The recent triumph of AI program AlphaGo playing against a human, signals just how far advanced analytics has come. What lessons can you learn to get a competitive edge for your business?
David Hodgson, March 15, 2016
Almost two decades ago, in 1997, IBM’s Deep Blue chess playing computer beat the reigning world champion, Garry Kasparov, in a six game match under tournament conditions. The world realized, perhaps for the first time, that HAL of “2001 A Space Odyssey” fame, was going to arrive at some point, though a few years later than cast by Kubrick.
Then, in 2011, IBM’s Watson computer stunned us by winning at Jeopardy. If you haven’t actually seen Watson playing Jeopardy click on that YouTube link; it’s truly awesome. The feeling of invasion is greater seeing Watson, perhaps because we can all imagine playing Jeopardy, and the question and answer approach is so “human”.
Which brings us to current events. Google’s DeepMind research team has developed AlphaGo which beat Fan Hui 5-0 last October. Hui is the current European Go champion and 2 dan master. This was impressive enough, but today saw AlphaGo win 4-1 playing Lee Sodol, the current World Champion, a South Korean 9 dan Grandmaster. Send Lee Sodol a message of support somehow, because being on the coalface of human defeat by computers must be tough.
What is happening here?
Closed system games like Chess and Go are complicated, but have simple rules and a known, although massive, number of variables. There are more possible Go board move sequences than the estimated 1080 atoms in the visible universe. This is a formidable problem, but a different sort to the open ended question and answer format of a game like Jeopardy
AlphaGo’s algorithms use a combination of value-weighted, Monte-Carlo tree search techniques and a neural network implementation. The DeepMind team’s approach to machine learning involved extensive training from both human and computer play. AlphaGo played itself to rapidly learn the outcomes of numerous different options.
Watson used Hadoop to store masses of unstructured information, including the entire text of Wikipedia, that it could search with analytical techniques in real time. Equally significant in Watson’s case is that it was responding to natural language questions that it had first to understand using similar search techniques.
A third powerhouse for change, Facebook is also experimenting with AI systems and has their own Go-playing system Darkforest, also based on combining machine learning and tree search techniques.
Between them and the numerous other AI projects underway in different domains, we have the building blocks for HAL’s arrival.
I hear some saying “So what David?”. “This is interesting to learn about, and with the election I had missed it in the news, but of what importance is it to me?”
DeepMind is targeting smartphone assistants, healthcare, and robotics as the practical outcome for their experimental work with AlphaGo. From their website:
“The algorithms we build are capable of learning for themselves directly from raw experience or data, and are general in that they can perform well across a wide variety of tasks straight out of the box.”
IBM has already applied versions of Watson to practical problems, offers it as a service for anyone to buy and a developer community to encourage experimentation. An example of a practical application is the partnership with Sloan Kettering to fine tune cancer treatment. Similarly DeepMind is partnering with the UK’s National Health Service to improve its services.
Although for specific solutions much secret sauce is often preserved, the framework of these systems is usually Open-Source software. An important component of Watson is the Apache Unstructured Information Management Architecture (UIMA) software. These same tools and techniques will be what disrupt your business soon and you will want to be an early adopter.
Fed with the right data, a Watson-type system could answer new questions that nobody yet knows the answers too. Or applied to real-world problems an AlphaGo-type system could decided on the best course of action given many variables and alternatives. Leading the field in practical solutions IBM calls this ‘cognitive business’ and it is definitely a part of our future.
You Control your Future
In the panorama of the Digital Transformation, AI is out there as a wildcard with seeming limitless possibilities. We are both familiar with, and scared of, these futures because of numerous science fiction dramas. HAL is not here yet, but its coming. For you it’s really a case of whether your company or the competition deploy machine learning systems first. You don’t need an AI system to answer that question.
Image credit: NYTimes