Week 1:

McKinsey AI value creation by $13 trillion by 2030.

AI can be bifurcated into:

- Artificial Narrow Intelligence (ANI): Eg. Smart speaker or self-driving car

- Artificial General Intelligence (AGI): Do anything a human can do

Machine Learning:

- Supervised Learning: This is A to B mapping or input to output. It has taken off recently due to
neural networks and deep learning and big data.

Acquiring data:

- Manual labeling
- From observing behaviors (eg: from human behavior on website: buy or not to buy)
- Download from website/partnerships

Use of mis-use of data: Don’t throw data at an AI team and assume it will be valuable

Data is messy:

- Garbage in, garbage out

- Data problems: Incorrect labels, missing values
- Multiple types of data: image, audio and text => they are called unstructured data (Structured
data is data that lives in a giant spreadsheet). Techniques are different to deal with structured
and unstructured data

Machine Learning vs Data Science:

- Data Science provides a list of insights into the data can help you make business decisions
- Field of study that gives computers the ability to learn without being explicitly programmed

Deep Learning:

- Neural networks are very effective in A to B mapping i.e. supervised learning. Neural networks
were originally inspired by the brain, but the details of how they work are completely unrelated
to the biological brain

What makes an AI company?

- Internet company is a company that does the thing that internet let you do really well: For
example, A/B mapping, short iteration time, decision making pushed to engineers and other
specialized roles
- AI companies are good at strategic data acquisition, unified data warehouses, pervasive
automation, new roles such as machine learning engineer and division of labor. To become good
at AI means, architecting for company to do things that AI makes it possible to do really well

What machine learning can and cannot do:

Machine learning tends to work well when:

- Learning a simple concept (less than a second of thought)

- Lots of data available
ML tends to work poorly when:

- Learning complex concepts from small amounts of data

- It is asked to perform on new types of data

Week II

Starting an AI project

- Workflow of projects
- Selecting AI projects
- Organizing team and data for the projects

Workflow of projects:

Key steps of a machine learning project:

- Collect data
- Train model
- Deploy the model

Workflow of data science project:

- Collect data
- Analyze data
- Suggest hypothesis/actions

How to choose AI project:

- Select projects both feasible and valuable for your business. To do this you have to go through
technical and business diligence. The technical diligence includes the following points: 1) Can AI
system meet the desired performance 2) How much data is needed 3) Engineering timeline.
Business diligence includes 1) lower costs and higher efficiency 2) increase revenue 3) launch
new product or business. The third type of diligence is ethical diligence.

Build vs Buy:

- ML projects can be in-house or outsourced

- DS projects are more commonly in-house
- Some things will be industry standard – avoid building those
- Build those things which are quiet or completely specialized to you or they allow you to build
unique defensible advantage.

Working with an AI team:

Specify your acceptance criteria: For example, the goal is to detect defects with 95% accuracy. The AI
team needs to be provided with a dataset on which to measure their performance. There is going to be
a training set and a test set. The training set will enable the machine to learn about the data and
perform A to B mapping. The test set is the one on which the algorithm in run to decipher performance.

Pitfall: Expecting 100% accuracy due to:

- Limitations of ML
- Insufficient data
- Mislabeled data
- Ambiguous labels
Brainstorming framework:

- Think about automating tasks rather than automating jobs

- What are the main drivers of business value?
- What are the main pain points in your business?

You can make progress even without big data:

- Having more data almost never hurts

- Data makes business defensible
- But with small datasets, you can still make progress

Week III

Case study: Smart speaker

Steps to process the command:

1) Trigger word/wake-word detection

2) Speech recognition
3) Intent recognition
4) Execute program

Case study: Self-driving car

Steps for deciding how to drive:

1) Image/radar/lidar/GPS/accelerometer/map
2) Car detection and pedestrian detection/lane detection/traffic light detection/obstacle detection
3) Trajectory prediction
4) Motion planning

Example roles:

