IBM's Steve Mills Goes Deep on the Cloud, Watson, POWER8

By Darryl K. Taft  |  Posted 2014-05-21 Print this article Print
IBM logo

I think this is potentially very profound as far as building momentum. And we’re way out in front of anybody else in terms of having a real integrated environment, a full system with all of the capability that one would want to be able to do something of value. So now we need to get more people participating. If IBM becomes the gate for participation we can only scale so much on our own. So we need more people to participate.

What is your reaction to the opinions that came out about Watson not faring so well because it has not made any money for IBM yet?

We’re generating revenue now. So people have been paying us money; they began paying us money last year. We’ve been collecting money. In the beginning you get revenue but the revenue doesn’t cover all your expenses. Then the revenue builds over time and it finally crests over and you get to profit. There’s a pattern for any software business out there. Look at all the startups. How long does it take them to reach profitability? Often there’s a plowback phenomenon. Assuming your investors are willing to let you do it, you keep plowing money back into the business because you believe you can make it even bigger. And from a venture perspective there’s a back end exit strategy, which is either IPO or M&A.

So there’s really nothing different about what we’re doing and what a venture capitalist would do. Except we’re our own venture capitalist, we see a big opportunity and we’re going to keep investing for a period of time out in front of the revenues. But we also have a pretty aggressive pathway to generate more and more revenue.

We’re not going to break Watson out from a financial dynamics perspective, though. It sits inside of the larger business analytics initiative that we have. And that’s the only way we’re going to talk about it financially.

What’s in your workshop? What’s in your research channel that’s close to coming through?

There’s a lot of Watson-related stuff in research. We’re focused on trying to understand human beings – attitude, sentiment, preference. With so many of these Watson projects having to do with businesses wanting to better serve their customers, we have to help them to know their customer. And there is technology that all of us that write anything can use. You get access to someone’s writing and you can begin to understand aspects of their personality and preference.

Consumer-facing companies want to know what are you tweeting about, what’s showing up on Facebook, are there blogs that you’re interacting with, etc. Some things are easy to understand and some are harder to understand. We have technology that allows us to provide a deeper level of understanding. And what they want to do is create preference for consumers. They’ll say, 'you seem to have a preference in this direction, what offers can I give you? Can I get greater loyalty out of you because I understand what you seem to care about by creating a preference profile?'

Also you have the whole five senses thing and all of these senses can be converted to numbers. You can put value on them. For instance, there are only a set number of flavors and everything else is combinatorial. And if you set the value, you can create menus. That’s what this Chef Watson thing does. As we create recipes we’re creating numerical value around taste. You can create numerical value around smell. You can do numerical value around audio or video, so we need Watson to see.

One of the challenges in teaching the systems is it’s a fairly limited input model. Whereas you take a child, a child has five senses so they're learning lots of things around lots of things through observation and interaction. And they figure out the association fairly quickly. They learn what wood is and everything that’s wood, they know that it’s wood. However, a computer doesn’t know anything, it just has statistics. So you have to teach it a whole lot about wood before it finally gets to the point where it always gets wood correct. It’s an asymptotic set of conditions where you have so much information on a topic that you're almost always right and you're never wrong. It’s not the way the brain works, but that’s the way the computer works. It becomes 99.9999 probability of correctness.


Submit a Comment

Loading Comments...
Manage your Newsletters: Login   Register My Newsletters

Rocket Fuel