A.I for everyone. Week 1 & 2 takeaways

* A.I is estimated to add 13 trillion US dollars annually by the year 2030.

* A.I will have an impact in all industries from agriculture to retail and healthcare systems.

* A.I can be classified into: ANI and AGI.

* ANI: Artificial Narrow Intelligence is like self-driving cars, A.I in factories, web searches ... etc.

* AGI: Artificial General Intelligence is the one that can do anything a human can do.

[Machine Learning]

* is a tool to achieve A.I.

* Types of machine learning: Supervised Learning, Unsupervised Learning, Deep Learning, and Reinforcement Learning.

* Supervised Learning: in supervised learning, the system is provided with the input data and the desired output data.

* Unsupervised Learning: in unsupervised learning, the system is provided with input data and then it's automatically going to look for patterns and relationship between the data and put them into groups.

* Deep Learning: deep learning or artificial neural networks ANN are taken from the neural networks in the human brain and is one of the ways to build ML models, but it is completely different from the neural networks in the human brain

* Reinforcement Learning: in reinforcement learning the system is provided with input data and then left to make the decision. If the decision is correct, the system will receive a reward, and if it is not, it will receive a penalty. This process is repeated hundreds of thousands of times until the system learns to make the right decisions.

* Machine Learning vs Data Science: the output of a DS project is the insight that can help you make a business decision. The output of a machine learning project is a software that makes the machine learns about a particular problem and make predictions.

[Data]

* Data is the facts, pictures, texts, values, audios ... etc. A dataset is a group of collected data.

* Acquiring the data can be through: observing user behaviors or other types of behaviors, manual labeling, download from the internet, IoT devices, mobile phones, laptops, .. etc

* Don't wait too long to collect the data. When you start collecting the data, hand it over to the A.I team to start building the models. The feedback you will get will include things like how to refine the data and what types of data is more needed.

* If you own a company, don't throw data at the A.I team and assume it will be valuable.

* Data is usually messy. If it's incorrect, then your A.I will predict wrong values.

* Data can be structured or unstructured. Unstructured data like audio, images or text.

[A.I in your company]

* Because you use some Deep Learning algorithms doesn't make your company an A.I first company

* An A.I first company has strategic data acquisition

* An A.I first company has unified data warehouse

* An A.I first company are very good at spotting automation chances

* To make your company effective at using A.I you should: 1) Execute small projects to understand what A.I can or cannot do for your company and what A.I project looks like 2) Build an in-house A.I team 3) Provide A.I training for everyone 4) Develop an A.I strategy 5) Develop internal and external communications so that every stakeholder understand how your company navigates in the rise of A.I.

[What A.I can and cannot do]

* One imperfect rule of thumb is that anything we can do with a second of thought, an A.I can probably do it.

* An A.I cannot do a market analysis and write a 50 pages report in a second.

* An A.I cannot write an empathetic paragraph.

* In self-driving cars an A.I can detect what's in front of the car and where are the other cars. However, an A.I cannot understand or interpret the intention of a human gesture in the street.

* In general if the problem is "simple" enough and a lot of data is available, then most likely an A.I can be built to solve a problem or automate something.

[Building A.I project]

* Steps in machine learning projects: 1) collect data 2) Train the model 3) Deploy the model

* Steps in data science projects: 1) collect data 2) analyze data 3) suggest hypothesis/actions 4) deploy changes

* To choose an A.I project: 1) Think about automating tasks rather than automating jobs 2) Identify the main drivers of the business value 3) Identify the pain points in your business 4) Will you build the project in-hous or buy one?

A common rule of thumb here is that the best project is feasible in A.I and brings values to your business.

[Working with an A.I team]

* Specify an acceptance area for the project

* A.I team think of data in two way: 1) The Training dataset 2) The Test dataset

* Don't expect 100% accuracy from the machine learning model

* A.I teams commonly use the following tools:

- Machine learning frameworks: TensorFlow, PyTorch, Keras, MXNet, CNTK, Caffe, PaddlePaddle, Scikit-learn, R and Weka.

- Research publications: Arxi.org

- Code sharing: github.com

?

The course is free if you'd like to take it:

https://www.coursera.org/learn/ai-for-everyone