3. What data do we need to collect?

© Blu Ltd 199

 Data
Depends on:
Meaningful
Customer
User
V0 Purpose
Engagement Users
Interaction
V1 Data Required
V2
Implementation
Cost
Time
V3

AI Usage
© Blu Ltd 200
© Blu Ltd 201
Source: [Link]
 • Two Chatbots
• One to respond to
common customer
questions
– Returns, refunds…

• Another chatbot to
help human customer
service agents
respond faster
• Chatbot passes the
conversation to a
human if it does not
know the answer

© Blu Ltd Source: [Link]/2020/02/25/amazon-tests-ai-powered-customer-support-agents-on-amazon-com/ 202

 • Bank of America launched ‘Erica’, a chatbot and virtual
assistant in 2019
• Erica accepts voice/text commands and combines predictive
analytics with NLP to help customers:
– Check balances, transfer money
– Search for past transactions and account info
– Track spending habits
– Help customers manage recurring or late payments

© Blu Ltd Source: [Link] 203

 • Based on your knowledge of romance novel/movie titles you
have seen: Class
– Take a minute or two and come up with ONE romance novel title Exercise
– Were there any factors that influenced why you came up with that title?

© Blu Ltd 205

 • Methodology
– Webscraper in Python
– Dataset: 20,000 romance novel titles under the Harlequin brand sourced from
[Link]

• Some of the more “interesting” titles:

– Secret Secret Baby
– Consultant Bride
– Sleeping Baby Man
– Under the Cowboy
– Sob Over the Boss
– My, Hot Doctor

Source: [Link]
© Blu Ltd 206
Source: [Link]
© Blu Ltd 207
 • ChatGPT: The ultimate autocomplete
– Developed by San Francisco based AI lab OpenAI ([Link]
– Trained on large amounts of text on the web (as of 2021)
– How it works is a secret – Billions of weighted connections between nodes in
ChatGPT’s neural network

• ChatGPT: [Link]
• In Hong Kong: [Link]
– POE = Platform for Open Exploration
– Developed by Quora

© Blu Ltd 208

• LLMs (e.g. ChatGPT) understand and generate human-
like text based on the input they're given (prompts)
• Trained on a massive amounts of public and
proprietary text
– Predict what comes next in a piece of text
• LLMs use this training to answer questions or generate
new text
– Draft emails or pull information from the internet or an
internal company database

© Blu Ltd 209

 • LLMs trained on internal company data
• Potentially higher accuracy, no data sharing
with external companies
• Two approaches
– Use external LLM model (e.g. OpenAI) through
an API (becoming more common)
– Build your own LLM

© Blu Ltd 210

© Blu Ltd Source: [Link]/story/artificial-intelligence-has-advanced-so-much-it-wrote-this-article-11617639437 211
© Blu Ltd 212
 • Company blogs, product descriptions, social media posts are
examples of text-based content
• A LLM can train on a database of company and product-
related info to ‘write’ content for websites and social media
• Of course, the output won’t be perfect… human review and
edits needed

• Benefits
– Overcome “writer’s block”
– Create 100’s or 1000’s of variations of the same content to see which
creates more customer engagement

© Blu Ltd 213

 Individual
1. What elements of the course do you like so far? Exercise
2. What elements of the class do you not like so far?
3. Do you have any suggestions for how we can improve your
experience?

© Blu Ltd 214

© Blu Ltd 215
 • Robotics
– Involves the design, creation, and operation of robots to automate certain
tasks

– Draws on the fields of electrical engineering, computer science and

mechanical engineering

© Blu Ltd 216

 Source: [Link]

© Blu Ltd 217

 Source: [Link]
Source: [Link]
© Blu Ltd 218
 • Boston Dynamics
• Precision movements, fine motor balancing…dancing?

Source: [Link]

© Blu Ltd 219

Robots doing parkour is cool but watching them fall is more fun

Source: [Link]
Parkour Video: [Link]
Fail Video: [Link]
© Blu Ltd 220
 Reinforcement Learning: Robot is ‘rewarded’ for good behavior (standing up and
walking), and penalized for bad behavior (failing to walk)

Robot dog started After 10-20 minutes, In an hour, it had taught itself
struggling on its back it learned to flip over how to walk.

In comparison, human kids start walking at 18 months…

Source: [Link]/2022/07/18/1056059/robot-dog-ai-reinforcement/
Alt. Source: [Link] 221
© Blu Ltd Video: [Link]
 What will it be in 30 more years?

Source: [Link]
© Blu Ltd 222
 Source: Tesla AI Day 2021
© Blu Ltd [Link]/watch?v=j0z4FweCy4M 223
© Blu Ltd 224
 225
© Blu Ltd Source: [Link]
 Source: [Link]

© Blu Ltd 226

 Hanson Robotics - Sophia AI Human-Like Robot Demonstration

Source: [Link]

© Blu Ltd 227

 1. Do Sophia’s facial expressions seem realistic?
2. If realistic, why?
3. If unrealistic why?
4. What use cases could Sophia be used for? Group Work

© Blu Ltd 228

 • Meet Pepper the Robot – from Softbank Robotics

Source: [Link]

© Blu Ltd 229

 Possible applications
• Customer service robots
– To guide customers in-store

• Customer education
– Help customers learn about products
– Creates a memory of your company
– Salespeople handle the detailed customer
interaction

© Blu Ltd 230

 Henn na Hotel (Japan) – “The strange hotel” in English…

Source: [Link]

© Blu Ltd 231

© Blu Ltd 232
© Blu Ltd 234
 • Computer Vision
– Training a machine to collect pictures and derive meaning
– Hardware required varies based on use case

Source: [Link]

© Blu Ltd 235

 • How it works
– From an image the Machine extracts features and classifies them
– Video = image + time
• Typical Financial Services use cases
– ATM video authentication
– Surveillance
– Trading indicators
• Satellite imagery for crop
• Wall mart car park
– on-site Customer behaviour (face emotion analysis)
• Fraud detection, customer waiting / satisfaction
• but also .. Casinos, Classroom, inside shopping malls, road traffic control

© Blu Ltd 236

 Dog or bagel (piece of bread)?
What do you think?

© Blu Ltd 237

[Link]
m/dog-food-comparison-
bagel-muffin-lookalike-
teenybiscuit-karen-zack

© Blu Ltd 238