- Software engineer: To write specialized software

- Machine learning engineer: To write software to do A to B mapping or building machine learning
algorithms needed for your product
- Machine Learning researcher: To extend state of the art in ML and may publish papers
- Applied ML scientist: To go through academic literature and finding steady VR techniques and
finding ways to adapt them to the problem the team is facing
- Data Scientist: To examine data and provide insights
- Data Engineer: Organize data and make sure that it is stored in an easily accessible, secure and
cost-effective way
- AI product manager: To help decide what to build; what’s feasible and valuable

AI Transformation Playbook:

1) Execute pilot projects to gain momentum

- More important for the initial project to succeed rather than be most valuable
- Show traction between 6 to 12 months
- Can be in-house or outsourced
2) Built an in-house AI team
3) Provide broad AI training
4) Develop an AI strategy
- Leverage AI to create an advantage specific to your industry sector
- Design strategy aligned with the Virtuous Cycle of AI
 -Consider creating data strategy:
- Strategic data acquisition
- Unified data warehouse
- Create network effects and platform advantages:
- In industries with ‘winner take all’ dynamics, AI can be accelator
5) Develop internal and external communications
- Investor relations
- Government relations
- Consumer/user education
- Talent/recruitment
- Internal communications

AI pitfalls to avoid:

DO’s:

- Be realistic about what AI can and cannot do given limitations of technology, data, and
engineering
- Pair engineering talent with the business talent and work cross-functionally to find feasible
and valuable projects
- Plan for AI development to be an iterative process, with multiple attempts needed to
succeed
- Work with AI teams to establish timeline estimates, milestones, KPIs, etc
- Keep building the team as there are many AI engineers

DON’T:

- Expect AI to solve everything

- Hire 2-3 ML engineers and count solely on them to come up with use cases
- Expect the AI project to work the first time
- Expect traditional planning processes to apply without changes
- Think you need superstar AI engineers before you can do anything

Taking first steps in AI:

- Get friends to learn about AI

- Start brainstorming projects
- Hire a few ML/DS people to help
- Hire or appoint AI leader
- Discuss with CEO/Board possibilities of AI transformation

Major AI application areas:

Computer vision:

- Image classification/object recognition

- Object detection
- Image segmentation
- Tracking

Natural language processing:

- Text classification: sentiment recognition

- Information retrieval: web search
- Name entity recognition
 - Machine translation
- Others: parsing, part-of-speech tagging

Speech:

- Speech recognition(speech-to-text)
- Trigger word/wake-word detection
- Speaker ID
- Speech synthesis (text to speech, TTS)

Robotics:

- Perception: figuring out what’s in the world around you

- Motion planning: finding a path for the robot to follow
- Control: sending commands to the motors to follow a path

General machine learning:

- Unstructured data (images, audio, text)

- Structured data (that in spreadsheet)

Unsupervised learning:

- Clustering

Transfer learning: helps you learn from task A and uses the knowledge on task B

Reinforcement learning:

- Uses a ‘reward signal’ to tell the AI when it is doing well or poorly. It automatically learns to
maximize the reward.

Generative Adversarial Network (GAN):

- Synthesize new images from scratch

Knowledge graph:

- It can be database that lists key information about certain things such as movies, celebrities
etc.

Week 4:

Goldlilocks rule for AI: Shouldn’t be too pessimistic or optimistic about AI

Limitations of AI:

- Performance limitations
- Explain-ability is hard
- Biased AI through biased data
- Adversarial attacks on AI

Why bias matters:

- Hiring tool that discriminated against women

- Facial recognition working better for light-skinned than dark-skinned individuals
- Bank loan approvals
- Toxic effect of reinforcing unhealthy stereotypes
Combating bias:

Transparency and/or auditing processes

Diverse workforce

- Create less biased application

Adversarial Attacks: Implementing ways to fool the AI system

Adverse uses of AI:

- DeepFakes
- Undermining of democracy and privacy
- Generating fake comments
- Spam vs anti-spam and fraud vs anti-fraud

How developing economies can build AI:

- US and China are leading, but all AI communities are still immature
- Focus on AI to strengthen a country’s vertical industries
- Public-private partnerships to accelerate development
- Invest in education