Artificial intelligence (AI) and machine learning (ML). It’s likely that you’ve heard both of these terms with increasing frequency over the past few years, often in the context of big data. You may have also noticed that they’re often used interchangeably, which is erroneous.
In short, machine learning is a subset of artificial intelligence. This means that any application of ML is AI, but this is not necessarily true the other way around.
In this post, we’ll take a deep dive into the definitions of AL and ML and discuss some applications of each.
What is artificial intelligence?
Searching around the web will lead you to multiple academic sources defining AI. While differing in specifics, they agree that AI is the discipline of computers simulating human intelligence. Simulating human intelligence can take a range of forms from actions based on pre-programmed rules, to black-box decision making based on complicated algorithms.
What are the subsets of artificial intelligence?
There is really no definitive list of the different branches of AI, but one way to classify AI branches is by matching them to the components of human intelligence. In addition to machine learning, some of these include:
- Problem solving – Taking a systematic approach to answering a question based on knowledge or information that has been preprogrammed.
- Natural language processing (NLP) – This broadly means reading and understanding human language. The complexity of NLP can range from matching related keywords in a web search to inferring subjective information about an individual through sentiment analysis.
- Planning – Developing a strategy to reach a particular goal. The computer is preprogrammed with facts about the situation and information about the results of individual actions.
- Movement – For years machines, especially those in manufacturing and warehouse environments, have been programmed to perform tasks and navigate through environments.
- Pattern recognition – This programming is often used in the context of vision or language. The computer is programmed to compare what it sees, hears, or reads with a predetermined pattern to make connections. For example, a computer might see handlebars, pedals, and two wheels and determine that it is seeing a bicycle.
What are some broad applications of AI?
Here are some examples of how AI (including machine learning) is used in everyday life:
- Translation – This is an application of NLP in which a computer translates text or speech from one language to another. Machine translation is much more than a foreign language decoder or dictionary, however. Programs must understand rules of grammar, syntax, and the culture of the speakers or writers to provide usable translations.
- Games – We often think of AI in relation to games as computers beating chess or go champions, but the technology also plays a big role in video games. AI is used in video games to change the behavior of non-player characters, based on the actions of the human player.
- Computer vision – This is an application of pattern recognition. Computer vision programs allow machines to capture and interpret images and videos in real time. A computer vision program might be used for security reasons by scanning an area and then notifying a human if something does not line up with pre programmed patterns.
- Sentiment analysis – As mentioned previously, sentiment analysis programs infer information about an individual through their text or writing. This can be used in a customer service context to route emails or chats to appropriate agents based on language cues from the customer.
- Financial trading – Very basic robo advisors are usually rules-based programs that execute buy and sell orders when individual stock prices reach a certain threshold.
What is machine learning?
Machine learning algorithms differ from other subsets of AI in that they learn dynamically through experience. They can modify themselves and improve their predictions or task executions without human intervention when exposed to new data.
How do machine learning algorithms learn?
Most machine learning models are built on sample or training data that it uses to make decisions. As the computer makes decisions or tries to reach a goal, it receives feedback or signals. Ultimately, the ideal machine learning model receives as little correction as possible.
How is machine learning used in industry?
Here are some examples of how machine learning models are used in different industries:
- Fraud detection – With the volume and ease at which frauders can wage attacks, it is necessary for banks and credit card companies to use machine learning to combat this problem. Machine learning models constantly watch and assess customer behavior and quickly send out an alert when suspicious activity is detected.
- Credit scoring – Traditional credit scoring models are typically rules-based systems, but in more recent years machine learning models have been used to incorporate additional types of borrower behaviors.
- Detection – Many healthcare providers are using computer vision models to aid in the detection of abnormalities in X-rays and images.
- Recommender systems – Streaming platforms like Netflix and Spotify provide personalized recommendations for television, movies, and music based on the user’s activity and the activity of individuals like them. As the system receives more information, it is better able to provide more relevant recommendations.
- Dynamic pricing – Machine learning models can help sellers predict a shopper’s willingness to pay a specific price for an item. Retailers are often able to make sales that they would have otherwise lost but just adjusting the price slightly.
- Web page personalization – Page layout can influence an individual’s likelihood to buy. Personalized site models are often based on previous shopping history along with how the shopper interacted with the website in the past.
So, artificial intelligence or machine learning?
It doesn’t make sense to make a comparison between a discipline and a subset of the discipline. One could argue however that machine learning is possibly the future of AI, as much of the current research in the field is focused on this particular area.