Game Over: Kasparov and the Machine
Imagine this – The year is 1997. In a scene out of a sci-fi movie, world chess champion Gary Kasparov faces off against a machine in a rematch of a game he had won a year before. In the second game, Kasparov struggles with the black pieces, but sets a trap that most computers fall for. Deep Blue doesn't fall for it and wins to level the match. The next three matches end in draws, with Kasparov appearing to weaken psychologically. Deep Blue goes on to win the decisive sixth game, marking the first time in history that a computer defeats the World Champion in a match of several games.
Deep Blue's win was seen as symbolically significant, a sign that artificial intelligence was catching up with human intelligence, and could defeat one of humanity's great intellectual champions. Subsequently, in an even more challenging competition between man and machine, Google DeepMind's AlphaGo program defeated the European Go champion Fan Hui in 2015 and then went on to defeat top-ranked Lee Sedol in 2016. While Deep Blue mainly relied on brute computational tactics to evaluate millions of positions, AlphaGo relied on neural networks and reinforcement learning.
What is artificial intelligence?
The concept of AI began with Alan Turing's seminal work, "Computing Machinery and Intelligence", which was published in 1950. In this paper, Turing, who is famous for breaking the Nazi's ENIGMA code during WWII and often referred to as the "father of computer science", asked the question, "Can machines think?" From there, he offered a test, now famously known as the "Turing Test", where a human interrogator would try to distinguish between a computer and human text response. While this test has been the object of much scrutiny since it was published, it prevails as an important part of the history of AI as well as an enduring concept within philosophy.
While definitions of artificial intelligence (AI) have abounded over the last few decades one of the noteworthy elucidations comes from Stuart Russell OBE and Peter Norvig who wrote and published ‘Artificial Intelligence: A Modern Approach’, which became one of the leading textbooks in the study of AI. In it, they classify AI into two approaches, which differentiate computer systems on the basis of rationality, and thinking vs. acting. These approaches are as follows:
- Human approach:
- Systems that think like humans
- Systems that act like humans
2. Ideal approach:
- Systems that think rationally
- Systems that act rationally
Alan Turing’s definition would have fallen under the category of “systems that act like humans.”
In its simplest form, artificial intelligence is a field which combines computer science and robust datasets, to enable problem-solving. It also embraces the sub-fields of machine learning and deep learning, which are frequently mentioned in the context of artificial intelligence. These disciplines are comprised of AI algorithms which create expert systems which make predictions or classifications based on the input data.
Today, as expected of any new emerging technology in the market, a lot of hype still surrounds AI development. As noted in Gartner’s hype cycle, product innovations such as self-driving cars and personal assistants, follow “a typical progression of innovation, from overenthusiasm through a period of disillusionment to an eventual understanding of the innovation’s relevance and role in a market or domain.” As conversations converge around the ethics of AI, we begin to see the initial glimpses of the trough of disillusionment. Lex Fridman, the Russian-American computer scientist, AI researcher, podcaster and new age poster boy for AI science, declared in his MIT lecture in 2019, ‘We are at the peak of inflated expectations, approaching the trough of disillusionment.’
The good news however, is that once we make it through this period, we can expect a greater popular grasp of the relevance and role of AI, across the board.
Deployment of Artificial Intelligence in businesses:
IBM has developed the AI Ladder for successful artificial intelligence deployments in businesses. The steps towards this end involve -
- Collect: Simplify data collection and accessibility.
- Organize: Create an analytics framework which is ready for business.
- Analyse: Build trustworthy and scalable AI-driven systems.
- Infuse: Integrate and optimize systems to span the entire business framework.
- Modernize: Bring the operational AI applications and systems to the cloud.
This system provides enterprises with the AI tools they need to transform their business systems and workflows, and significantly improves automation and efficiency.
