Artificial Intelligence¶

Branch of computer science which designs ‘intelligent’ machines capable of behaving, thinking and making decisions like a human.

Rapid Rise¶

Due to data availability from digitalization

Key Areas¶

1. Machine Learning
2. Natural Language Processing
3. Robotics
4. Object Detection
5. Speech Recognition

Applications¶

1. Fashion and Art
2. Science
3. Games
4. Music and Sounds
5. Videos and Images
6. Business and Finance
7. Security and Safety and the list goes on...

Why AI?¶

• Replicate human intelligence to provide precise decisions and outcomes
• Solve knowledge-intensive tasks to reduce human workload
• Establish an intelligent connection of perception and action

Types¶

There are 2 types of artificial intelligence

Limitations¶

Don’t be too optimistic or pessimistic about AI

1. Performance

2. Explainability: AI finds it hard to justify its decisions

3. Biases due to biased data

4. Attacks on AI
5. Adverse use of AI

Responsible AI¶

1. Fairness: Ensure AI does not perpetuate/amplify biases
2. Transparency: Making AI systems and their decisions understandable to stakeholders impacted
3. Privacy: Protecting user data and ensure confidentiality
4. Security: Safeguard AI systems from malicious attacks
5. Ethical use: Ensuring AI is used for beneficial purposes

AI Company¶

This section is not relevant from the AI courses as such, but is important to know.

Company + Deep learning \(\ne\) AI company

Features¶

1. Strategic data acquisition: some companies release non-monetised products just for data collection
2. Unified data warehouse: makes it easier to connect and link
3. Task Automation
4. New roles (ML engineer) and division of labor

Roles¶

Roles aren’t concrete

1. Software engineer - program/task execution
2. ML engineer - Trains the neural network
3. ML researcher - extend state-of-the-art in ML
4. Applied ML scientist - does roles of ML engineer and researcher
5. Data scientist
6. examine data
7. provide insights
8. presentations
9. Data engineer
10. organize data
11. ensure data security, accessibility and cost-efficiency
12. AI product manager
13. decide what to build
14. feasibility and value

Transformation¶

Steps for a company to transform into an AI company

1. Execute pilot projects

2. focus on success rather than value of pilot projects

3. some sort of progress in 6-12 months
4. In-housed/outsourced
5. Build in-house AI team
6. Provide AI training to all employees

Curate (online courses, books, etc) instead of create training 4. AI strategy 1. use AI to create an advantage specific to the required sector 2. strategic data acquisition

1. unified data warehouse

2. network effects and platform advantages

   Eg: social media - more users join, more lucrative for prospective users

3. aligned with ‘Virtuous Cycle of AI’

flowchart LR

a[More<br/>Data] -->
b[Better<br/>Product] -->
c[More<br/>Users] -->
a

1. Communications: everyone should be aligned with how company is ‘navigating' with AI
2. internal
3. external
   1. investors
   2. governments
   3. customers
   4. talent

Projects¶

Outsource whatever does not require specialisation/customisation - especially industry-standards

Choosing a project¶

Brainstorming¶

1. Automate tasks instead of automating jobs
2. what are the main drivers of business value
3. what are the main recurring weak points in your business

Proofing¶

• Techincal Diligence - AI experts
• AI must be able to perform
• how much data is required
• resources required
• Business diligence - domain experts
• valuable
• does it
  • lower costs
  • increase revenue
  • allow to launch new product/business
• Ethical diligence
• does it benefit/harm society

Implementing¶

1. Specify acceptance criteria
2. do not expect 100% accuracy
   1. limitations of ML
   2. Insufficient data
   3. mislabelled data
   4. ambiguous labels
3. provide AI team with
   1. training set
   2. test set
4. to measure performance
5. Eg: 95% accuracy

Implementation¶

IDk

• Augmentation: Help humans with a task
• Automation: Perform task without human

Potential

• Business Value: Does this significantly fasters, cheaper, more consistently create substantial value
• Technical Feasibility: Can AI do it?

Comments