Where to start in AI

AI technology has been on the rise more so than ever, with AI trading platforms, China’s ‘skynet’ surveillance network, and Google DeepMind’s AlphaGo just to name a few. But what exactly is AI? And how does one who has little exposure to the field go about learning it? To start off, I believe it is key to first understand the correlation and differences between the following terms: Artificial Intelligence, Machine Learning and Deep Learning. All of these are often used interchangeably by people, potentially causing confusion to beginners trying to find a path to start learning about AI.

To best understand the relation between the three, as is commonly explained, one can imagine concentric circles where AI would be the outer circle, Machine Learning the middle and Deep Learning the inner circle.

Before digging into the inner circles it is essential to understand what Artificial Intelligence is. AI is an overarching concept that encompasses everything related to making a computer being able to act more human. It is allowing a computer to accomplish a task that would otherwise be easy for humans to do. Tasks such as speech recognition, facial recognition, and emotional interruption are a few of the tasks that AI could accomplish.

AI was originally coined by John McCarthy back in 1956 where he defined AI, or “Machine Learning” as:

What the quote says is that AI works to simulate human behavior. There are several definitions out there, but McCarthy’s helps to properly steer into the inner circles of Machine Learning and Deep Learning.

Machine learning can be considered a more concentrated, specific type of AI. It focuses on the learning aspect of machines, specifically around how it receives data and can learn from it; changing algorithms by themselves in order to efficiently process and act on the data. This is all possible thanks to neural networks.

Neural networks are virtually modeled after the human brain. They are programmed to recognize patterns in data, categorize it, and decide whether to continue using the same algorithms or change them in order to increase efficiency. Examples of machine learning can be seen as email filtering or facial recognition software.

More interesting types of machine learning can be seen in robots learning how to lie in order to be rewarded. In an experiment that was originally meant to test teamwork between robots, some of the “guide” robots learned that they could harvest more reward if there were not as many robots in the same area, so they would purposely lead other robots into non-reward areas. They were not originally programmed to behave in this manner, but instead they learned it on their own.

Following the topic of machine learning leads into the concept of deep learning. Machine learning uses only one neural network with a number of algorithms, whereas deep learning uses two or more neural networks paired with exposure to large quantities of data.

One could argue that deep learning uses neural networks as decision trees. When one neural network produces a result it can lead to another neural network processing for a different set of results as a consequence of the first result. These neural networks allow the machine to “think”, and when exposed to large amounts of data it can lead to accurate results. An example is the Google Brain, which is able to immediately identify everything about a cat after being fed millions of data points on cats. Another example is the ability of driver-less cars to study a sign from afar and immediately determine which type of sign it is, followed by what action it should take.

Language:

It is Impossible to talk about AI without mentioning about specific languages, but don’t go ahead rushing to learn a new language. If you do not have a large amount of time on hand, start out with, the language you are currently most comfortable with first. Try and learn a language in tandem while you explore basic AI programs, or defer it until a later date when you have more free time. Ultimately, it is recommended to pick up languages such as Python, R, Java… before moving too far ahead as they are the more preferred languages for AI related development which results in more online resources and ML libraries available.

What to code:

At the core of each AI program is its algorithm, therefore here is a of break-down of what you could be doing based on the difficulty of an algorithm.

Novice: you should be starting out with simple programs that revolve around a static algorithm. My personal first project was an AI that plays tic tac toe based on the minimax algorithm. Ideally the first two or three programs you write has online resources that guide you through the process.

Intermediate: Once familiar with the basic structure of an AI program, try delving into self-improving algorithms. A good place to start is to try and adapt the static algorithm projects you created to incorporate self-improvement. You may have begun to try writing your own algorithms from scratch and experimented or researched a little into how neural networks are written

Advanced: By now, you should be familiar with writing your own algorithm by scratch and learning to integrate them into a neural network. If you find yourself comfortable doing so, try layering multiple neural networks together.

Beyond this point, you are well into DL where the forefront of AI technology currently is.

Bear in mind, this is only a rough road-map of what you should be doing. In the end, it is my opinion, and you do not have to follow it. Other things to note, it is not necessary to expand your skills until you’ve learnt DL. Just because DL is the most complex does not make it the best, as some tasks may not require such complexity therefore unnecessarily wasting time. How far and much you learn should be ultimately dependent on your interests and what you wish to pursue.

Related posts:

Cyber Monday is coming , Huobi Pool is awarding a limited amount of free EOS Accounts through EOS Red Packet. At the bottom of this passage, there will be the promotion code for free EOS account. EOS…

Imagine you come home after a long, hot, humid day. You take your boots off, loosen your tie, and turn your AC on (you wouldn’t be doing the last step if you had the Cielo Breez Plus), but smell…

Over the last couple of years, we’ve experienced a huge influx of speech recognition technologies — to the extent that it feels like we have another person in our homes. Voice platforms are…