One cannot talk of Artificial Intelligence without wondering about the technology which brings it about. Machine learning is a branch of artificial intelligence (AI) and computer science which uses data and algorithms to mimic the manner in which humans learn, gradually improving in accuracy. Arthur Samuel is credited for coining the term, “machine learning” with his research around the game of checkers. His inspiration, Robert Nealey, the self-proclaimed checkers master, played the game on a computer in 1962, and lost to the computer. Compared to what can be done today, this feat seems trivial, but it is considered a major milestone in the field of artificial intelligence.
In this new century, the technological advances in storage and processing power have enabled some excitingly innovative products based on machine learning, such as Netflix’s recommendation engine and self-driving cars.
Data science is practically based on the foundation of machine learning. With the aid of statistical methods, algorithms are devised in this science, to make classifications or predictions, and to uncover key insights in data mining projects. Decision making within applications and businesses are driven by these insights, and positively impact key growth metrics. As big data continues to expand and grow, the market demand for data scientists is increasing. They are required to help identify the most relevant business questions and the data necessary to answer them. In other words, the horsepower of machine learning needs the driving skills of data scientists.
Machine Learning vs Deep Learning
Aside from machine learning, deep learning is also a term often bounced around in the field of AI. Deep learning and machine learning are often used interchangeably, so it is worth noting the subtle nuances between the two. While both deep learning and machine learning are sub-fields of artificial intelligence, deep learning is actually a sub-field of machine learning.
The basic difference between deep learning and machine learning lies in how each algorithm learns. Machine learning which is not ‘deep’, is more dependent on human intervention to learn. Deep learning automates much of the feature extraction aspect of the process, and does away with the manual human intervention required. This enables the use of larger data sets. You can think of deep learning as "scalable machine learning" as Lex Fridman noted in the same MIT lecture mentioned above.
"Deep" machine learning can distinguish different categories of data from one another. Unlike machine learning, it doesn't require human intervention to process data, allowing us to scale machine learning in more interesting and productive ways.
How machine learning works
UC Berkeley breaks down the learning system of a machine learning algorithm into three main parts.
- A Decision Process: A recipe of calculations or other steps that takes in the data and “guesses” what kind of pattern the algorithm is looking to find.
- An Error Function: A method of measuring how good the guess was by comparing it to known examples (when they are available). Did the decision process get it right? If not, how do you quantify “how bad” the miss was?
- A Model Optimization Process: A method in which the algorithm looks at the miss and then updates how the decision process comes to the final decision, so that next time the miss won’t be as great.
Machine learning methods
Machine learning models fall into different categories.
- Supervised machine learning: Supervised machine learning is defined by its use of labelled datasets to train algorithms to classify data or predict outcomes accurately. As data is fed into the model, the model weighs its options and adjusts its decisions until it has been fitted appropriately. Problems such as classifying spam in a separate folder from your inbox can be solved with the assistance of supervised learning. Scaled up, this translates to time, effort and money saved for organizations. Neural networks, naïve bayes, linear regression, logistic regression, random forest, and support vector machine (SVM) are some of the methods used in supervised learning.
- Unsupervised machine learning: Unsupervised machine learning uses machine learning algorithms to analyse and cluster unlabelled datasets. In the case of these algorithms, hidden patterns or data groupings can be discovered without the need for human intervention. This method is ideal for exploratory data analysis, cross-selling strategies, customer segmentation, and image and pattern recognition, as a result of its ability to discover similarities and differences in information. It is also used to reduce the number of features in a model through the process of dimensionality reduction. Two common approaches to achieve this aim, are principal component analysis (PCA) and singular value decomposition (SVD). Neural networks, k-means clustering, and probabilistic clustering methods are other algorithms used in unsupervised learning
- Semi-supervised learning: To avoid the pitfalls of either supervised or unsupervised machine learning, a happy medium emerged in the form of semi supervised learning. In this system, a smaller labelled data set is used to guide classification and feature extraction from a larger, unlabelled data set, during training. Semi-supervised learning can resolve the problem of not having enough labelled data for a supervised learning algorithm, hence also reducing the cost of acquiring such data.
