Machine Learning: A Beginner’s Guide

You may think you know about AI, but what about machine learning? Often, the two are confused. What you think is AI may be machine learning getting the job done in everything from email filtering to speech recognition and more. Let’s look at machine learning, AI and how it’s all related.

First, what is AI?

Before understanding machine learning, you must know a little bit about AI.

When we think of artificial intelligence or AI, we often think of computer systems or devices that can, to some degree, behave like humans, at least in how they perform specific tasks.

AI consists of various computer solutions, including machine learning.

The difference between AI systems and traditional computer programming is found in the name — artificial intelligence. The human brain may not be as fast as a computer or as easy to access information. Still, the flexibility of our cognitive capabilities has long kept us ahead of computers–at least until recently.

AI is designed to mimic the one thing computers don’t have: human intelligence.

Computer scientists have long looked for a way to create AI that mimics human intelligence, specifically our decision-making ability and a few other skills. The human brain is nimble. It can make choices and predictions based on past experiences, resulting in better performances than many computers for certain tasks.

AI is a broad term used to describe a computer system made of many different algorithms and solutions to create a more extensive, central system that mimics human intelligence to increase the performance of certain tasks. AI’s replication of specific human behaviors, in theory, increases speed and productivity, etc. For a solution to be AI, it must have a very high level of autonomy in decision-making, i.e., it should be able to do tasks the way a human would in certain situations without actual human intervention. Artificial intelligence automates many tasks, powering its algorithmic system with human-like reasoning, perception, the ability to learn from experience, how to infer, and human language skills, to name a few.

 

Different algorithms and tools are used by computer scientists so that AI systems can be used to automate specific tasks and make predictions based on “experience” (i.e., past interactions.) It can recognize speech, do facial recognition, translate, and make decisions.

Virtual customer service and AI

One typical example of AI you’re familiar with is automated customer service systems. A virtual customer service rep using AI listens to your request, speaks in your regular voice, then sends your call to the correct department. AI systems like these and others are powered by various AI subfields and AI-related tools, including natural language processing and machine learning, which we’ll discuss more below.

Machine learning and how it relates to AI

To clarify, all machine learning is a type of AI, but not all AI uses machine learning. That’s because machine learning is a component of AI systems—it’s one of the tools used to make AI. Machine learning, deep learning, natural language processes, neural networks, and a few other subjects are related to AI–but not the same.

 

Devices and tools using AI systems are often described as having speech recognition, translation, grammar check, and more. These now standard device and app features are only available via various algorithms and solutions used to create AI. Many apps are only possible because of the different types of machine learning used to build AI.

Think of machine learning as a nesting doll

Often, machine learning’s relationship to AI, deep learning, and other AI-related topics are best explained using a Russian nesting doll example. We use this now to depict how various tools used to create AI systems relate to each other.

 

  • AI – the most giant Russian doll; everything nests within it. AI is a combination of technologies designed to replicate human decision-making to help autonomous systems make better choices and subsequently maximize results
    • Machine learning – the next doll lives just inside AI. It’s a subfield of AI. There are different methods for building machine learning, including supervised and unsupervised ML. Nested within machine learning are neural networks and deep learning.
      • Neural networks – a subfield of machine learning, these algorithms are a vital part of what makes AI networks “think” like humans.
        • Deep learning – a type of machine learning, is a subfield of machine learning. It differs in the type of data it uses. Other machine learning uses data sets that are more structured. Deep learning requires less human intervention, less help deciding what data to use, labeling of data, and more.

 

So, machine learning is a subfield of AI, one of several tools different computer scientists use to create the different types of AI systems available. Meanwhile, deep learning is a subfield of machine learning. And neural networks are algorithms that help AI systems seem more human.

More facts about machine learning

As a branch of artificial intelligence and computer science, machine learning algorithms are used to help teach AI how to “learn,” how to understand and imitate the way humans learn, to improve system accuracy and efficiency.

