Early this year, IBM formed the Watson Group, investing $1 billion to commercialize the Watson cognitive computing technology that captured the public’s imagination in 2011 when it beat two former Jeopardy! champions at their own game. Watson is already solving some challenging problems in healthcare, finance and other professions, and the Watson Group is moving quickly to empower developers of every ilk with the tools needed to build Watson-fueled applications.
So, is Watson ready for your world? Could it help you make faster, smarter decisions about the most important aspects of your business today? How will Watson be melded into other parts of the vast IBM product line for enterprise IT? And how can you start experimenting with Watson today?
[ Related: What Makes IBM Watson So Smart? ]
Mike Rhodin, senior vice president of the Watson Group, answered those questions in a conversation with IDG US Media Chief Content Officer John Gallant at the organization’s new headquarters in New York City’s Silicon Alley. Rhodin explained how his team has worked with early Watson customers and the steps he’s taking to speed the technology’s rollout.
[ Tech Titans Talk: The IDG Enterprise Interview Series ]
CIO: Let’s start by talking about why IBM set up a separate Watson business unit earlier in 2014. What’s the goal?
Rhodin: I think it’s important to step back a little bit farther than that. When we did the demonstration [of Watson] on Jeopardy and played the game show successfully, a buzz started in the market that hasn’t actually died down, which is unusual in the tech industry. The tech industry tends to be built out of lots of products that have very short shelf life. But the concept of Watson and the branding of Watson really have staying power.
When I go on campus and I’m talking to students or doing lectures, I get mobbed with questions, starting from the Jeopardy context. It wasn’t that Watson played Jeopardy. It was that people who watched that demonstration saw a glimpse of a future technology that they thought was going to be meaningful for a new generation. That’s what powered the brand and why it still gets a lot of press. The Jeopardy match stood for the beginning of a new era – a new era of systems that were going to be knowledge-based vs. application-based.
We don’t declare eras of computing very often. We think this is only the third in the hundred-plus year history of the IT industry. The first era was tabulating machines that counted stuff. The second era, which started in the late ’40s, early ’50s, was programmable systems, electronic computers that ran programs. We’ve gotten really good at making those computers smaller, faster and better connected.
That’s what we’ve done for the last 50 or 60 years. But the basic model is the same. The apps that run on your iPhone are running on the same programming models that were around for many, many years. We invent new languages and the engineering has been spectacular over that time period. It’s changed businesses and business models, it’s transformed entire industries and that’s great.
But there are areas of human life and business that that class of systems didn’t touch. We think part of that is because these are areas of human life and professions that are judgment related. You need to have the ability to reason through problems, not just write a program to give you an answer off a piece of data.
[ Related: IBM Watson Analytics Now Open for Business ]
CIO: As well as context.
Rhodin: Right. Context and clarification and dialog and ever-changing information sources and most of the information is not what programs are good at dealing with -- standard relational rows and columns. The information itself is completely unstructured. We didn’t really have systems that knew how to do anything more than store and categorize unstructured data. We could search it, we could find lists of documents but we didn’t have systems that understood the contents of documents.
Coming out of the Jeopardy match we had to do what any technology company does when they have a really amazing piece of research technology -- and at that point in time we had a great piece of technology that knew how to answer one question for one user in under 3 seconds. To be commercializable, you have to answer tens of thousands of questions for tens of thousands of users simultaneously, with all sorts of robustness and tools and you can’t require people to buy a supercomputer. It needed to be a cloud-based system with APIs that enabled people to reach out and use the services.
We spent the better part of two years after the Jeopardy match doing what I would call classic market and technology validation. We evolved the technology, we made it more robust, but at the same time we were working with a half-dozen clients in healthcare and financial services -- brave companies that wanted to work with really raw, emerging technology. Their view was they could get a lot out of learning with the technology along the way. Our view was that we needed help figuring out what the use cases were going to be.
Then last summer we came to the conclusion that the technology had evolved enough and there was a second wave of technology starting to come out of IBM Research that was additive to it. That [told us] it’s time to start the next phase, we really need to start a commercialization project around this. We put that plan in place over the fall and then we launched it in January with the formation of the Watson Group.
When we were putting the plan in place, we came to the conclusion that this new business needed to have a freedom of action different than just another new business unit within the organization. It needed to really act and operate like an emerging startup that had the freedom to make quick decisions, to twist and turn in the market as things were evolving and really move forward at a quick pace. It’s not the first time we’ve done that. Probably the most well-known example was the PC company, when we segregated that and put it down in Boca back in 1981. That became the basis for an entire industry.
Many people today still call them IBM-compatible PCs even though we got out of the business 10 years ago. But the architecture and the machines that we created then spawned an entire industry of machines and then clearly gave way to Microsoft and Intel becoming very dominant in the PC industry. We did a similar thing in ’91. We created a wholly owned subsidiary called ISSC. ISSC was the seed for IBM Global Services. It grew to a $50+ billion business over the course of a decade.
