Chapter 1

Introduction
Non-AI vs AI products
Without AI:
Although the code on these IoT devices is becoming smarter with every
upgrade, the devices are not exactly learning autonomously
Advancement in AI will help solve this problem, and soon we will have devices
that will learn from the input of their human creators, as well as from their own
mistakes
→ Today we are surrounded by code, and in the near future, we will be
surrounded by embedded artificially intelligent agents.
No one-size-fits-all solution

Every organization is different, and it is important to remember not to try to apply

techniques like a straitjacket. Doing so will suffocate your organization. This
course is prepared with a mindset of best practices. Although best practices will
work in most cases, it is important to remain attentive and flexible when
considering your own organization's transformation. Therefore, you must use your
best judgment with each recommendation we make. There is no one-size-fits-all
solution, especially not in a field like AI that is constantly evolving
FANUC (text book)

https://www.youtube.com/watch?v=EcpTRU4GiuY

Robots builds robots

- Deep learning model to identify and pick the correct devices from a bin
- Consistent 90% accuracy
How Facebook fight against fake news

https://about.fb.com/news/2018/05/hard-questions-false-news/

https://www.facebook.com/MetaforDevelopers/videos/10155608056828553/

https://about.fb.com/news/2017/12/managing-your-identity-on-facebook-with-face-
recognition-technology/
Discussion

What can be learned from these case studies?

How they identified the problem?

How they found the solution?

How they implemented the solution?

How they verified the solution was good enough?

An overview of the AI roadmap in business

The journey to adopt AI promises to bring major changes to the way your
organization thinks and approaches its future

The plan for achieving this goal can vary from organization to organization, but the
main steps invariably remain the same
An overview of the AI roadmap in business

Defining the
Ideation Data curation Prototyping Production
project

What problem are you Outline specifically Which data is Build an initial version Complete your
trying to solve? which improvements necessary for this of your system prototype and put it in
How do you believe you plan to attain. problem? Improve it time by time action
your organization will Make a project plan as How to obtain the Adjust your approach if Perform technological
be able to benefit from a guiding document data? any problem arises evaluation
AI? for the implementation How to improve the See with actual results, Build user security
What AI technologies of your project. data quality? whether the plan you models
are available to solve defined in the previous Establish testing
these problems? steps have promise framework to ensure
the system works well
Thriving with an AI lifecycle

Maintain your AI system while consistently looking for ways to make it better

- Improve the current system with latest technology

- Apply the system to other parts of your organization

Save your experience gained during the implementation of your first AI system for
future projects
Materials

● Textbook: Artificial Intelligence for Business - A Roadmap for Getting

Started with AI, Jeffrey L. Coveyduc, Jason L. Anderson, Wiley 2020
● References: Artificial Intelligence in practice - How 50 successful companies
used artificial intelligence to solve problems, Bernard Marr, Matt Ward, Wiley
2019
● Lecture slides
How to choose your group project
Look for an organization, an enterprise, a company
- That you know in real life
- That you read from books, websites, … (e.g. the reference book)
- Or even a business that you want to create, why not?
Think about the products, the services it provides, or its working process.
What does not satisfy you? Any idea to improve?
Please focus on AI aspect (something can be improved by AI, machine
learning)
Or, can you reproduce their idea from scratch?
Example of group project idea

FaceID in iPhone could not recognize you when you wear a facemask. It’s very
inconvenient when you are on the road

- Any alternative AI technology to unlock iPhone rather than facial recognition

- Any machine learning algorithm to help verify a face with facemask (half-face
recognition)
- If there are no algorithms available, could we divide this problem into sub-
problems to solve, e.g. eyes recognition, forehead recognition, …
- How to integrate your algorithm in a real iPhone
Example of group project idea

Imagine you run a Facebook page of a student club, you post your club activity
news everyday and many students comment on those posts. You don’t want the
hate/profanity speech to be appeared.

- Is there an algorithm to filter out those hateful comments

- If you can build these kinds of comment classification model, can you
integrate to your Facebook page?
The Goldilocks rule

The human brain loves a challenge, but only if it is within an optimal zone of
difficulty.

The Goldilocks Rule states that humans experience peak motivation when
working on tasks that are right on the edge of their current abilities. Not too hard.
Not too easy.

Something in-between is just right.

Grading policy

Attendance: 10%

Midterm test: 20% - multiple choice questions

Project checkpoints: 20% - Group presentation

Final exam: 50%

Plus points during the course

Attendance score policy

- The default attendance score for each student is 8

- Do not pay attention to the lecture → 1 minus point
- Volunteer to answer a question / discuss a topic → 1 plus star
- 2 stars = 1 attendance plus point
- 2 attendance plus points = 1 midterm point
- The maximum attendance plus point is 6
Fundamentals of AI

Let’s remind ourselves about the fundamentals of AI in the previous course

1. AI definition
2. AI capabilities
3. Machine learning basis
4. Machine learning algorithms
5. Deep learning
6. AI ethic
AI definition

● A branch of Computer Science, which is mainly concerned with automation of

Intelligent behavior
● Strong AI (General AI) vs. Weak AI (Narrow AI)
● The Turing test: measure the performance of an intelligent machine against
humans, for its intelligent behavior
● Techniques in AI: data representation & learning
AI capabilities

● Computer vision (CV)

○ Face Recognition, Self driving cars, Smart Parking
● Natural language processing (NLP)
○ Google Translate, Opinion Mining
● Speech recognition
○ Siri, Google Assistant
Machine learning

● A computer program is said to learn from experience E with respect to some

task T and some performance measure P, if its performance on T, as
measured by P, improves with experience E (Tom Mitchell - 1998)
● Experience E:
○ Data
○ Attribute
○ Label
○ Feature
● Supervised vs. Unsupervised learning
Machine learning

● Task T:
○ Classification: predict a discrete value / category / class of objects, e.g.:
■ predict a person is positive/negative to Covid-19
■ predict a movie is Romantic/Action/Horror
■ predict a word is Verb/Noun/Adjective
○ Regression: predict a continuous value, e.g.:
■ predict the price of an apartment
■ predict the age of a person
■ predict the stock price
○ Clustering: group data samples into clusters to discover/understand them
○ Others: recommendation systems (Amazon, Youtube,...), generation (translation,
summarization,...)
Machine learning

● Performance measure P
○ The idea of Turing test
○ Regression: MAE, MSE, AUC, ROC
○ Classification: Accuracy, Precision, Recall, F1-score, Confusion matrix
○ Train/test split, k-fold cross validation
○ Underfit, overfit, normal fit
Machine learning algorithms

● Supervised learning:
○ Linear regression
○ K nearest neighbor
○ Decision trees
○ Naive Bayes
● Unsupervised learning:
○ K-means clustering
Deep learning

A subset of Machine learning that use neural networks with hidden layers
Deep learning vs. traditional Machine learning

Require a huge amount of data

Require enormous computational power (GPU, TPU, parallel computing)

Useful tool:

- Deep learning frameworks: Tensorflow, Keras, Pytorch

- Research publications: arxiv.org, paperswithcode.com
- Open source repositories: github, gitlab
AI ethic
● AI is transforming the business world, there are challenges, but also
opportunities
● AI discrimination/bias
○ AI hiring tools discriminated against women, Face recognition works
better on light-skinned than dark-skinned people
● Adversarial attacks on AI
○ Paraphrase some words to pass the plagiarism test, Add some noise to
the video streaming to pass the copyright regulations
● Adverse use of AI
○ Synthesize video/photos of people doing things they never did,
Undermining of democracy and privacy