Is Machine Learning the future?
What is Machine Learning ? What could be the future of Machine Learning ? Well, first of all, let us all get to know what this entire thing is all about.
- What is Machine Learning all about?
The question that has been asked for ages is: Can computers think like human minds? Most of us are led to believe that no matter what a computer can do, it can never be equated with the nature of thinking of the human mind. However, what if a computer or a machine can think and predict results based on analysis of data? Interesting, isn’t it? Let’s see what this is all about.
- With Machine Learning, computers or machines predict results based on the data which is provided to them; not on the basis of a pre-determined equation or model. In short, Machine learning is a process where the outcome of the particular situation or scenario is usually predicted with the help of data provided.
- It is a part of Artificial Intelligence (AI).
- With the progress of Machine Learning, computers would be able to decide the end result based on past results and patterns of data. When no equation can be designed for a particular situation or no code can be programmed to decide the outcome of a particular result, it is at this time that the techniques of Machine Learning come into utility.
- The progress of Machine Learning over the past few years
- Till the 1990s, Machine Learning followed a knowledge-driven approach.
- Post-1990s, Machine learning moved towards a data-driven approach. It was at this juncture that algorithms were being created for computers to analyze results based on data.
- In the year 1997, Deep Blue, a computer designed by IBM, beat the then world chess champion.
- During the 2000s, the concept of Deep Learning started to progress.
- Post-2010, Machine Learning techniques evolved rapidly. Features like facial recognition, voice recognition and many more were being implemented.
- Types of Machine Learning
Let us have a look at some of the types of Machine Learning
- Supervised Machine learning: In this, a set of data is provided to the computer. The computer is trained to make decisions based on this data and arrive at the conclusion predicting what the outcome or the result could be. A practical example of this could be predicting the health conditions of a patient based on his or her past medical history.
- Unsupervised Machine learning: In this, a computer or machine does not arrive at a conclusion on the basis of data provided to it. Rather it observes the data and based on some patterns and relationship between the data, it creates clusters or groups of data. The thing it cannot do is, identify the data group. Example: If images of onions and tomatoes are presented to the computer, it can create data groups of onions and tomatoes and separate them. What it will not be able to do is, identify, which group is of onion and which is of tomato. When a fresh dataset is provided, it will create a group based on pattern observation or add it to an already existing data group. This data group is known as a ‘cluster’ and the technique is called ‘clustering’ of data. One of the useful examples of Unsupervised Machine learning technique is voice recognition, wherein the machine can identify the voice after observing the data patterns and relationship between the data.
- Reinforcement Machine learning: It may not be possible for a machine to always determine outputs on the basis of data sets provided to it. Also, it may be cumbersome for a machine to determine the relationships between the different data patterns and come to a conclusion. In such cases, Reinforcement Machine Learning comes into play wherein the computer (agent) is trained to interact with the environment and predict the output based on certain situations. The best example of this would be when the computer plays a game of tennis against us. It decides according to the situation which would be the best possible shot which can be played. A computer playing a game of tennis expertly is indeed a wonder!!!
- Practical applications of Machine Learning
- Machine learning has exciting possibilities in an evolving world. Machine Learning is already finding practical applications in furthering medical science where the medical diagnosis can be made much better by building a decision set based on multiple real-life medical histories, diagnosis data and treatments. What is getting tested out is if machines can indeed diagnose diseases and suggest treatments better than real medical practitioners.
- Similarly, research in the field of automotive engineering is proving to be a good testing ground for applications like driverless cars. With multiple global companies investing in research in developing technologies linked to driverless vehicles, we are bound to see applications of machine learning in aspects of vehicles to navigate, sense obstacles, change its course and thus human intervention redundant.
- Another example which is an apt illustration of how machine learning works is the success of AlphaGo, a program computed by a London based company called DeepMind, in the game of Go. The rules of the game were fed to the machine along with a past history of games which human had played. The machine was able to pick up the nuances of the game in a short period of time and subsequently, defeat the players in the game.
- Is Machine Learning everything that we need?
With all these exciting applications linked to Machine learning, it is pertinent to take a moment’s pause and reflect upon some of the challenges or dilemmas in this space. For long, the question of Morals or Ethics has posed a question to researchers. Though a decision can be worked out by the machine based on trends and analysis, how does one ascertain that the decision in itself conforms to the moral or ethical decision-making framework that humans would rely on? There have been cases reported wherein the decisions taken by machines have reflected the biases of the researchers who have fed the data sets for the machine to learn from. Whether end users would be in a position to even understand that the decision from a machine involves bias and how the same is to be set right, at any time in the future, are difficult questions to answer and for which, there are no clear answers today. One hopes that with more and more research and funding committed towards development of machine learning techniques, human beings can arrive at applications which enrich human life and in turn, which do not have an adverse impact on human decision making.