Machine learning is a technique in data science that uses algorithms to ‘teach’ computer systems to recognize patterns and to respond in an appropriate way. It goes something like this:
- Data is collected and supplied to the algorithm as inputs.
- The algorithm identifies patterns based on all data received.
- A response is triggered.
- The response can simply be alerting data scientists to patterns in the data or it can be something more complicated such as activating a program or returning a particular message.
- Data scientists can ‘teach’ a machine learning algorithm to perform its tasks more accurately by indicating whether the algorithm’s results are correct or not.
This might sound pretty abstract, but concrete examples of machine learning can be found all around us! We’ll cover just a few of them below.
1. Automated Chat Bots
Many companies use online chat bots to pare down the responsibilities of their customer service teams. You may have seen one of these while online shopping or browsing a service site. A chat box appears with a message like, “Hi there! What can I help you with?”
Sometimes there is a real person on the other end ready to answer your questions but, more often the first response you get will be from a chat bot. When you enter a message, the chat bot takes the text you typed out as an input and tries to match it with a pattern they are familiar with to supply the correct response. For example, if you type, “does your website have any sales going on?” the chat bot’s algorithm might recognize the word “sales” and decide to direct you to the Sales page.
Input: User-generated text
Response: Answers to questions, useful information, links to relevant web pages
2. Voice Assistants
Voice assistants like Siri, Alexa, and Cortana use a similar type of machine learning as chatbots. However, these devices have the added challenge of translating spoken words to a data type they can process. When you speak a command like, “hey Siri, play classical music,” your device first identifies words it recognizes and then performs the correct response, which would be playing classical music.
Input: Spoken commands
Response: Spoken answers, web results, connecting to an app, making a phone call, playing music, etc.
3. Social Media Feeds
Apps like TikTok, Twitter, Instagram, and Facebook use machine learning algorithms to present users with the kind of content they will like. These algorithms use inputs such as user engagement (likes, comments, video views) to ‘learn’ what type of content each user most enjoys. These apps also look for patterns to categorize content. They may learn to recognize pictures of shoes and categorize them as ‘fashion’ content.
Machine learning in social media does its job correctly by correctly identifying content types and by presenting the right content types to the right users.
Inputs: User engagement, text/image/video content
Response: Categorize content and show it to users who like that type of content
4. Email
Email doesn’t require machine learning to work but, recently, many email platforms, like Gmail, are using machine learning to provide suggestions to users. They can also use machine learning to identify certain messages as spam.
When writing an email response, you may see the email platform suggest a variety of prepared responses like “Sounds good.” or “Thank you!” These responses are developed and proposed using machine learning that looks for patterns in the email received and proposes a variety of correct responses for you.
Inputs: Email text
Responses: Categorize emails (spam, promotions, etc.), supply response options
5. Map Apps
Over the past ten years or so, map apps like Apple Maps and Google Maps have gotten incredibly good at providing accurate and efficient directions to get users from Point A to Point B. Machine learning has contributed to this by identifying patterns in user travel and factoring these patterns into suggested routes.
For example, if a map’s algorithm detects that users have completely stopped travelling over a segment of road that was once heavily-trafficked, it might flag the segment as possibly being under construction.
Inputs: User location data
Responses: Indicate potential road closures, deliver travel routes that avoid obstacles
6. Auto-Correct
Auto-correct can be a controversial tool. Some people want to tear their hair out everytime they have to change “cat” back to “car” but others rely on auto-correct to such a point that they wouldn’t be able to type coherently without it. Many auto-correct tools use machine learning to find patterns in the words that users type and so, learn which words are commonly used on purpose, and which words are misspellings to be corrected.
This most often applies to proper nouns like names and locations. For example, if you live on Carper Road and find yourself typing out your address often you may find that auto-correct frequently replaces “Carper” with “carpet” or “camper.” However, after you ignore or undo the auto-correction a certain amount of times, you may find that the tool eventually recognizes “Carper Road” as a correct phrase and will stop suggesting changes.
Inputs: User-generated text
Responses: Identify misspellings, replace misspelled words with the correct spelling
As technologies evolve to become more effective and accurate and machine learning algorithms are improved upon, we will likely see it used in more elements of our everyday lives in the future. For example, some cities are piloting ‘smart’ traffic lights that can take in data about traffic flow and ‘learn’ to adjust their timing to allow cars to pass more efficiently. Can you think of any other applications of machine learning technology that you use?
Want to learn more about this subject area? Check out our overview page on machine learning!