By Steve Morris, 9 Oct 2012
If you've ever used Siri, Apple's natural language voice interpretation system, and found it wanting, perhaps you should try Watson.
Watson is an Artificial Intelligence (AI) capable of understanding natural language and answering questions on any topic.
Watson shot to fame in 2011 when it competed on the American TV show Jeopardy and beat Brad Rutter, the show's biggest earner and Ken Jennings, the show's longest championship winner. It didn't just beat them. It annihilated them, winning $1,000,000 compared with their respective $200,000 and $300,000. In Jeopardy, the contestants are given clues in the form of answers, and they have to respond with the correct question. Subjects include art, science, popular culture, history, literature and language.
Jeopardy is an interesting challenge for AI. Unlike a chess-playing machine, the rules are much more diverse and ambiguous. It requires cognitive abilities not normally associated with computers, such as the ability to understand double meanings of words, puns, riddles, and inferred hints. It also has to operate in real time.
Of course, the enormity of the challenge was why IBM chose Jeopardy to showcase Watson's capabilities.
Natural language processing, hypothesis generation & learning
Clearly to compete and win in such a situation, some impressive AI must be going on, combined with an encylopaedic knowledge of the world. In fact during the game, Watson had access to 200 million pages of material using four terabytes of disk storage and including the full text of Wikipedia, but it wasn't connected to the internet. You could argue that carrying an encyclopaedia into the game should have been against the rules, but on the other hand that is what the human contestants do in their brains, and Watson's disks are its brain.
Watson was developed by IBM as a tool for helping businesses. It combines natural language processing, machine learning, and hypothesis generation and evaluation to give direct, confidence-based responses.
First it understands natural language. This allows it to read text like Wikipedia pages. Then it generates hypotheses using the vast number of structured and unstructured rules it has created. It assigns a probability to each possible answer, giving not only its best answer, but other possible answers, and its confidence in each one. It also learns from feedback about which answer was correct.
Machines vs humans
This process of finding possible answers and selecting the best candidate is perhaps similar to what humans do when they use their "intuition". It's a pattern-matching process similar in some ways to what happens in the brain's neo-cortex.
One day we may create AIs that operate more like human brains, and on the same scale as human intelligence, but for the moment Watson is the best we have. Of course, machines like Watson may be able to help us make progress towards smarter-thinking machines, in the way that in the Hitchhiker's Guide to the Galaxy, the computer Deep Thought designed its own successor.
Applications - healthcare & finance
In real life, organisations are already using Watson in the fields of healthcare and finance. With Reuters publishing 9,000 pages of financial news every day, no human could ever absorb the available information, so Watson clearly has a head start on investment analysists. I don't know if Watson is allowed to invest the million dollars it won on Jeopardy, but it's certainly helping finance professionals to quantify risk and make decisions.
IBM is reputed to be working on a version of Watson that could run on a smartphone or tablet computer. The present version of Watson runs on ten racks of IBM Power750 servers with a processing power equivalent to 6,000 desktop PCs. That's orders of magnitude too big to fit in a tablet.
One possible solution would be for Watson to do the hard work in the cloud, with a simple voice and image recognition front-end running on the mobile device. Such a system (dubbed Watson 2.0) could make Apple's Siri look like the class dunce.
Looking to the future, it's clear that the capabilities of Watson and similar systems can only increase. Computer processing power and the amount of digital information in the world both double every two years. And as Watson learns from its mistakes and the team at IBM continue to improve its algorithms, who knows what a future Watson might be able to do in a few years time? Perhaps it will be writing articles about technology, and I will be looking for a new job.
Got a question? This is the place to ask it!
Please don't ask a question that has already been asked. Duplicates will be removed.
Please do not use swear words or offensive language, and please, no advertising!
Comment by ted
on 1st Oct 2014
Must be nice :) thanks for the info and the page Respectfully T.M.R.