Diving into Artificial Intelligence

A summary of Andrew Ng’s “AI for Everyone” course taught in Coursera, By Kevin Anderson

Artificial Intelligence, AI, has become a rising aspect of technology these days. Multiple companies, big and small are applying AI concepts within their business to either sell AI services to the market or increase efficiently of internal production. So, what is AI and how can we define it? Simply put AI has not yet been given a most definite introduction, for intelligence itself is a big topic on its own. To this day human intelligence is still in its analytical phases as humans, together with their environment and lifestyle, have continued to evolve overtime.

According to Andrew Ng, AI can be divided into two categories come, one being artificial narrow intelligence and the other being artificial general intelligence. Artificial narrow intelligence (ANI) is the most common form of AI that is used today. This can be seen in smartphones smart cars AI based websites and machinery. When it comes to artificial general intelligence (AGI), this is where we have a machine being able to do anything a human can do.

There are also common terminologies using artificial intelligence among them being machine learning and data science. In machine learning, you applied a concept of feeding input and expecting an output. You train a module to gather data from whatever source it can get the data from and provide a specific output. This allowed spectators to learn without being explicitly programmed. In data science, this is the science of gathering information or data in order to provide a predictive. This means you can use data science to gather information such as activities, human behaviors, charts and numbers in order to predict what would be the best course of action to take depending on what the purpose of gathering the data was. this can be defined as a science of extracting knowledge and insights from data.

So, we know that both of these concepts use data but how do we define data? Data, according to AI, can be defined as a collection of raw input gathered by a module or by the creator of the module, which would be used to train a module to come up with an output or an analytical result. They take a very from text input, pictures, audio input, documents and many more. Based on what an angle is or what data is available, you can determine as to whether or not you can apply machine learning to your business or to your project. Unfortunately, with today’s level of technology there are still some limitations to machine learning or AI in general. For instance, we are not yet at a point where we can train a machine to have self-imposed emotional responses or empathetic responses to a user’s input. Andrew gave a good example of how it would be difficult to train the machine to respond to someone’s email while mentioning and acknowledging their daughter’s birthday.

In order to create a machine learning project, you need to 1st gather data for your project I if you want to detect if an animal such as a cat or dog you can gather different images of dogs and cats and then train the module there was supervised learning to predict which picture is that of a cat and that of a dog. Projects such as self-driving cars use machine learning where the driving module would gather information from the car’s environment. This data includes distance from one object to another, distance from the object to the car, pedestrians, positions of other cars, heights of objects, how long is the stretch of the road and many more. Once you have gathered all the data you can then train the module by giving the input and it’s the expected output. You can give your module a large number of pictures of cats and tell it that these are cats and the same thing can be done with dogs. Once Training is complete you can’t go ahead and deploy the model, and this is where you would test its predictive abilities and determine whether or not you would need to update the model.

For data science, the data collected is initially analyzed first before providing a result. You may have noticed that a lot of mobile applications have integrated Facebook login for the user profile dot. Before Facebook had started allowing user authentication two other applications, these applications would have a long sign up process which could discourage her user from proceeding further into registering their account. By noticing this activity applications have applied that use of external application authentication to their application and they further noticed that it would be better to have the user feel their profile information later on rather than during initial setup or sign up. this process uses data science to synthesize the correct action. After an action has been predicted, the data will still be recollected and further analyzed.

Ideally every job function can use either machine learning or data science simply by identifying which part of the function works either one of the aspects. When it comes to choosing run AI projects, they are several factors to consider. One of them is to know what artificial intelligence can do and what is important to the business and see whether there are almost 10 minutes commonalities between the business value and the things that AI can do. You can also determine which tasks can be automated as well as knowing what the main drivers and the main points in your business are in your business. Keep in mind that having more data usually never hurts but you can also make progress using smaller datasets. If you have an AI team you can discuss on the different diligences in determining whether a project is feasible for example on a technical aspect you can ask questions like:

Is the project feasable or doable?

How much data is needed?

What is a time frame for the project?

Andrew Ng

In A business aspect you could ask:

will the project be cost effective?

Will the project provide an increase in revenue?

Will the business provide a new product?

Andrew Ng

You can also determine whether or not you need to outsource development services for either data science projects or machine learning projects. Most data science projects are company done in house. There is some functionality that are industry standard which do not require further development.

AI projects can be achieved by team AI team. Depending on the number of members there are specific roles that can be divided among each member. Below are some of those roles;

Software Engineer – Creates quality software that will be applied to the end product  

Machine Learning Engineer – Creates the A to B concept model

Machine Learning Researcher – extends the state or the machine learning model

Applied machine learning scientist – researches on machine learning functionality and feasibility

data scientist – examines data and provides insights

data engineers – organize data and make sure that data is saved and accessible, secure and cost effective

AI project manager – helps decides on what to build.

To work with your team, you must first specify or determine what is your acceptance criteria. You also need to provide a training set and a test set. Both sets can have labeled inputs so you can test the accuracy. Make sure that your test set id different from you training set. You can start gathering your AI Team with a small number of members.

You can get started with a small AI team

Andrew Ng

Artificial intelligence can be as complicated or as easy as you would like it to be. I believe that there is still a lot more to learning about artificial intelligence and more to discover in regards to the topic as we get closer and closer to mimicking human intelligence. AI can by far be used to improve the life of humans and time moves on.

AI is a super power, and understanding it allows you to do things that very few people on the planet can. 

Andrew Ng