An absolute beginner’s guide to machine learning, deep learning, and AI

07-07 15:08
  1. Tech


Meet Samantha. She’s your friendly assistant from 2025. She sorts your mail, sets up your meetings, and orders groceries. She paints and writes poetry. She’s your best friend. She’s also an artificial intelligence from the movie Her, which imagines how a juiced-up Siri will change our lives.

Now, tech companies large and small are racing to make this a reality. You’ve read the news. You’ve heard the jargon: AI, machine learning, deep learning, neural networks, natural language processing.

Maybe it’s all a little confusing. So here’s a primer on these concepts and how they’re interrelated.

What is artificial intelligence, or AI?

AI, simply put, is an attempt to make computers as smart, or even smarter than human beings. It’s about giving computers human-like behaviors, thought processes, and reasoning abilities.

There are two kinds of artificial intelligence:

Narrow, or weak AI

That’s AI focused on one narrow task. Weak AI is already all around us. It has beaten us at chess, Jeopardy, and most recently, Go.

Digital assistants like Siri and Cortana are giving us the weather, and self-driving cars are on the road. But they have limits. A self-driving car can’t play chess. Siri can’t read and delete your unimportant emails. Weak AI has a narrow scope: It can’t go beyond its original programming.


Cortana is an example of weak AI.

Strong, or general AI

Here we enter the realm of science fiction. Samantha is the very definition of strong AI. She can learn new things and modify her own code base. She can beat you in chess and drive a car.

The anatomy of AI

So we know general AI is the end goal. How do we get there? Here are five areas it needs to master:

  1. Perception: Like us, a computer needs five senses to interact with the world. But it isn’t just limited to five. It can have senses people don’t possess. X-ray vision? Sonar detection? All possible with machines.
  2. Natural language processing (NLP): Beyond sensing the world, AIs need to interpret spoken and written language. They’ll need to parse sentences and understand their nuances, accents, and meanings. That’s a notoriously hard task given how the same sentence can have different meanings depending on the context.
  3. Knowledge representation: Now that it can sense things – objects, people, concepts, words, and mathematical symbols – it’ll need a way to represent the world in its own brain.
  4. Reasoning: Once it collects data via its senses and connects concepts together, it can use the data to solve problems logically. For example, a chess software senses moves on the board and then works out a gameplan.
  5. Planning and navigation: To be truly human-like, AI must not only think like humans. It should live among us. So a big concern with researchers is to help AIs navigate the three-dimensional world and plan the optimal route. Autonomous vehicles must do this well, because errors can cost human lives.
An absolute beginner’s guide to machine learning, deep learning, and AI

Singapore’s Prime Minister getting a ride in Google’s self-driving car. Image credit: Kenji Soon, MCI.

You can see how these areas play together in fields like machine vision, which is the use of imaging and the analysis of imagery to solve problems. Facebook, for example, parses photos you upload on the social network to suggest who you should tag. And it has become eerily accurate.

A self-driving car will perhaps be the most complex implementation of machine vision yet. It needs to recognize road signs, observe lanes, and look out for cars, objects, and people. It has to work in weather conditions with poor visibility, in both night and day, on worn-out roads as well as brand new.

Tools to get there

These concepts aren’t new. They’ve been described as early as 1956 at the Dartmouth Conferences, which was the seminal event that founded the field of AI.

While it took decades before technology could catch up with our imagination, we appear to be finally on the cusp of an AI revolution, with more venture capital investment, more big tech companies getting involved in R&D, and more everyday use of AI in our lives.

Obvious factors contributing to the rise of AI include Moore’s Law, which has packed more computing power into smaller, more efficient chips. Computing power has reached a point where AI is both functional and cost-effective.


Google make a breakthrough with cat videos. Photo credit: zbeads

Big data is another trend that led to the rise of AI: Google made a breakthrough in 2012 when it fed a neural network tons of data, consisting of stills of 10 million YouTube videos.

