Introduction

A summary of what artificial intelligence, data science and machine learning is and some real life examples.

Definitions

So what are AI, data science and machine learning?

There are no universally accepted definitions for artificial intelligence (AI) and data science; machine learning (ML) is generally better defined.

It is not unusual to see all 3 used interchangeably in industrial, commercial, or non-expert settings. Many product descriptions, companies and media outlets use these terms in a very loose fashion.

A recent study analysed over 2,800 European start-ups claiming to use AI and found that only 40% of these actually did. This should alert you to the fact that there are lots of different definitions and there is potential for a lot of argument about what AI, data science and ML are, or are not.

Given the confusion, some clarity is necessary. For this reason, we provide the following simple definitions to give you some sense of how they differ from each other.

Artificial intelligence

Theories and techniques developed to allow computer systems to perform tasks normally requiring human or biological intelligence. (As you’ll see later the intelligence is very limited.)

Data science

A multidisciplinary field that combines statistics, mathematics, computer science and domain expertise to extract relevant insights or knowledge from data.

Machine learning

A field that aims to provide computer systems with the ability to learn and improve automatically without having to be explicitly programmed.

So AI is about performing tasks intelligently, data science is about discovering insights from data, and ML is a means to achieve both through automatic processes. It is easy to see how terms can be muddled: you may ask for AI and get ML because ML is the method that is applied to achieve the AI.

The diagram below shows all three topics in relation to each other and some applications and techniques that are applicable to them.

Examples

Here are some examples of daily activities that you might be doing and that rely on AI, data science and ML.

Search engines

Google, Bing, and other search engines use sophisticated ML methods to find and rank webpages that match your search criteria. These engines not only use ML to provide relevant results for you, they also combine data science and ML so every time you search for something, the algorithms at the backend will monitor your responses:

  • which pages do you open

  • how many do you open

  • how long you stay in each

This way, these engines can tailor the search results to you.

Virtual personal assistants

Have you used Alexa, Siri or Google Home? All of these virtual personal assistants apply data science to complete tasks such as answering simple questions, telling you the news or weather, or playing music or podcasts. To do this they collect information about what you are saying, and also about when, where and how you are saying things. These assistants then use this information to produce results that are tailored to your preferences. They also use ML to:

  • understand you (speech processing and understanding)

  • improve their performance based on your previous interactions

  • communicate back to you (dialogue management)

Traffic status

Ever wondered how a traffic or map app can tell you which section of your commute will have heavy traffic? That is because they are using the GPS location and speed of their users, then adding it to a central server managing traffic. Data science methods are then used to build maps of current traffic and to estimate the density of the traffic. For areas in which GPS information might not be available, ML can be used to predict regions with heavy traffic using historical data.

Loan approvals

Banks and other financial entities collect extensive information about customers who are applying for loans. Data science is used to find relevant data, while ML is used to classify the customer as eligible or not for a loan depending on their history and the history of people with a similar profile.

Activity trackers

Physical activity trackers, such as Fitbit, collect a vast array of information about their users. Data collected includes:

  • steps covered

  • floors climbed

  • calories burned

  • sleep stages

  • heart rate per minute

Data science is then used to create health stats which, if the user allows, may be shared with external partners (such as health professionals and insurance companies) so they can provide a better and more personalised service.

Chatbots (online customer support)

More and more websites now provide customer support using chats, but quite often the person you are chatting to is a chatbot not a person. Businesses like IKEA, Hotels.com, and E.ON use bots to filter any customers who might need to contact them. These bots use ML to identify relevant information in your text and provide possible answers to your queries. If the bots are not able to provide the information customers need, then they are transferred to a human representative. Duolingo, an app for learning new languages, uses chatbots to help users practice their newly-learned language skills via text messages. They also use data science to collect information about their users and apply ML to classify their personalities and learning styles, with the idea of allocating them to a chatbot that best matches them.

Recommendation systems

Have you ever received an email from Amazon with products that could interest you? Or have you ever seen the ‘Recommended for You’ section on Netflix? These are 2 examples of recommendation systems. These systems collect and pre-process data from your activity within their site, for example:

  • what you search for

  • what you look at

  • for how long

  • what you place in your wish-lists or your cart

  • which parts of a movie you rewind or fast forward

This data produces recommendations based on how your behaviour compares to the rest of users on the site. Using data science, they are able to group customers according to behaviour and share recommendations amongst each group. So, if several people with behaviours similar to yours have watched a movie that you have not, Netflix will recommend it to you.

And of course there are plenty of applications in a professional context including:

  • Classification: for example, classifying images as containing vehicles, people etc

  • Recognition: a common application is facial recognition

  • Filtering: taking large volumes of images, video or documents and selecting those that contain certain images, objects or references

  • Anomaly detection: for example, analysing large quantities of engine performance data and identifying possible anomalies that could indicate a fault

  • Prediction: for example, where we may want to predict when the food is likely to go bad

And so on, the range of applications is growing so fast this list could go on and on.