Machines Can win over Humans?

Machines Can win over Humans?

We may think, winning the world league is like reaching the pinnacle for humans. World champions win these world leagues.

But have you ever realized that even Artificial Intelligence (AI) programs can play games and can win over the world champions?
There are many games where Artificial Intelligence programs have won over world champions!
When you hear this the first time it looks weird! Right!!
Yes, for anyone hearing this the first time, it looks ridiculous, Machines winning over Humans!!??

Well, Artificial Intelligence aided machines can do miracles. These machines can think and apply logic on the fly like humans do!

Today Artificial intelligence (AI) has entered our mainstream lives. We are able to see complex to complex human tasks being demonstrated by powerful AI programs.

One of the biggest proof of the pudding of AI programs coming of age demonstrating human level has been in the domain of games. This is an area where humans have excelled in digesting the potential moves and dynamically deriving winning strategies

This is far beyond the conventional view of automation of many of the things and tasks, which Artificial Intelligence has been demonstrating its capability for a few decades.

Let’s explore some of these complex, human-like tasks, demonstrated in recent past by complex Artificial Intelligence programs.

  1. Poker

We saw in early 2017 how a complete Artificial Intelligence program was able to defeat mankind in playing Poker

Poker is a card game which is oriented towards gambling, involving skill and an extremely complex strategy.  It also involves dynamic human thinking, on the fly strategy formations, flexible optimizations to choose the right path among millions of possible paths. This game is very complex and is played by people who bet according to their whims. Two Artificial Intelligence programs have proved that they can be better in playing Poker. They have recently beaten human professional card players at a popular poker game of Texas Hold’em. The team behind them is known as DeepStack.

  1. AlphaGo 

A similar achievement was demonstrated in yet another complex game namely

GO in October 2015

GO is a highly complex game involving a maze of choices and paths.

Go program is played by two people like in Checkers, One player plays with black stones and the other with white stones.

Players of Go use their logic and understanding of the context, to place the stones.

AlphaGo is an AI program developed by Alphabet Inc’s Google DeepMind in London. This program was used to play the board game Go in October 2015. It became the first Computer Go program to beat professional Go player.

  1. Jeopardy

Before this, we had a highly publicized game show named “Jeopardy” in which the IBM AI software program won the world Jeopardy championship. In 2011, IBM Watson won over the two of the world greatest Jeopardy Champions. Ken Jennings had the won 74 matches and Brad Rutter had won the biggest prize of $3.25 million. These two Jeopardy players were the best in the decade and had won over millions of dollars in the past decade. Watson was named after the founder of IBM.

Watson (the unknown Jeopardy) had spent years of work to contest against these two human champions. This opponent of the two world champions

  • never smiled,
  • nor had any emotions.
  • was kept in a black room with his answers transmitted to the computer in the studio.
  • never spoke, his answers were shown in the text format.
  • was producing a good amount of noise and was kept away from humans
  • Lights on the Watson’s representative desk (with the software connecting to the server) turned green when it was a right answer and orange when it was a wrong answer

Watson consisted of multiple racks of ten Power 750 servers. Most of the time, it was Watson who played correct and not the world champions.

This is a classic example to prove that AI-aided machines can over beat human beings.

The complexity of Jeopardy is in the variety and complexity of the questions that can be generated from the immense world knowledge database.

Watson software was able to digest this volume of world knowledge and convert into answers to the questions thrown at it at an extremely fast pace.

Watson stage replica in Jeopardy! contest, Mountain View, California

Interns demonstrating Watson capabilities in Jeopardy! exhibition match, 2011

IBM Watson

The forebearer of these achievements was the remarkable win of the world chess champion Grand Master “Kasparov” by IBM AI program Deep Blue. This game was played between the world champion Kasparov and an IBM supercomputer-aided with Artificial Intelligence called DeepBlue. The first game was played in Philadelphia in 1996 and won by Kasparov. The second was played in 1997 in New York City and won by Deep Blue. This was the first defeat of a human by a Machine.

With this win of Deep Blue, it was symbolic that Artificial Intelligence is about to catch up to human Intelligence. Some of the critics blatantly denied the power of Artificial Intelligence pointing that Deep Blue relied on the brute computational force to evaluate millions of positions.

  1. Chinook – Checkers

Even before this win, in the past decade, there have been successful instances of AI programs winning over human champions in games. For Ex: The Chinook software defeated the World Checkers Champion.

English Draughts, also known as American 8 x 8 checkers is a group of strategy board games for two players., involving diagonal moves of uniform game pieces and mandatory captures by jumping over the opponent pieces.

This game involves good logic and strategy to win over the opponent. Several notable advances in AI playing games have started with this English draughts. In the 1950s, Arthur Samuel created the first board playing program. In 2007, scientists at University of Alberta created a program called Chinook which won over a human world champion in Checkers.

Exploring the above cases, one may wonder if we can expect AI programs to play physical games like “Kabaddi”, Cricket, “Kho-Kho” etc!! To play games like these, we homo sapiens use our body and mind. In all the above examples quoted, we had a software program and no physical body involved. Well, it is possible to have a Robot play instead of an AI program. It does not look out of reach.