ML even gives artificial intelligence the ability to “learn” on its own, without explicit direction by humans. Machine learning uses mathematical modeling to help AI systems learn and act without direct instruction. The AI systems are trained and improved based on each set of data it ingests. Algorithms used for machine learning are built on the training data—sample data used to help them make decisions or predictions without being told to do so by humans. These algorithms look at mountains of data, leveraging the info to improve the performance of specific tasks, just as a person learns from their mistakes.

Neural networks

Neural networks get their name from the neurological component of humans related to the brain. That’s because artificial neural networks (ANN) are designed to help AI mimic the system used by the brain. Neural networks are made of individual node layers that send data to the next network layer when activated. Some early forms of neural networks date back as far as 1958. Neural networks are a tool that machine learning is composed of. There are several types of artificial neural networks, but they’re often called neural networks or ANN.

Deep machine learning

Deep machine learning or deep learning is a type of ML. Machine learning becomes deep learning based on the number of layers in the neural network used to make it. Deep machine learning is traditional machine learning with more layers.

 

Deep learning or deep machine learning also differs in the unstructured data it uses. Deep learning requires less pre-processing of data associated with regular machine learning.

 

Both traditional ML and deep ML have their uses. But due to the ways they differ, one method may be better suited for certain tasks while another for other activities.

What is natural language processing?

Natural language processing (NLP) is a type of machine learning that helps AI read and understand how humans communicate. NLP uses elements of various fields, including linguistics, computer science, and AI, to help computer systems communicate with humans more naturally. It allows computers to understand written and spoken words the way people use them.

So even though a computer uses coding languages based on 0’s and 1’s, its system can understand your verbal or written statements. This lets AI understand human language and successfully extract information, glean insights, categorize, organize, and more to complete the necessary task. However, not all types of NLP use machine learning for their training.

Some ways the discovery of natural language processes has evolved computers include:

  • Autocomplete and predictive typing (how search engines like Microsoft Edge and Google Chrome know what you’re typing before you’re done)
  • Machine translations – language translation apps like Google Translate, etc.
  • Auto-correct for word processing – Grammarly, Microsoft Word, etc.
  • Chatbots – NLP lets computers talk to you during customer service interactions, more
  • Handwriting recognition is another example of how NLP is used
  • Virtual personal assistants – Siri, OK Google, Cortana, and Alexa are just a few using NLP to understand your requests and respond in a similar, human-like manner.
  • IVR tech – interactive voice response technology used in call centers, i.e., automated phone systems you can speak to before being connected to a “real” human
  • Predictions – using data to figure out what’s next, used in financial/investment systems, for analytics, etc.
  • Recommendation engines – like BBC, SVT, Netflix, Apple Music, etc.
  • Dynamic pricing systems
  • Image classification
  • Anomaly detection
  • And more

 

Any app you use that understands your written commands, talks/listens to you, translates, corrects, recognizes your handwriting, and makes predictions or decisions for you—making life easier—probably using some form of NLP, a machine learning tool used in AI.

Past experiments in machine learning

One of the earliest demonstrations of machine learning was published in 1952. Anthony Oettinger created the AI called “Shopper” from the University of Cambridge.

The shopper was a simulated version of a mall with eight shops. The AI could “search” the mall and visit shops until the correct item was found. Each time the system “shopped,” it would memorize a few more items stocked at each shop. This is how the AI learned. Eventually, it could immediately go to the correct shop without looking at random shops, just as a person would.

Early computer games led to a win for machine learning theory

Early examples of artificial intelligence with machine learning can be found in the first computerized versions of draughts (checkers) and chess.

The first successful AI program came about in 1951, written by Christopher Strachey, who later became a director of the University of Oxford’s Programming Research Group. By using a University of Manchester England computer called Ferranti Mark I, Strachey created AI that could play draughts or checkers. In 1959, Arthur Samuel, another writer of early AI computer games, first popularised machine learning during his research on self-learning programs.

