Artificial intelligence (ai), machine learning and deep learning: what is the difference?
Delve into ‘What is AI’ by exploring the distinctions between AI, Machine Learning, and Deep Learning. Comprehensive, insightful and geared to help you navigate the jargons, join our journey to AI mastery.
What is AI Table of contents
The computer can easily diagnose cancer, drive a car, and learn. Why haven’t the machines taken over humanity yet?
We use Google maps, allow sites to select interesting movies for us and advise us on what to buy. And, in general, we have heard that under the hood of all these smart things — artificial intelligence, machine learning and deep learning. But can you tell one thing from the other right away? Let’s look at the examples.
What is artificial intelligence?
Artificial intelligence is the ability of a computer to learn, make decisions, and perform actions that are characteristic of human intelligence.
In addition, AI is a science at the intersection of mathematics, biology, psychology, cybernetics and a bunch of other things. She studies technologies that allow a person to write “intelligent” programs and teach computers to solve problems independently. The main task of AI is to understand how human intelligence works and model it.
There are subsections in the field of artificial intelligence. These include robotics, computer vision science, natural language processing, and machine learning.
What is artificial intelligence?
Researchers usually divide AI into three groups::
Weak AI (Weak, or Narrow AI)
Weak intelligence — the one that we have already managed to create. Such an AI is capable of solving a specific task. Often even better than a human. For example, like Deep Blue — a computer program that beat Garry Kasparov in chess back in 1996. But such a Deep Blue does not know how to do anything else and will never learn it. Weak AI is used in medicine, logistics, banking, and business:
- Artificial intelligence from Google was able to beat experienced doctors in the accuracy of breast cancer diagnosis. To do this, we used hundreds of thousands of screening results. According to the American Cancer Society, doctors fail to diagnose cancer in about 20% of cases and often make a false diagnosis. AI not only made a more accurate diagnosis than doctors — by 9.4% — but also more often pointed out the disease where oncologists failed to recognize it.
- Amazon, one of the world’s leading AI companies, has developed a Fraud Detector system. It helps fight online fraud, which causes people and businesses to lose millions of dollars. The algorithm monitors user actions in real time, finds them, and reports anomalies — For example, it marks suspicious orders that need to be checked before making a payment. This can be used in banks, online stores, and large companies.
- Thanks to machine learning, Waymo self-driving cars are able to move on real roads without harm to passengers and passers-by. By the way, such cars — though from Toyota — will be used at the next Olympic Games in Japan to transport guests.
These are a few examples, but in reality there are many more applications.
Strong AI (or General AI)
What a strong artificial intelligence would look like can be seen in the game Detroit: Become Human.
In the Detroit universe, robots are able to learn, think, feel, be aware of themselves, and make decisions. In a word, they become like a person. And in everyday life, the closest thing to General AI is chatbots and virtual assistants that mimic human communication. Here the key word is imitate. Siri or Alice don’t think — and are unable to make decisions in situations they haven’t been taught. Strong artificial intelligence is still a dream.
Superintelligence
Not only have we not created superintelligence, but we don’t yet have the faintest idea how to do this, or if it can be done at all. These are not just smart machines, but computers that are superior to humans in everything. Simply put, something from the field of science fiction.
Machine learning: how AI learns
Machine learning is one of the branches of AI science. It uses algorithms to analyze data, draw conclusions, or make predictions about something. Instead of coding a set of commands manually, the machine is trained and given the opportunity to learn how to perform the task itself.
Three things are necessary for a machine to make decisions:
- An algorithm is a special program that tells the computer what to do and where to get data from. For example, we can write a program that sorts pizza: “Margarita”, with mushrooms, with sausage.
- Data set — examples on which the machine is trained. It can be pictures, videos, text, or anything else. In our case, you will need thousands of photos of different pizzas. The more examples, the richer the experience-just like people.
- Signs — what should the computer look at when making a decision? If we are engaged in machine learning with a teacher, we manually select mushrooms and sausage slices. When learning without a teacher, we merge all the data into the program and let the computer figure out where everything is, and correct it if necessary.
There are many different algorithms in machine learning. One of the simplest methods is linear regression. It is used if there is a linear relationship between the variables. Example: the larger the order amount, the more you will leave a tip. Based on the available data, you can predict the amount of tips in the future. Basically, simple math.
There are Bayesian algorithms. They are based on the application of Bayes ‘ theorem and probability theory. These algorithms are used for working with text documents — for example, for spam filtering. The program needs to provide data sets for the “spam” and “non-spam” categories. Then the algorithm will independently evaluate the probability that the words “Free tours for pensioners” and” Book a tour for mom, please ” belong to one category or another.
And then there are neural networks, you’ve probably heard of them. They relate to the methods of deep machine learning, and more on this in a little more detail.
Deep learning: deep learning for different purposes
Deep learning is a subsection of machine learning. The Deep learning algorithms don’t need a teacher, just pre-prepared (marked-up) data.
Artificial neural networks (ANN) are the most popular, but not the only method of deep learning. They are most similar to how the human brain works.
Neural networks are a collection of connected units (neurons) and neural connections (synapses). Each connection transmits a signal from one neuron to another, just like in the human brain. Typically, neurons and synapses are organized into layers to process information. The first layer of the neural network is the input that receives data. The last one is the output, the result of the work. For example, there are several categories, one of which we ask you to include what was sent to the entrance. And between them are hidden layers that perform the transformation.
In fact, hidden layers perform some kind of mathematical function.
We don’t set it, the program learns to output the result itself. You can teach the neural network to classify images or find the desired object in the image. Remember how reCAPTCHA asks you to find all the images of trucks or traffic lights to prove that you are not a robot? The neural network does the same thing as our brain-it sees familiar elements and understands: “Oh, I think it’s a truck!”
Neural networks can also generate objects: music, texts, and images. For example, Botnik fed the neural network all the books about Harry Potter and asked them to write their own. The result is “Harry Potter and the portrait of what looks like a huge pile of ashes.” It sounds a little strange, but at least from the point of view of grammar, this essay makes sense.
Today, neural networks can be used for almost any task. For example, in cancer diagnostics, sales forecasting, face identification in security systems, machine translations, and photo and music processing.
To train a neural network, you need huge sets of carefully selected data. For example, to recognize cucumber varieties, you need to process 1.5 million different photos. You can’t just merge random images or text from the Internet — you need to prepare them: bring them to the same format and delete what doesn’t fit exactly (for example, we classify pizza, but we have a photo of a truck in the data set). Data markup-preparation and systematization-takes thousands of person-hours.
To create a new neural network, you need to set an algorithm, run all the data through it, test it, and repeatedly optimize it. This is difficult and time consuming. Therefore, sometimes it is easier to use simpler algorithms — such as regression.
Let’s sum up the results
Artificial intelligence is both a science that helps create “smart” machines, and the ability of a computer to learn and make decisions.
Machine learning is one of the areas of artificial intelligence. The MO uses algorithms to analyze data and draw conclusions.
And deep learning is just one of the methods of machine learning, in which a computer learns without a teacher implicitly, using data.
Leave a Reply