- Reinforcement machine learning: This is the most exciting machine learning model that is akin to supervised learning, differing in sample data not being used for training the algorithm. This model uses trial and error to learn as it goes. A series of successful outcomes to the resolution of a problem will result in a reinforcement of decision and provide the best recommendation.
A good example of this is the IBM Watson system which won the Jeopardy! challenge in 2011. When a question is put to Watson, more than 100 algorithms analyse the question in different ways, and find the different plausible answers to it at the same time. Another set of algorithms ranks the answers with a score. Watson finds evidence that may support or refute each possible answer. For each of hundreds of possible answers, it finds as many bits of evidence and then with hundreds of algorithms grades the degree to which the evidence supports the answer. The answer supported by the best evidence earns the highest rank, which then becomes the proposed correct answer. All of this is done by the Watson computer in just about three seconds.
Ken Jennings, famous for winning 74 games in a row on the TV quiz show, acknowledged the obvious with a quote from the ‘Simpsons’ - “I, for one, welcome our new computer overlords”, he wrote on his video screen.
Real-world Machine Learning and Artificial Intelligence Applications – The future is here:
Machine learning and in turn, Artificial Intelligence are here to stay and are already very much a part of our lives. A few examples of ML and AI you might encounter everywhere these days, are -
- Customer service: Chatbots are replacing human agents online, everywhere along a customer’s journey, changing the way we think about customer engagement across social media platforms and websites. Frequently asked questions (FAQs) about topics such as shipping are frequently answered by chatbots, and they also provide personalized advice, cross-selling products or suggest sizes for users. Other examples include messaging bots using Slack and Facebook Messenger, virtual agents on e-commerce sites and tasks usually done by virtual assistants and voice assistants.
- Recommendation engines: Using past consumption behaviour data, AI algorithms can help to home in on trends which can be used to develop more effective marketing strategies. This approach is used by online retailers to make relevant product recommendations to customers during the checkout process for online retailers.
- Computer vision: Computers can derive meaningful information from digital images, videos, and other visual inputs through computer vision, and then take appropriate action. This application is powered by convolutional neural networks, and has applications in radiology imaging in healthcare, self-driving cars in the automobile industry and photo tagging on social media.
- Speech recognition: This application is also known as automatic speech recognition (ASR), computer speech recognition, or speech-to-text. It has a capability which uses natural language processing (NLP) to generate written text from human speech. Many mobile devices incorporate speech recognition into their systems to conduct voice search or to improve accessibility for texting. Eg. Siri, Alexa etc.
- Fraud detection: Machine learning can identify suspicious transactions for banks and other financial institutions. A model can be trained using information about known fraudulent transactions through supervised learning. Transactions that look atypical and deserve further investigation can be identified through anomaly detection.
- Automated stock trading: AI-driven high-frequency trading platforms make thousands or even lakhs of trades per day without human intervention, through automated stock trading which is designed to optimize stock portfolios.
ML and AI are here to stay
The world would have stopped and stared dumbstruck when Gary Kasparov was defeated by Deep Blue, but frankly, the incident probably didn’t have coverage enough to warrant such a global reaction. Today however, every little chess-move or activity anywhere in the world can be observed in minute detail by every single person on earth. Conversely, each bit of information about each of us is available to corporations and industries which can significantly alter the precision of our need fulfilment. We casually use equipment and services which tap into big data to optimise our business experience and don’t even notice the impact of ML and AI on these new technologies. But the fact of the matter is that we have come to a stage of development in civilization where we cannot go back to crushing information manually. Machines learn, and are developing an ‘intelligence’ which will anticipate and fulfil our peripheral needs, thereby freeing us for bigger and greater achievements in the process of evolution.
The future is a fantasy land of possibilities in which AI and ML not only play a primary role, but could possibly even, under the instruction of data scientists, run the actual show.