 

Decades later, in 1997, Deep Blue became the first computer system to beat a world chess champion, Garry Kasparov. And in March 2016, Google’s AlphaGo beat Go champ Lee Sudol, the first time a computer had won the complicated game against a human without using game handicapping. By the 2010s, this same system was teaching itself how to play many Atari games without human intervention.

Data training and machine learning

Data scientists and developers use maths, like probability, statistics, calculus, and data, to create models to train their machine learning algorithms. When it comes to AI and machine learning, there are generally two methods of training.

  • Supervised learning
  • Unsupervised learning

The two differ in how systems using each method will use data. During supervised learning, a data scientist interacts with the algorithm, teaching it what conclusions it should make using labeled data sets. The algorithm reads the tagged information to determine items without designated outputs or labels.

During unsupervised learning, there is less human interaction and less data labeling. The system would partially or wholly teach itself to make determinations based on the data sets, not by what it’s told.

Machine learning helps AI make predictions

Earlier predictive technologies depended on having lots of historical data, but now machine learning can make predictions with less info than used by past systems.

Today’s robust machine learning tools empower AI technologies to provide various predictive tools for everything from automated stock trading and predictions to apps that determine lifetime customer value and more.

On a larger scale, the World Economic Forum recently reported that similar predictive statistical algorithms one day would be used to successfully forecast extreme events like earthquakes and pandemics using machine learning algorithms, even with limited on-hand data for training.

Some of the latest ways AI and machines are solving problems

AI and machine learning can be used for all sorts of things.

Whisky-making and machine learning

In 2019 Swedish whisky maker Mackmyra released Intelligence, touted as the first whisky in the world created using an AI program. The distillery worked with executives at Microsoft and their Microsoft Machine Learning Studio to create the unique whisky using recipe #36, taken from an algorithm the system developed.

Streaming services and search use AI to make recommendations

Take SVT, BBC, or other streaming services and their recommendations–created by AI systems. AI looks at your past viewing habits to “guess” what you’d like to watch next. Search engines like Microsoft Edge and Google Chrome use AI to curate your feeds for news results, entertainment, and advertising. Social media curates your feeds based on what you’ve clicked on in the past.

Companies like these and others use machine learning and AI to better understand customer needs.

How AI makes personalized, programmatic advertising possible

AI and machine learning can be applied to about anything and improve it. This is especially the case with digital advertising, which depends on data and fast decision-making for the best results.

Machine learning significantly improves digital advertising systems by mimicking the decision-making abilities of humans. AI and machine learning can read and analyze the data to make cost-effective ad buys, auto-select the best keywords, optimize, test, and more.

It can even write the ads for you–all in the blink of an eye.

AI and machine learning are how we at Amanda AI can offer real-time advertising tailored by geo-location and other buyer traits. If the data says it should, it’ll create multiple versions of an ad based on locale, etc., for you–without asking. That’s why digital advertising performs so much better than regular static ads.

 

Many of these systems:

  • Offer better personalization tools
  • Automate the decision-making for better bidding, keyword optimizations
  • Create tailored ads based on improved data
  • And more

 

Amanda AI uses machine learning to energize and optimize your ad campaigns

Our advertising robot will help with your digital advertising using the power of machine learning and the experience of advertising pros. Contact us today to learn more about the best digital advertising options for you.

Related content

Google Ads Consent Mode v2 is dominating conversations in the advertising space. This new framework will provide further protection for user data while promising to help advertisers better measure conversions. But is it all it’s cracked up to be?
Many businesses struggle to get results from paid search because they excessively or exclusively bid on brand keywords. This article will show you why you need to target generic keywords instead.
The Iceberg Effect can hurt the profitability of your PPC ads because you end up paying for irrelevant search terms. But what role does copy play in the phenomenon? Let’s find out.

In this article

Stay updated

Get on the (email) list!

Stay updated

Join our newsletter