We recognize that there are times in any large corporation’s history that a new idea needs to be set aside in order for it to grow at the speed it needs to grow at. We felt this was one of those moments. We decided to set it up as a separate group. Group structure in IBM is a [specific] designation. IBM Global Services is a group. There are only a half dozen of them in the company. It is a separate P&L, separate structure, it’s autonomous.
We moved 150 researchers, bringing over a whole next wave of products. That [represents] the single largest movement of IBM Research personnel in IBM’s history. We brought the market validation team, the team that had been working the technology forward, obviously, and we also brought over some related big data technology from our software business to form into the new group.
We brought over consultants from our consulting business, delivery people from our cloud business so that we had the ability to create strategies, design, build and run Watson-based systems for clients without having to stitch together pieces of different organizations to make that happen -- a very clear line of sight on what we had to do. That’s why we created it as a separate group. That’s worked out pretty well. We’ve been able to move very quickly.
CIO: A lot of what we read about Watson involves big engagements with big brand-name customers, some really amazing things that you’re doing. I think that leads to a couple questions from customers. Is Watson only for the big guys and does every project involve this deep consulting and engagement with IBM?
Rhodin: No. One of the things that we realized when we set up the commercialization project is that there were going to be three parallel tracks to market. There was what you just described, which I would call big transformative solutions, industries that are going to go through pretty robust transformations because of these information-based platforms.
We also recognized there was going to be a class of very repeatable applications that would evolve, much like the ones that we launched a couple of weeks ago at the IBM conference, Insight, like IBM Watson for Wealth Management, even our Chef Watson application. We’re targeting widespread adoption, not requiring heavy consulting engagements up front, more off-the-shelf product sales.
The third level is that we recognize that at our core we’re an enterprise B2B company. But cognitive computing is going to be a great technology for consumers and we needed a way to reach them. The way to do that is to open it up as an ecosystem platform that allows startups and entrepreneurs to build applications and businesses on top of the platform. Those applications and businesses can target other businesses, or target consumers free, and we set up the system in a very contemporary way, which is that you can use the platform for free. If you build a business on it we get a cut of the revenue.
CIO: I want to drill into that. Since you mentioned it, is that both as a development sandbox but also a Platform-as-a-Service (PaaS) where people can buy and run these applications?
Rhodin: Exactly. The idea is you start in the sandbox, you promote it to production and we run the production system for you. The work we do for our revenue share is that we’re actually providing the technology and running the system for the clients you’re selling to.
CIO: Is there a model like that in IBM today?
Rhodin: Our Bluemix platform is that model. Amazon Web Services is another PaaS model. Azure is another PaaS model. There are a number of those PaaS models. Watson is only available right now on the Bluemix platform.
CIO: Do you ultimately see it going on to other platforms?
Rhodin: I’m busy enough right now that I’m not worrying about it.
CIO: I want to go back to the big transformative ones. Talk about that brainstorming process. How do the ideas take shape?
Rhodin: One of the things we discovered over the course of the last year is that this isn’t about industries as much as it is professions. So we started to think about what professions exist where the amount of information being produced is overwhelming the ability of the professional to consume it? So that’s target No. 1.
The other thing that seems to be common about the really big, transformative solutions are professions that are taught through the time-old tradition of apprenticeship, the passing down of best practices from one generation to the next. That applies to Watson because it’s a voracious reader. It can just keep reading and connecting the dots across what it reads. Then it learns as it goes through a training cycle with experts. Those experts are actually teaching it best practices. Those best practices become how it learns and becomes expert in a particular domain and that’s when you build the app.
What’s interesting about this new model, this knowledge-based system model, is that most of the work goes into the creation of the knowledge base, not the application. In the prior domain it was the other way around. That takes people a while to get their heads around. You’re going to spend six months working on your information and you’re going to spend two weeks writing your app. It’s really that dramatic.
The reason it takes a while to figure out the information is that curating the information is often difficult. You have to curate information from many, many sources, synthesize it down to something you can create a concrete article out of. Watson has to go through that same process. We have a process and a product called Watson Curator that crawls and searches all of your enterprise data and helps you build curated datasets for ingestion into systems like Watson.
The best practice piece is done the same way our children learn. If our children read a book, the way that we know they understood the book is we ask them questions. So after Watson reads a lot of books, we ask it questions and we see how good it is at getting the answers.
If it’s not good, then we try to figure out why is there conflicting information in the underlying knowledge base. Usually it has something to do with the way we’ve constructed the knowledge base underneath. But you have to be careful not to train the system to answer the questions you know because then you’re just writing a rules engine.
You always have a blind set of questions that no one has seen and those run against the knowledge base and they let you know how accurate you are against the blind set. Then you know whether the knowledge base is trained correctly. So you go through these training cycles to build up your knowledge base. Once the knowledge base is built, the APIs are clear, it’s very easy to build the applications at that point.