In my first article about AI, I talked about the emergence of this field. How John McCarthy put before us, a Pandora’s Box to explore. The second article in the series focused on the division of Artificial Intelligence into branches. It talked extensively about Machine Learning and Robotics. This piece will focus on the other two subdivisions of AI- Deep Learning and Expert Systems.
When you hear the term deep learning, just think of a large deep neutral net. Deep refers to the number of layers typically and so this is the kind of the popular term that’s been adopted in the press. I think of them as deep neural networks generally.Jeff Dean, Google Senior Fellow in the Systems and Infrastructure Group
This field is not so commonly heard by people. The common public is restricted to the first three terms usually- AI, ML, and Robotics.
But it does sound similar to Machine Learning. Doesn’t it? Well, Deep Learning is simply a broader class of Machine Learning. It trains a machine to do what comes naturally to humans.
The Human Brain
For instance, let’s focus on the things a baby can do when they are small. They can identify your words. They may not be able to speak fluently, but they can understand you. If two languages- Hindi and English- are spoken in a household, then they can even translate from one language to another. A 5-year-old baby will know that milk and dudh are the same. They will even identify things and ask embarrassing questions.
“They are looking, listening, and imitating from the time they are born. Stick your tongue out at a baby, even an infant just hours old, and he or she may do the same back at you“, says Sarah Lytle of the Institute for Learning and Brain Sciences at the University of Washington.
Did someone tell the baby to do these things? Did the surgeons who delivered the baby program them to imitate you? No. A big fat no.
All of it came naturally to the baby. Why? Because a baby’s brain is miraculous. The brain involves the functioning of Neural Networks. Which is exactly what is required for Deep Learning. What I am trying to say is the brain naturally builds algorithms to perform deep learning.
Deep Learning focuses on these aspects of Artificial Intelligence. It involves recognizing speech, identifying pictures, and even making predictions.
Implications of Deep Learning
At this point you might want to know, how are Machine Learning and Deep Learning different? Well, Machine Learning focuses on organizing data to make predictions on inferences from the data. Deep Learning is more than that, as stated above. Deep learning sets up basic parameters about the data and trains the computer to learn on its own by recognizing patterns using many layers of processing.
Siri, Alexa, and Cortana are amazing implications of Deep Learning. Every time you say something to your smartphone. It records your voice to understand your instructions better. It assembles data like location and search history and the next time you wanna go out, it shows personalized recommendations. This satisfies your needs and makes you happy. The entire field of Deep Learning is focused on one word- Understanding.
Future technologies like Self driving cars, personalized medicines, and facial recognition software are enabled due to Deep Learning.
Modern-Day Applications of Deep Learning
As of today while it does have a multitude of uses, its uses can be summed up in 4 categories.
- Speech Recognition– This doesn’t need much explanation. Companies like Amazon, Apple, and Google are continuously working to make this aspect better. So, soon we might use our hands for just holding our phones. Lazybones, aren’t we?
- Image Recognition– When criminal activities need to be traced, Image Recognition comes handy. Recently, in China, police identified a man kidnapped 24 years ago from his home. And, reunited him with his biological parents. All they had was a baby photo. They used an Image Recognition software to develop a modern-day picture of the baby and lo, behold!
- Natural Language Processing- This involves data mining. Another big word. I understand your pain. It simply means identifying data patterns in a large number of data sets. Common factors in complaints registered by customers to companies are identified via NLP. You really thought they have the time to read all your mails? Sorry to break your shell.
- Recommendation Systems– This application is pretty personal to me. OTT platforms like Netflix, Hotstar + Disney, and Amazon Video use this feature to understand what we might prefer to watch. Amazon Video’s recommendation system doesn’t work as it should. On the other hand, Netflix does an amazing job. This is precisely why, shares of Netflix are rising continuously, in spite of having so much competition.
This was enough to tell you, why tech giants need to embrace Artificial Intelligence as fast as they can. Or, they will lose the war.
Let’s decipher the meaning of this one by ourselves. The Expert System in layman’s language can be referred to as Expert Computer System. The addition of the word computer does make things easier now. A computer system/program that solves problems within a specialized domain is an Expert System. These domains usually require human expertise.
Today medical diagnosis, financial investing, and even petroleum engineering employ expert systems.
To understand better how an expert system work, let’s take an example. An expert system usually contains two parts- data sets/knowledge base and an inference engine. And, the system analyses the circumstances by comparing it with the knowledge base and then use the inference engine to give us results. Let’s say, there is a medical patient with an eye condition. The expert system after taking reports of all medical eye tests will determine the diagnosis. Even here, it will show a sequence of diagnosis in a hierarchial manner. The one at the top will be the most preferable and so on.
Why do Organisations use Expert Systems?
To be fair, the above example makes everything clear. But there are more reasons why even government institutions use Expert Systems-
- Proven AI for all data and processes
- Accurate, scalable and multi-language NLU
- Predictable Costs
- Release Of Information (ROI) in weeks, not months
- Human capabilities become more efficient
- Explainable AI
Those finding the above factors complicated, need not worry. I have pretty much summarised those factors in the above example.
But, there is one disadvantage that comes up when considering expert systems. Expert Systems sometimes make life-saving decisions. So, they need to have a wide range of information. In other words a vast and expansive data set/knowledge base. Acquiring this data set is a tedious task and takes a lot of human effort. Therefore, there is a continuous push for the development of tools to make data acquisition easier. But in no way has human interference in the field declined.
We have talked about the major fields of Artificial Intelligence, and seen their relevance in the modern world. But, there are still aspects of AI that need to be looked upon.
The next article in the series focuses on Internet of Things. Hope to catch you there.