Deep Learning

Deep Learning refers to algorithms that can become self-learning if they get enough data. A Deep Learning algorithm works in a way that is inspired by the human brain. The way Deep learning is designed architecturally is called neural networks.

The first is a slightly more comprehensive explanation of what Deep Learning is. Deep Learning is by far the hottest in Artificial Intelligence, and is also the branch of development that could move your business the most.


What makes Deep Learning special is that you train the computer system.


Usually, you create IT by creating lines of code in the form of a computer program or an app. That is not how you create Deep Learning applications.


In Deep Learning, you have several sub-components that look for patterns that can explain a given behavior. Sometimes a subcomponent is of great importance to the behavior. Other times it has none or little significance.


Conceptually, one often draws a Deep Learning system in the same way as a brain that processes information. Here’s how:


What is smart about processing data in this way is that the patterns that can describe a user’s behavior the best arise in a learning process.


For example, let’s say that we would like to train a Deep Learning algorithm to understand human handwriting.


You do this by showing such a system may be a thousand different ways in which you can write the number “3” or the word “dog” with various people’s handwriting. Then the system learns which patterns characterize whether a number is “3” or what determines how the number “10” is written in hand.



If you ask a person why they can read the number “3” almost no matter how it is written by hand, then very few people can articulate why they can identify what they see as the number “3”.


But we do not doubt that what we see is the number “3” and not “6”. Deep Learning works with the same kind of logic.  Humans recognize the number “3” correctly because we use our intuition.


Intuition is defined as “the ability to understand something instinctively, without the need for conscious reasoning”. As humans, we use our intuition all the time.


Deep Learning works in ways that are similar to our intuition – It does not use logic as we know it from the IT systems that we use every day.


Deep Learning algorithms work similarly to human intuition. There are no rules in a Deep Learning system that determine how to define the number three written as handwriting.


A Deep Learning system determines that a given number is “3” based on the knowledge from an algorithm. This algorithm has gained this knowledge by learning perhaps a thousand different ways in which the number three can be written, and maybe 10,000 ways in which the number should not be written. So the AI makes a deduction. This number is “3” when confronted with a number “3”.


Deep Learning can predict human behavior

Another great strength of Deep Learning is that it is a technology that can be brought much closer to our everyday lives than conventional IT can.


Conventional IT systems are designed as math and logic-based tools. For them to work, they assume that mathematical formulas can describe the environment they are looking at. It is a useful approach if you develop spreadsheets or design a banking system.


But if you want to make services that interact with people on our terms, that is, speech, text, and non-verbal communication, then the mathematical problem-solving approach will give you problems. Deep Learning is already technologically superior to all conventional IT systems in such areas.

Deep Learning can even be made so advanced that it could make predictions about human intentions and rational behavior. Think about it and its significance, that’s a big deal. There are various examples in the book of how it would be practically possible to work with technology that way for you.

That knowledge could be based on information about what day the meeting was held, the flow of the phone call, how long the conversation was, how many clarifying questions the customer asked, at what point in the conversation did the customer inquire information about the cost? All of the above is information about the customer’s behavior.


Our AI can make predictions by looking at patterns from past customers’ buying behavior and comparing them with our new customer’s behavior.


The AI will look for significant patterns in how the new customer behaved. The system will then make predictions based on this knowledge. Such forecasts could include an expected output of your relationship with this customer, such as how many visits you need before they buy from you, whether you can make upsells and whether they are sensitive to price or not.


This means that we can create a set-up where we could be more knowledgeable of a specific customer’s buying preferences than they are aware of. So, we know in advance how they will react to the core elements of the meeting with us.


Your employees would thus be able to know which products the specific customer prefers without even talking to the customer, and the best ways to close sales with him.


This level of knowledge will change the premises for the level of customer service that you could provide.


For example, your Sales Reps will be able to zoom in to your client’s needs quickly, and you will appear professional in the client’s eyes. Or maybe you could offer a virtual sales experience that matched the level of service that an actual sales rep could provide.


It is because of examples like this that some conceptually describe Deep Learning as systems with intuition. The system cannot explain why it thinks a given customer will be interested in a given product. However, if the system has enough data to learn from, it will be able to predict a given customer’s behavior with consistent accuracy.


The fact that AI can learn the meaning of behavior and make qualified proposals for actions based on the knowledge acquired is a big thing.





If you are looking for a more technical approach to Deep Learning, the we suggest you read the book Deep Learning (MIT Press Essential Knowledge series) by John D. Kelleher


You can find the full definition at this Wikipedia page:

This research from the three Stanford professors Andrea Stevenson Won, Jeremy N. Bailenson, Joris H. Janssen shows as an example of why Deep Learning is superior to IT technologies in the most complex area, non-verbal communication.

Carlos Perez describes the concept of Deep Learning and intuition well in his book ”Artificial Intuition: The Improbable Deep Learning Revolution.”

Neural networks


More from the AI Wiki