As a result, it learned what a cat is without anyone teaching it. It achieved 75 percent accuracy in identifying our feline friends. This wouldn’t have been possible without a corpus of 10 million videos.

When machines learn

Now let’s sort out a couple of concepts that are often confused with one another. Machine learning is an AI technique concerned with learning insights from data and using them to make predictions about the world.

An absolute beginner’s guide to machine learning, deep learning, and AI

Decision tree algorithm. Image credit: Wikipedia

Machine learning is done with algorithms. An algorithm is a series of steps to accomplish a task. Examples of such algorithms include decision tree learning and association rule learning.

The machine learning algorithm that’s setting the world ablaze, however, is the artificial neural network, a technique inspired by how our brain’s neurons function.

It’s even entered popular culture: in the comedy series Silicon Valley, the startup Pied Piper runs its compression service on a neural network.

Here’s a simplistic breakdown: a neural network consists of several layers of neurons. Inputs are passed into the first layer. Individual neurons receive the inputs, give each of them a weightage, and produce an output based on the weightages.

The outputs from the first layer are then passed into the second layer to be processed, and so on. The final output is produced.

Then the magic happens. Whoever runs the network defines what the “correct” final output should be. Each time data is passed through the network, the end result is compared with the “correct” one, and tweaks are made to the weightages until it creates the correct final output each time. The network, in effect, trains itself.

This artificial brain can learn how to identify chairs from photos, for example. Over time, it’ll learn what the characteristics of chairs are, and increase its probability of identifying them.

Facebook’s AI director Yann LeCun explains neural networks with an analogy:

A pattern recognition system is like a black box with a camera at one end, a green light and a red light on top, and a whole bunch of knobs on the front. The learning algorithm tries to adjust the knobs so that when, say, a dog is in front of the camera, the red light turns on, and when a car is put in front of the camera, the green light turns on.

You show a dog to the machine. If the red light is bright, don’t do anything. If it’s dim, tweak the knobs so that the light gets brighter. If the green light turns on, tweak the knobs so that it gets dimmer. Then show a car, and tweak the knobs so that the red light gets dimmer and the green light gets brighter.

If you show many examples of the cars and dogs, and you keep adjusting the knobs just a little bit each time, eventually the machine will get the right answer every time.

An absolute beginner’s guide to machine learning, deep learning, and AI

Image credit: a16z

Now we come to deep learning, which is simply a set of methods for training multi-layered artificial neural networks. It has been found to be especially effective in identifying patterns from data. Whenever the media talks about neural networks, it’s likely referring to deep learning.

A great explainer on machine learning and deep learning:

Deep learning has advanced AI significantly. It’s now being used in many industries beyond software.

Facebook is using deep learning in M, an AI-driven virtual assistant that helps users complete any task – do research, book a flight, and get coffee.

Google is using a deep learning system called RankBrain to filter search results, alongside more traditional signals. As Bloomberg describes:

The system helps Google deal with the 15 percent of queries a day it gets which its systems have never seen before. For example, it’s adept at dealing with ambiguous queries, like, “What’s the title of the consumer at the highest level of a food chain?”

The system has become so useful that it’s now the third biggest signal for Google’s search results, aside from linkbacks and content.

The neural network that identified cats? That’s deep learning.

From Siri to Samantha

Deep learning could be a key puzzle piece leading to the creation of smarter, more human-like AI.

Google’s cat-scanning brain required 16,000 computer processors to run. AlphaGo, the program that beat Go champion, Lee Sedol, was run on 48 processors. In the future, a neural network could be run on a cheap mobile phone.

Deep learning could improve all facets of AI, from natural language processing to machine vision. Think of it as a better brain that’ll improve how computers learn.

It could enhance virtual assistants like Siri or Google Now to deal with requests they’re not familiar with. It could process videos and generate short clips summarizing the content.

Who knows, maybe one day it’ll win an Oscar.

Check out this weird short film written by an AI:

Want to get a great job? Add CV