Machine Learning: A Very Good Wing-man

Source: Deep Learning on Medium

Machine Learning: A Very Good Wing-man

It is very common to find data science related jargon everywhere these days. People use a lot of terms inter changeably and cause a lot of confusion to a beginner in the field. Today, through this blog post I would like to talk about the three most popular terms Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL).

“Data is the new oil” , is a very common statement you hear in most lectures related to data science. Why did data become so important? Oil was present since centuries, but its importance was realized after the industrial revolution with cars and factories needing oil to power them similarly companies now need data to power them, hence the importance.

Data was present with companies two decades back also, but they didn’t have access to proper tools or resources to mine the data and draw insights. What is the use of oil being present underground when you can’t mine it?

Machine learning and deep learning are two ways of mining the 21st century oil for insights. In fact the idea behind these techniques were first presented in a ML conference around 60 years back, but most companies didn’t have the compute resources to implement them. Thanks to compute becoming cheaper, techniques like machine learning and deep learning can now be used by companies to mine data. With storage also becoming cheap, the rate at which data is being captured by companies has increased many folds. It is like generating new oil rich areas for later use. Many companies are building their non-curated data rich areas which are commonly known as data lakes now.

Data science is the field related to collecting, storing, describing and modelling of data. As you can see from this definition data science is a very broad field covering a lot of things. Machine learning and deep learning fall under the modelling component of data science. Modelling in data science can be done using a basic statistical approach or algorithmic approach. You generally take the statistics heavy route if interpretations are more important to you and take the algorithmic route if the result and not how you reached to a particular result is important. If you use machine learning or deep learning you are following an algorithmic approach which generally involves making predictions. One point to note here is that some statistical approaches like logistic regression and linear regression are also considered as machine learning models universally.

The word machine in machine learning doesn’t imply the machine does everything here. It is called machine learning because as you feed more data to the model, the model learns over time and can(the word can is important here because if you feed some noisy data to the model, the model won’t learn and predictions would go wrong) give more accurate predictions. Deep learning is a sub-field of machine learning. Deep learning models can capture more complex relationships and generally need a lot of data to work compared to a classical machine learning model. From now on whenever I use the word machine learning it also includes deep learning.

You must be thinking “Oh wait? You told me what machine learning is but what about AI?”

Just like data science, AI is a much broader field. Let’s first define AI and then talk about how it has an intersection with data science.

Al is a field involved in building intelligent systems capable of replicating what a human can do.AI involves the following seven components problem solving, knowledge representation, reasoning, decision making, communication, perception and actuation. The last four out of the seven can be achieved using machine learning.

Machine learning is mainly responsible for the intersection of the two fields of data science and AI. It is acting as a bridge that binds these two fields together. Machine learning models need data which makes knowing the art and science behind the data to build these models important. With the help of these models built using the knowledge of data science it is now possible to build more powerful AI systems. A wing-man is a friend who supports you in forming a relationship with someone. If you consider machine learning to be a friend of data science, then machine learning is like a very good wing-man who made a nice relationship between data science and AI possible.

I believe one component in AI which is causing most of the confusion around these terms is decision making. Data science is required to build a machine learning model and every model used to make a prediction assists in decision making or helps in communication, perception or actuation therefore it can be considered as an AI system. Even a simple classification or regression model built can be marketed as an AI system based on this perception. I am not saying this is completely wrong, but companies should clearly state what the capability of their “AI system” is to the customers so that they won’t be disappointed later. Setting customer expectations right upfront is important because there is a greater chance that people might lose the faith in using machine learning to build intelligent AI systems.

I have also seen people use the words AI and deep learning as synonyms. This confusion was mainly created because deep learning played a major role in solving problems related to communication (which includes natural language understanding and generation), perception (which include computer vision and speech technology) and actuation (putting a machine into mechanical action or motion). Classical machine learning failed to solve most of these problems whereas deep learning worked wonderfully on text, images, and audio. What deep learning has done is that it has brought us closer to artificial general intelligence that is building machines capable of understanding the world as good as humans and with the same capacity to learn a wide range of tasks. What we see in sci-fi movies can maybe become a reality sometime soon because of deep learning.

With advancements in technologies like cloud and open source platforms everyone has the capabilities to build these models at scale irrespective of the size of the organization. If your company is not going to do it, some other company is going to do it and will gain competitive advantage. The cars known as data science and powerful AI, with machine learning as the engine and data as fuel, have the potential to transform the world. I just hope that powerful countries don’t wage war against smaller countries over data which is the oil for these models.