Let’s take a wide-angle look at where Watson’s been, where it’s going, and how it’s doing.
It was way back in 2011 that Watson burst onto the scene, beating the Jeopardy! champions. Since then, we’ve had a continuing flow of information coming out of IBM about Watson and its growth. But what’s really happening today, and how is Watson being put to work in the real world
Think You Know How Watson Is Being Used?
To find out Watson’s been up to, all you have to do is peruse some of the news sources out there and look for Watson’s name.
For example, Goldcorp and IBM Canada have just announced the development of a new product, IBM Exploration with Watson. This tool uses Watson to analyze mining-site data-set information to help determine the probability of finding gold. The goal, of course, is to increase the “hit” rate in predicting where to dig.
At the same time, IBM has partnered with the open-source framework MITRE ATT&CK to improve the IBM QRadar Advisor with Watson. The goal here is to tie QRadar in with a large database that summarizes types of cyber-security threats and the specific actions that can be taken to counteract them.
Colombia (a part of Times Internet, not the country) is one of the largest ad network platforms in APAC. It chose Watson to improve regional language options in their content.
Guiding Eyes for the Blind even used it to determine what puppies would be the best guide dogs for the visually impaired.
And finally, just to show that Watson isn’t just another nerd, Lexus used it to develop a commercial that should be showing up pretty soon. Reviewing 15 years of video and text related to car advertising, Watson chose the elements that were required to set the tone and content of the spot.
But one of the biggest areas of involvement for Watson has been in healthcare.
With its ability to deal with unstructured language and its question-and-answer architecture, Watson seemed a natural to handle diagnostic work or develop treatment plans. And the main focus in this area has been taking on a real challenge: oncology.
A recent example of this type of diagnostic decision system is the partnership between IBM and Guerbet, where Watson is integrated into Guerbet’s Contrast & Care initiative for liver cancer treatment.
Similarly, Perficient Health Services is using Watson in its cancer diagnosis and treatment decision-making.
Oncology has long been working toward adapting Watson to the medical environment, and IBM and Memorial Sloan Kettering (MSK) Cancer Center have spent the last six years “training” Watson in this area. As of this past summer, MSK felt that Watson was at the 90 percent mark in terms of MSK’s plan matching that of the review board.
Did I Hear a Discouraging Word?
Not all the news about Watson is positive, though.
Take the IBM Watson Health division, for example. Just recently, Deborah DiSanzo, former CEO of Philips Healthcare, who was heading up this division, stepped down and took a position with IBM’s Cognitive Solutions team instead. By itself, of course, that is not necessarily a sign of anything relevant to Watson.
At the same time, however, while IBM revenues as a whole declined 2 percent last year, those of the IBM Watson Health division were down 6 percent. Again, not necessarily Watson’s fault. Or is it? Hard to say.
There have been some negative reports on Watson. For example, an article in STAT, which caters to the life science and medical crowd, recently reported that Watson had recommended some unsafe treatment plans. (Full disclosure: I did not see this originally in STAT but on extremetech.com.) This information apparently came via IBM when working with Memorial Sloan Kettering. The cause of the bad plans was blamed on the training that Watson was given. Initially, it was trained on nonexistent, fictitious cases. Since that time, Watson has been trained on actual patient data. Has it made a difference? I guess we’ll have to wait a year or two until data comes in so we can see.
It’s No Surprise
Actually, stories like the one from STAT should not come as a surprise to us for three reasons.
First, training is the most difficult and time-consuming part of developing a Watson API. It’s also the most critical; it’s the one part that, if not done well, renders Watson’s performance sub par. I haven’t seen anything like it yet, but I fully expect “Watson Training Engineer” to be a valid job title in the future.
Second, all the hype notwithstanding, AI is still in its infancy. Do we really think that in seven years mankind has made the leap to near-sentient software? People who expect everything fast are liable to be disappointed because this area of technology has a long way to go yet.
And third, point number two is especially true when AI is trying to solve problems as complex and multi-faceted as cancer. Not only can there be different types of cancer in different organs, but the way cancer attacks each specific organ can vary from what happens in another organ. The amount of data required to describe something like this is enormous, and the time required to train a machine to make sense of all of the complex and interacting aspects of the disease is daunting. It’s a lot more complex than analyzing bumper damage.
So What Do We Know?
So much uncertainty. Always in motion is the future. But what do we know right now?
First, we know that whatever else it is, Watson is a tool of amazing breadth and power. Some AI tools tend to be focused on one area. Perhaps visual. Or maybe related to interacting with people. But Watson addresses all of these things: visual, auditory, tone, content, etc. And on top of that, one of its key strengths is the ability to deal directly with unstructured data. Yes, whatever else Watson is or isn’t, it is a formidable package of options.
Second, unlike many things that have come along, IBM has a very vested interest in perfecting and improving Watson. Maybe it won’t always carry that name, but the architecture is solid and likely to only grow stronger.
Third, though many kinks need to be worked out, AI and Watson are going to be major players in building the world our children will live in. Sure, the terms and names might change, but the basic concepts and architecture will persist while getting more sophisticated.
So you have to ask yourself, what am I doing, what is my company doing, to be ready to take advantage of those capabilities? It’s not an idle question but rather a starting point for our planning for the next decade.