According to IBM, Watson’s performance has improved by 240% since it rose to prominence by beating the reigning human champions at the popular US quiz show, Jeopardy, two years ago. The supercomputer is named after IBM co-founder Thomas Watson, and is a project of IBM’s research labs.
Speaking at an open lecture at Wits University recently, IBM’s senior VP and director of IBM Research, Dr John Kelly, said the original intention with Watson was to create a system that would be “as good as humans” at answering any question in any domain. The supercomputer can take a question in natural language and search all of the data that has been fed into its system and find the correct answer through statistical ranking.
Kelly emphasises that Watson, at its core, is a learning machine. “It literally gets smarter the more it’s used. One thing the other Jeopardy contestants didn’t realise is that Watson actually knew more about them than they knew about themselves. It had studied what they knew and their game behaviour. So, in a sense, it had some perception of the environment it was in.”
According to Kelly, Watson 1.0 was the first generation of a cognitive computing system, with the ability to exercise judgment in game theory, and some perception of its environment.It’s Watson 1.0 that has been put to work at Wellpoint and Memorial Sloan-Kettering Cancer Centre for over a year now and, last week, IBM announced the launch of the first two commercial applications that have been created as a result of Watson’s “residency” at these centres.
The supercomputer has been ingesting more than 600 000 pieces of medical evidence, two million pages of text from medical journals and clinical trials, and 1.5 million patient records, while also being taught to analyse and interpret the information.
It might actually be able to debate and present new information to decision-makers. That will have a profound impact on the role of IT in literally every industry.
“One might imagine it would be possible for Watson to pass medical exams in the not too distant future. It has the complete ability to ingest all medical information at any point in time and answer any question a doctor may have against that.”
Looking at what’s next for Watson, Kelly says: “We are working very diligently around multi-modal unstructured data and being able to feed images into Watson, to glean more information.
“The next-generation Watson is a big deal. A really big deal,” says Kelly. “The first-generation Watson in Jeopardy took a single question and presented a single answer. But that is not the way most complex problems present themselves, they certainly don’t present themselves in healthcare that way.
Kelly says images will be the next big thing for Watson in the medical field: “The goal here is to produce a system that will be as accurate, or more accurate, than a radiologist.
“Where we want to go with Watson is not just question and answer, and not just using paths to find things. We have very dense research going on in complex analysis and complex interactions with Watson.”
“When it comes to the Turing Test – the notion that if a computer were to be communicating with you from the other side of a wall, would you know if it was a human being or a computer – in my mind, there’s no question that Watson is very close if not already capable of passing that test today.”
Kelly notes, however, that Watson still has a long way to go in terms of catching up to the human mind. “The brain consumes roughly 20W of power; it’s a very efficient machine considering what it does. By comparison, Watson in Jeopardy consumed 850 00W – it is still leaps away from what the human mind and brain can do.”
Kelly adds that the intention is not to try and recreate the human brain. “But we must produce computer systems that will allow human beings to live in a big data world. If we don’t – if we don’t extract this information – we’re starting to leave a lot of knowledge on the floor, and I think we’ll just be completely overwhelmed by data and actually start to make bad decisions.”
When asked if he foresees supercomputers such a Watson ever actually replacing humans in certain situations, Kelly says: “There will be an explosion in machine-to-machine interactions, and I can see a Watson-like machine taking over decision-making in that type of non-critical situation where the risk level is reasonable. Where I do not see Watson or cognitive systems going is in replacing the final human judgment – that is a very difficult thing to do.”
Kelly says an important factor is that so far, they have found no evidence to suggest computers are capable of being creative. “Humans have this ability that no matter how much data and experience we have, we can still create outside of our knowledge. While Watson has surprised us and we’ve thought ‘How on earth did it know that?’ we’ve always been able to trace it back to where it found the information, we’ve never found it being creative.”
In healthcare, Kelly says Watson will never replace a doctor: “You will always want a doctor as a final decision-maker. But we will reach a point where, as a patient, you will demand that your doctor has access to a Watson because the amount of information that is out there is simply beyond what a human being can possibly know.
“We’re not trying to replace humans; we’re trying to bring a new set of tools to the party that will allow humans to be much more effective in this world of enormous data. As long as in those critical situations which require judgment and creativity, we have a human being involved, it will be just fine.”
According to Kelly, the way healthcare and medical best protocols are produced in large medical institutions, currently involves a group of highly experienced doctors and experts sitting around a table and discussing and sharing different protocols and outcomes, and then deciding on best practice based on that.
“Think about having a very intelligent Watson at that table, as not only a resource to search massive amounts of data and statistical rankings of different protocols. But actually being able to say: ‘You decided that this was the best protocol, but let me tell you why you would probably want to reconsider that.’
“It might actually be able to debate and present new information to decision-makers. That will have a profound impact on the role of IT in literally every industry.”
In terms of the next generation of cognitive computing and Watson 2.0, Kelly says IBM is doing some advanced research based on studies of the human brain.
“We’re trying to understand the patterns within the brain. Neurons and synapses are not a clear network compared to the way we structure today’s computers with 1s and 0s on a bunch of layers of capabilities.
“We are trying to physically produce a new underlying structure for a computer system that will be truly cognitive. We will still be off by about three orders of magnitude in terms of the density of synapses and neurons in the human brain; we’ll also still be off by about two orders of magnitude in terms of power consumption. But this will give us, literally by the end of this year, a new platform upon which to experiment in learning techniques and new architectures that are based on these massive networks rather than traditional computing architectures.”
Kelly says so far the system is capable of recognising simple images, numbers and letters. “This is all with no programming – just pure learning. We think it is possible to build a very interesting architecture that will be more human-like and more biologically inspired than what we’ve built in the past with brute force.
“By the end of the year, we will have an extremely powerful chip and when it’s all said and done we will put hundreds of these devices together into something like a box and hopefully we will have a device that will be a truly learning system and one that is the next step in the era of cognitive computing.”
Our comments policy does not allow anonymous postings. Read the policy here