The differences between Data Science, Artificial Intelligence, Machine Learning, and Deep Learning
Data Science, Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are closely interconnected.
The Venn-diagram shown below visualizes overlapping AI-related terminology.
Here in this post, we will shed light on each one of the following terms one by one:
1. Data Science
2. Artificial Intelligence
3. Machine Learning
4. Deep Learning
By the time you finish reading this post, you will have a fairly good idea of what these terms mean and how they differ from or relate to each other.
Data Science
Data Science, as the name suggests, is about data. It’s a multidisciplinary field focused on drawing INSIGHTS that can help an organization make better decisions.
For instance, by analyzing data collected by a bank, you may discover creditworthy customers or identify financial products that new customers buy.
Today, the availability of huge volumes of data implies more revenues due to Data Science. Using predictive analytics, it is possible to identify hidden patterns in data that you didn’t know even existed.
For instance, a travel eCommerce company may discover that people flying American airlines to Amsterdam opt for a luxury canal cruise tour in the city.
Using prescriptive analytics, the company may further learn that people flying first-class prefer evening cruise while those who fly economy-class book bike tours.
Such data-driven insights can be extremely valuable for targeted advertising or cross-selling strategies.
Different organizations use Data Science in different ways. Here’re some examples of Data Science applications.
You may be wondering why so many of Data Science applications sound like AI applications. Well, this is because Data Science overlaps the field of AI in many areas.
A data scientist uses tools such as statistical modeling, visualization methods, hypothesis testing, and Machine Learning algorithms (AI uses Machine Learning as well).
A typical data science product is based on the application of knowledge in different fields, as shown in the picture below.
In another post, I’ve briefly explained the Data Science Pipeline. You can read it to understand how data scientists build models by applying Machine Learning algorithms on data (No prior knowledge of the field is necessary).
Artificial Intelligence (AI)
To see the big picture, understanding the historical context is important.
The concept of AI has been around since antiquity; there have been numerous myths and stories of inanimate objects endowed with intelligence by expert craftsmen.
The seeds of modern AI, however, were planted by classical philosophers who tried to describe human thinking/intelligence as a symbolic system. The programmable digital computer, the machine based on abstract mathematical reasoning was first invented in the 1940s based on this work.
The term ‘Artificial Intelligence’ was coined by John McCarthy, a computer scientist in 1955. The research in modern AI formally began in 1956 at a workshop in Dartmouth College, a private Ivy League research university in New Hampshire, United States.
AI research at that time was centered on neural networks, inspired by the way neurons work in the human brain. The people who attended this workshop later became leaders in the field.
Up until the 1970s, the field was booming with discoveries but they were far from creating intelligent machines. Initially, they had hoped a machine as intelligent as a human being would exist by the 1980s that could read and understand the world around us, and reason as we do.
But, building artificially intelligent machines wasn’t so simple; the limitations in computing processing power also hindered the progress of AI. For many decades, AI was restricted to research labs.
Machine learning became more popular from the late 1980s to the 2010s. Funding and interest in AI peaked in the early 2000s as major tech giants began building supercomputers and investing in AI. Deep learning became the focal point for AI researchers around the world.
The picture below demonstrates the evolution of AI over the last 6–7 decades.
AI is still under evolution and considered a really wide-term; Machine Learning and Deep Learning are subsets of Artificial Intelligence.
There is no single definition for AI. You can look at AI as the ability that we can impart to a machine to enable it to
- Understand/interpret data
- Learn from data, and
- Make ‘intelligent’ decisions based on insights and patterns drawn from data
An AI-powered machine can enhance its ability based on ‘new data’ that was not a part of the data set first used to train it. For example, an AI-powered CCTV surveillance system for monitoring traffic signal violations can enhance its ability to detect offences on the basis of new camera feeds and corresponding traffic violation tickets.
On a fundamental level, AI is a collection of mathematical algorithms that enable computers to understand the correlation between various data elements. In the surveillance system example discussed above, the data collected and analyzed in real-time can be related to traffic lights, turn indicators, the position of vehicles at a traffic signal, traffic lanes, the gap between vehicles, etc. to arrive on an actionable conclusion i.e. ascertain a traffic signal violation and issue a ticket.
In that respect, an AI-driven machine carries out tasks by mimicking human intelligence. But often, AI goes beyond what is humanly possible.
Machine Learning (ML)
The term Machine Learning (ML) was coined by Arthur Samuel in 1959.
ML is a subset of AI. It is used in scenarios where you need machines to learn from huge volumes of data. The knowledge thus gained is applied to a new set of data. ML gives a machine the ability to learn from (or about) newer datasets without giving explicit instructions.
You can say that ML is the implementation or current application of AI. Some of the most common methods deployed to ‘make machines learn’ are:
- Supervised learning
- Non-supervised learning
- Reinforced machine learning
In some methods, the machine is told beforehand about independent (input) and dependent (output) variables. The machine learns the relationship between these two types of variables by analyzing a set of data referred to as the ‘training dataset.’ Before training a data model, a series of data pre-processing steps are undertaken.
Once a machine has been ‘trained’ enough or when an ML model is ready, it is applied to a fresh set of data, referred to as the ‘test dataset.’
The ML model goes into production mode only after it has been tested enough for reliability and accuracy.
ML involves the use of various algorithms such as simple linear regression, decision tree regression, polynomial regression, K-nearest neighbors, etc. The ML algorithms can be used to approach regression problems, prediction problems, classification problems, etc.
Since ML libraries (e.g. SciKit) have evolved a great deal over the last few years, even programmers with no background in statistics or no formal education/training in AI can start using these libraries to build, train, test, and deploy ML models. But, it’s always helpful to know how exactly different ML algorithms work so that you fully understand what you are doing.
Deep Learning (DL)
You can look at DL as a subset or advancement of ML. DL comes into play when ML cannot fully deliver desired outcomes. Generally, ML is suitable when your dataset is relatively small.
DL is the preferred choice when -
- The data has too many features
- The data is huge in size
- Extremely high level of accuracy is required
In comparison to ML, DL can solve more complex problems but is more difficult to implement, requires specialized hardware (e.g. GPUs) to run, and demands more time to train the model.
DL uses neural network models to understand a large amount of data.
Siri, Alexa, and Google Assistant are some applications that use DL to understand your requests. When Facebook recognizes your friends in a picture or Netflix recommends just the right kind of movies, it’s deep learning at work.
From news aggregation and fake news detection to self-driving cars, natural language processing (NLP), visual recognition, and virtual assistants, DL based applications are now being deployed in many areas.
Deep learning breakthroughs are driving the AI boom. So, yes, Deep Learning IS a big deal right now.
Follow me on: LinkedIn. Twitter.
If you’ve any questions about this topic, please drop them in the comment section and I will be glad to answer any questions or clear doubts that you may have.
