The Game of Checkers

Back on our early AI days, we looked closely at problem solving for games.  Notably the work of Arthur Samuel in machine learning.   This piece below on the seemingly simple game of checkers. Good about the people involved, but not enough about the nature of the AI.  See the last link for that, I apologize for the pay wall there.

How Checkers Was Solved
The story of a duel between two men, one who dies, and the nature of the quest to build artificial intelligence ..... " 

About Marion Tinsley, the world's greatest checkers player
by Alexis C. Madrigal in the Atlantic  .... 

Checkers was the first significant classic board game that had a non-human champion.  But it never defeated the long time human champion in a match.

See Checkers in Solved.  http://science.sciencemag.org/content/317/5844/1518  (Requires Login)

Cooperation of Humans and AI

A key element of neear future tech.   How do we work with AI?   Will it be able to solve 'societal' issues?    Or is that something we have to do ourselves?

The Partnership of the Future
Microsoft’s CEO explores how humans and A.I. can work together to solve society’s greatest challenges. By Satya Nadella

Advanced machine learning, also known as artificial intelligence or just A.I., holds far greater promise than unsettling headlines about computers beating humans at games like Jeopardy!, chess, checkers, and Go. Ultimately, humans and machines will work together—not against one another. Computers may win at games, but imagine what’s possible when human and machine work together to solve society’s greatest challenges like beating disease, ignorance, and poverty.

Doing so, however, requires a bold and ambitious approach that goes beyond anything that can be achieved through incremental improvements to current technology. Now is the time for greater coordination and collaboration on A.I. ... " 

A History of Machine Learning. What is Machine Learning?

Bernard Marr in Fortune: A short history of Machine Learning.   This certainly alerts managers to the direction and speed of how this area is starting to move.   The examples shown are excellent.

My only caution is that 'learning' has two parts, first the ability to find pattern in data and store it away.  And second how to find and extract that data in a given context  and then reapply it to  improve some real world work or decision.    AI some people call it.

So one example given,  inventing a clustering algorithm, is a clever and useful thing.    It has many applications from marketing to manufacturing. The Checkers and Chess playing methods cited used forms of pattern recognition, but these are trivial compared to say, understanding and directing the operation of a company's supply chain operations.

So I caution.  Just like in humans.  Just because we learn, and computers can learn much more and much faster,  does not mean we can now directly solve every problem.  Machine learning is primarily a set of very useful component tools.   Often to augment human decision making.    And even sometimes to replace them.   But much work is yet required.

Machine Learning Reshaping our World

I like the enthusiasm of this, but I disagree somewhat.  True, ML is a kind of learning.  But its still not necessarily easy to get it the right data to learn with, or to configure the ML to make it learn the right thing, or to take the solution and install it into the right business process to provide value.

In DSC, Machine Learning Reshaping our world
by Bernard Marr

Machine learning is inherently different. Rather than telling a computer exactly how to solve a problem, the programmer instead tells it how to go about learning to solve the problem for itself.
Machine learning is really just the very advanced application of statistics to learning to identify patterns in data and then make predictions from those patterns.  This website has a gorgeous visualized walkthrough of how machine learning works, if you are interested. 

Machine learning started as far back as the 1950s, when computer scientists figured out how to teach a computer to play checkers. From there, as computational power has increased, so has the complexity of the patterns a computer can recognize, and therefore the predictions it can make and problems it can solve.  .... "

DeepMind can Beat a Top Go Player

First it was checkers, then chess, and now finally the Japanese game Go.  AI continues to evolve. From Google.  AlphaGo by Deepmind.   Another example of leveraging neural net methods.

" .... This paper published in Nature on 28th January 2016, describes a new approach to computer Go that combines Monte-Carlo tree search with deep neural networks that have been trained by supervised learning, from human expert games, and by reinforcement learning from games of self-play. This is the first time ever that a computer program has defeated a human professional player.

The game of Go is widely viewed as an unsolved “grand challenge” for artificial intelligence. Despite decades of work, the strongest computer Go programs still only play at the level of human amateurs. In this paper we describe our Go program, AlphaGo. This program was based on general-purpose AI methods, using deep neural networks to mimic expert players, and further improving the program by learning from games played against itself. AlphaGo won over 99% of games against the strongest other Go programs. It also defeated the human European champion by 5–0 in tournament games, a feat previously believed to be at least a decade away.

In March 2016, AlphaGo will face its ultimate challenge: a 5-game challenge match in Seoul against the legendary Lee Sedol, the top Go player in the world over the past decade.  ... " 

John McCarthy, Father of AI and Lisp Dies

We met John McCarthy at Stanford University, sometime in the 1980s.  He was a very enthusiastic promoter of AI and in particular the use of game playing systems to deliver 'intelligence'. Checkers and Chess were early implementations.  We used his language Lisp on Sun Workstations to implement knowledge based systems for Product development applications. More.

War Games and AI

I just watched the 1983 film War Games. I closely follow the history of computing, and this is a fun watch to see how much things have changed since 25 years ago. This was one of the first films that positioned AI and security in a generally plausible way. In the realm of that day it was difficult for the average person to search for knowledge to perform a task. At that time the Internet did exist, I had used it as part of the Darpanet. The Web did not, and a high school student would have had a very hard time getting the knowledge they needed.

So what has happened? Now it is possible to search very broadly and very quickly. Defense systems are on separate networks and likely much more secure. Likely. Some of the suggestions of the interaction of real world and simulated systems are impressive for the time, though the security gaps are not credible.

Shortly after I first saw the film I was part of a corporate AI team that sought to implement some of the Artificial intelligence implied by the film. Large scale strategic gaming directed and optimized by computing systems. Taking simple games like checkers and chess and scaling them up to real interactions that include choice making, simulations and fallible people. The promise was a big one. Systems that could run large portions of corporations in a 'lights off' mode. Making product fast, cheaper, better.

Although there were lots of small victories that resulted from these corporate efforts, the big win never happened. What worked best were very narrow applications of AI to corporate process. Difficult too was the attempt to build intelligence that could be readily re-applicable. Bigger wins made it necessary to link many systems, many choices and many indeterminate interactions.

You can readily construct individual rules, simulations, pattern tasks and computations. You can string them together in a program or declare them as a set of knowledge statements. But eventually this falls apart. The sum of these knowledge nuggets becomes less useful in aggregate without some combining structure. Though there is AI theory that covers this, its hard to implement in real systems.

So it's not easy to build generalized intelligence systems. It was also seen early on that it was not easy to insert people in the loop to interact with an intelligent system. More recently, linking decision makers with very complex simulations has shown promise. The AI factor is still not readily available in a generalized way. Will it happen? I think so, but there is still some some fundamental first principle work to do, perhaps another 25 years.

I would love to see examples of how other corporate attempts at the broad implementations of AI succeeded or not. And in particular, what were the key barriers to this kind of work? Part of it is a better understanding of how to integrate low level intelligence tasks.

The film, despite it's age, covers some interesting space and is worth a watch. The final resolution, though, is laughable.