I know what AI is, I mean ML! No no…DL?! Okay, now I’m confused!
It all goes back to the 1950s.
After being familiarized with robots through science fiction movies (ex: Metropolis silent movie in 1927), the concept of artificial intelligence was implicitly developed in the scientists' minds.
Alan Turing was the first one to ask himself:
If humans use available information and logical reasoning to solve problems, why can’t machines?
That’s what he tried to answer through his paper ‘Computing Machinery and Intelligence’.
Back then, it was hard to proceed with his theory because of two main reasons:
- Computers were so expensive, and
- Computers couldn’t store commands or, in other words, they could execute a command but wouldn’t remember what they executed.
During World War 2, Turing built a machine able to break the Enigma code through a machine called the Bombe machine. So far, none of the Artificial Intelligence or Machine Learning terms were used, but scientists, philosophers, and mathematicians were digging deep into this ‘thinking machines’ field, as it was referred to, until the Dartmouth Conference.
So how did we end up with all these terminologies: artificial intelligence, machine learning, and deep learning? That’s what we will be talking about!
But first, let’s check the questions we will answer in this blog:
- When was the term ‘Artificial Intelligence (AI)’ introduced for the first time?
- How is machine learning (ML) different from AI?
- What are some examples of AI that are not ML?
- What is Deep learning (DL)?
A little bit of history…
In 1950, John McCarthey introduced the term ‘Artificial Intelligence’ instead of ‘thinking machines’, through a workshop titled The Dartmouth Summer Research Project on Artificial Intelligence. He defined it as follows: “The science and engineering of making intelligent machines”.
From that time onwards, many definitions of artificial intelligence were developed, but they all align with the fact of AI being a science that studies different ways of building programs that can accomplish a human task, and be considered intelligent. It includes any way allowing rule-based systems to be smart or mimic human behavior, like reasoning, planning, robotics, natural language processing NLP; and experience-based systems using machine learning (ML).
So machine learning is only ONE part of AI and the term was first introduced in 1952 by Arther Samuel when he developed software that learns how to play checkers.
The question here is: What’s the difference between ML and other AI components? And what are some examples of AI that are not ML?
AI non-ML components are mainly about systems to which we feed the necessary rules to complete a certain task (that’s what I meant by rule-based systems).
Like a robot equipped with a sensor, designed to move backward if the distance separating it from an obstacle is less than 100 cm; so it’s more like feeding it a rule as follows: If the sensor measurement is less than 100 cm, move backward.
But some things can’t be explained and learned by rules, but rather by experience. In machine learning, we’re no longer talking about explicit knowledge but more like tacit knowledge.
So all ML is AI but not the opposite!
If we take back the same robot example, machine learning includes the robot learning how to grab a toy by trial and error, or what’s commonly known as ‘without being explicitly programmed’.
AI is the large bubble in which machines can carry out a task and thus be considered intelligent, whereas ML is simply giving the data to machines and letting them learn by themselves.
Before we jump into another terminology, let’s see another example of AI that is not ML!
Here we talk about chatbots. In this scenario, we define a knowledge base of a large number of Q&A, and the chatbot responds to the user to a limited extent.
Try to communicate with a bot of a travel agency, most probably, all questions related to travel and tourism are answered since they form the knowledge base of the bot defined by developers. But if the user asks a question that isn’t included in the knowledge base, the bot will answer something like this:
So what’s different about ML?
You can think of ML as software trying to find hidden information in the data you provide. For example, we all know heart diseases are related to many factors such as BMI, age, blood pressure, etc…. But we don’t know how each of these characteristics is affecting the risk of having heart disease. Here’s where ML comes to the rescue!
We create an algorithm, for example linear regression, assuming that the relation between the probability of having heart disease (Y) and the characteristics (xi) is linear. We feed a sufficient amount of data of patients having specific characteristics with how likely they are to have heart disease. The algorithm starts with random initialization (or any other value) of the parameters of the following relation: Y = a1x1 + a2x2 + …+ anxn + b;
During the learning process, the algorithm uses the data samples to fine-tune its parameters (a1, a2…. an ) aiming to achieve accurate predictions.
But all of this is still very far from what humans can do or how they learn right?
This is where we introduce Deep Learning (DL).
The real question here is not ‘What is Deep Learning?’ but rather ‘How humans learn?’. If we understand how our brain works, we can replicate this process to create more intelligent models.
Think of how we think… 😕
Our brain is made up of millions of neurons having the following structure.
The dendrites receive the input signals x1, … xn and the cell body of the neuron sums up these incoming signals. If the sum exceeds a certain threshold value, the neuron fires, and an electrical signal travels to the next part of the neuron, the axon. At the axon terminal, dendrites of the next neuron receive the outputs y1, … yn.
To imitate how humans learn, we created an artificial neuron that has the same structure and function as the biological neuron.
Here, the first part of the neuron receives the incoming signals, multiplies them with certain weights (wi) to emulate the connections between neurons, and calculates their weighted sum - just like what the cell body does. The second part of the artificial neuron represents the threshold to activate it. This is implemented through a function called the activation function.
Creating many neurons that are connected together gives us the famous Neural Network.
So deep learning is a subset of machine learning in which models are designed in a way to replicate the structure and function of the neuron through a neural network. The more layers the network has, the deeper it becomes and that’s why it’s called Deep Learning.
Conclusion
After reading this blog, we know that Artificial Intelligence (AI) is the science of making machines intelligent by any possible means. It includes applications that can learn and reason as we do.
Machine learning (ML) is a specific subset of AI in which we feed data to machines and let them learn by themselves without any rule or being explicitly programmed.
Deep learning (DL) is a more specific subset of ML where the machine or the model to which we feed our data resembles the structure and function of our brain and is known as a neural network.
You can think of DL as a more complex or sophisticated approach to ML.
Do you mean I can apply this type of learning to improve the performance of my robot? Yes of course! Can you see where the confusion comes from?!😅
Don’t forget to support with a clap!
Start your AI journey today by registering your interest in our 4 weeks Artificial Intelligence Bootcamp
You can join our efforts at Zaka and help democratize AI in your city! Reach out and let us know.
To discover Zaka, visit www.zaka.ai
Subscribe to our newsletter and follow us on our social media accounts to stay up to date with our news and activities:
