Introducing the Watson and AlphaGo
Over the past decade a majority of readers might have come across frequent online discussions about “Artificial Intelligence (AI)” and “Machine learning” and A lot has been said about, how AI is going to revolutionize society, put us out of work and create a new world economy. The most recent addition to this conversation was the defeat of world Go champion — Lee Sedol by Google’s Alpha-Go system has brought the been termed as a major milestone in the AI community. Similarly, for more than ten years, IBM’s Watson has been iteratively improved and has been used extensively to make “human like” decisions in wide variety of fields ranging from scientific research to business situations to tackling world social, health and economic issues. As a quick example, these systems have been made to look at microscope slides of blood samples and determine the health of the patient, without the necessity for a trained technician. So what exactly are Alpha-Go and Watson made up of ? What are the building blocks that comprise these oft talked about but little understood systems . This is what I try to explore in this post. Before we get into details on what Alpha-Go and Watson are, let us delve into some finer details on what the terms AI, machine learning and deep learning mean.
Deep Learning, Machine Learning and Neural Networks — The Blood and Soul of Watson and AlphaGo
Artificial intelligence is the development of computer systems to perform tasks that generally required human intelligence. While, AI has been in existence since the 1950’s, building a computer close enough to human intelligence is a momentously difficult task that has not been achieved to this date. For example, if we are shown a distorted image of human face or two images of twins, we would easily recognize the human face or that the people in the pictures are twins. But for a computer, it will need immense computing power, well structured algorithms and a lot of computing time to do the same. Since the 1950’s lot of progress has been in both increasing the computing power multiple folds and in developing algorithms that increase the speed in computation. Several approaches to these problems have been proposed and numerous computational and mathematical tools have been developed, of these tools neural networks and deep learning are some of the prominent ones which are currently dominating the space..
Machine learning is the ability of computers to learn without explicitly being programmed. Computers can use a variety of algorithms to learn in addition to access to large amount of data for learning. Neural networks are algorithms which train the computer based on pairs of inputs and outputs. They are designed similar to how neurons in the human neurons in the brain form stronger connections to learn a specific task or process In simpler words, neural networks infer their own set of rules based on example data to deliver an output. Deep learning is a collection of statistical machine learning techniques that are used learn features and are based on neural networks.
These computers which are built on complex networks are generally tested on complex board games such as chess, go etc. to showcase their abilities and test their limits. IBM Watson has currently been released as a technology platform that can be used to integrate machine intelligence capabilities to software applications just as easily as developers can integrate social network features. The platform is highly useful for revealing insights into huge amount of unstructured data using machine learning and natural language processing algorithms. As a demonstrative example, Imagine simple words, all your company’s legal documents such as PDFs, word documents and web pages are loaded into Watson. Then question and answer pairs are added to train Watson on the subject matter, this training session gives watson its own inherent understanding of the data and updates itself updated with the new information. After this Watson answers questions by searching all possible answers, sorting them using an algorithm and rank them based on quality of evidence. In 2011, Watson’s capability was tested in a game of Jeopardy against former Jeopardy winners in which it won the first prize. AlphaGo is another program developed by Google DeepMind to play the board game Go. The number of combinations in Go game are greater than the atoms in the universe-which makes it the most complex game. AlphaGo uses a search algorithm to find its moves based on knowledge previously learned by machine learning, using an artificial neural network for training, both from human and computer play. AlphaGo’s underlying algorithms are more general purpose in comparison to that Watson’s algorithms.
Digital transformation and Big Data fast track the progress of AI
Both IBM Watson and Google Deepmind AlphaGo have the demonstrated the capability of AI systems to tackle real world problems. The progress in the field of AI undoubtedly demands a large amounts of data for the learning algorithms to work. The digital transformation and the BigData revolutions make vast amounts of data ubiquitous, which acts a catalyst for the progress of AI and thus ensure a disruptive innovation in the fields of healthcare, economics and environment. In my upcoming blog posts, I will try to feature some of the changes happening in these fields due to AI.