It may be possible that we will soon start watching the Cricket finals of Homo Sapiens Vs Robots!!

About me: I am Neelima Vobugari, founder of Tarah AI. I have more than 15 years of experience in IT in marquee technologies ranging from Mainframes, web technologies, XML to Data Science, ML and AI. In this blog of mine, I will publish articles and videos which help you understand, learn what is ML, AI. These articles are not aimed just for people who want to build their career in AI but also for the entrepreneurs, CEOs who wish to use AI in their businesses. By 2020 I want to help at least 100000 people to learn ML, AI, and use.

Neelima Vobugari

B.Tech, MBA, Data Science Specialist from John Hopkins University

Strategic Lens for Machine Learning

With more digitization in place, data generated and collected is increasing day by day. Each of us uses social media applications like Facebook, Twitter, Linked in etc. Facebook alone generates 500+ terabytes every day. This is the information revealed by Facebook sometime in 2012.

Right now, the number of users using any of these social media sites has almost tripled since that time. The number of Internet of things (IoT ) devices used in various operations in our day to day life has also increased. To process all this data and to get good business insights from them, we use Machine Learning techniques. Machine Learning is defined as something that gives the computers the ability to learn without explicitly programming, from experience.

Machine Learning is a process where we explore, examine and visualize data to get relevant insights, and understanding to solve a variety of learning problems like correlation, prediction, summarization, pattern matching and fault detection.

There are two ways we can classify the algorithms, first based on the Learning type, and the second based on the type of Data the Algorithms are applied on.

Machine Learning applies different algorithms to data to solve problems. These algorithms help computers to learn insights from the data. There are two broad popular categories of Learning

  • Supervised Learning: Supervised Learning is applied to solve those problems where a target or output variable is present, that is, the output variable to predicted on or classified upon is present.
  • Unsupervised Learning: In this learning, there is no concept of Target or output variable. When the algorithm is applied to the data, a pattern is extracted from the data. This data pattern further helps in understanding the data better. A classic example where unsupervised learning is used is the Market Basket Analysis for retail shopping.

The other category based on which the problems can be classified is based on the type of data, Structured or Unstructured.

Data can be structured or unstructured.

  • Structured data is defined as that data which has columns for each data item and the data is labeled. Ideally, a table can be used to represent the same.

Ex: Data from an Excel sheet, RDBMS, Time Series Data, Transaction data and other databases

  • Unstructured Data: This data is free-flowing data without an inherent structure, or This data type is audio, video, text and image format.

Example: Emails, Facebook feeds, surveillance videos, traffic videos, speeches, images from cameras etc.

These twin dimensions give us a lens to view Machine Learning algorithms. ML algorithms can be viewed as a combination of

  • Supervised on Structured Data
  • Unsupervised on Structured Data
  • Supervised on Unstructured Data
  • Unsupervised on Unstructured Data

Machine Learning Decision Matrix

Let us examine each quadrant.

Supervised Learning on Structured Dataà Some of the examples in this category are:

Predicting the time taken to reach your office. Suppose we have a dataset with many parameters like Starting Location, Target Location, Vehicle Type, Vehicle Type, Brand of the Vehicle, Age of the person riding, Gender of the person etc.  It being a Supervised learning, the output variable, i.e., the time taken to reach the office is also part of the DataSet.
Predicting if It would rain or not: Predicting if it rains or not based on the data given. In the input dataset, there could be variables like humidity, cloudiness etc.
Predicting the price of a Stock in Stock Market, based on parameters like past price, segment etc.
Predicting if a student passes the exam or not based on marks, GRE scores etc

B). Unsupervised on Structured Dataà Some of the examples in this category are:

Market Basket Analysis based on supermarket data.
Customer Segmentation based on Structured Sales Data

C). Supervised Learning on Unstructured Data à Some of the examples in this category are:

Predicting if a mail is a spam or a genuine
Classification of digits of images of handwritten text into numbers
Sentiment Analysis using tweets or the text in an unstructured format.
Predicting cancer, based on the image recognition. Predicting if a cancer is malignant or benign

D). Unsupervised on Unstructured Data à  Some of the examples are:

Automatically figuring out if two images are the same creature
Recommendations for movies based on reviews

The other examples in this quadrant are:

Finding different groups of tweets segregated by the topics

Figuring out the trending topics based on the Facebook statuses and/or tweets

So, what we presented is a simple 2 X 2 lens for viewing Machine learning.

The above matrix helps Business Analysts take a systematic approach to analyzing the applicability of Machine Learning for the relevant Business Case. Further down, it can serve as a feed to the Data Scientists to set up the appropriate Machine Learning pipeline to solve the Business problems. These can range from simple segmentation of the data to complex feature detection in videos.

We have termed this Matrix as “Machine Learning Decision Matrix”. Explore it and learn!!

Neelima Vobugari

B.Tech, MBA, Data Science Specialist from John Hopkins University