IBM's Steve Mills Goes Deep on the Cloud, Watson, POWER8
So the more ways you can get input into the system the faster you can build up domain knowledge. And teaching is one of the bigger challenges we have here. How fast can you teach it? How fast can it get to the point where then it becomes self-learning in the sense that it becomes self-correcting and self-reinforcing. Because it learns entity, attribute and relationship. And within entity, attribute and relationship you teach it the truth. When it has enough truths then it can deduce the next set of truths out of any new piece of information that’s presented. It uses classic AI constructs. And the problem with a lot of AI technologies over the years have been how hard it is to train the environment and then can it be multi-domain. You end up with fixed ontologies versus variable ontologies. We have a lot of things going on with ontologies and finding ontologies, surfacing ontologies independent of being taught the ontology. That’s a tough problem. We’re also doing more language work. We’re doing English pretty well, but we have to do all the languages. And language is really hard. Is SyNAPSE part of this? The SyNAPSE capability is really about scale. It’s a mimicking of the number of connections in the human brain. If you can create enough connections that has an indication of your ability to feed patterns and connect those patterns up. It’s a scale and performance-enabler in effect. So it’s more of a mimicking of a human brain.Yes. The good news in that space is that the technologies are such that within a reasonable physical size you can pack an enormous number of connections in. IBM lately has been making a lot more emphasis on design. How big of a push is that now within IBM? It’s huge. We’ve always had a user-centered design initiative in IBM, so it’s not new. What we’re trying to do is invigorate the design community in IBM so that for a lot of the work we’re doing we want to start with the user experience and work our way back. We have a lot of products where that’s not the approach. If I’m trying to extend the scalability of DB2, I don’t need to worry nearly as much about the design. I mean there’s a DBA issue and there are some user issues, but it’s not profound. But when you think about analytics tools, which have to do with line-of-business people, visualization, how they want to see it, does the insight that you want them to derive jump out at them or is it somehow masked by the complexity of your UI? So that design orientation is critical for lots and lots of products. And as we try to extend into more line-of-business people, people who have less and less traditional IT experience, they’re less tolerant of UIs and obscure workflows that make sense to somebody with a degree in computer science but makes no sense to them. Because they’re not thinking like computer scientists. So we’re putting all the development teams through the knothole of forcing design up front, design verification. You want to get the design right before you code, as opposed to you code and then think about how a customer is going to use this thing. Can I graft a UI on it that makes sense? It’s bringing that part of our development community up to a level above where they’ve been. Because oftentimes when pressure built up in a development project their voice was not getting heard because the team became totally fixated on getting their development function requirements out the door rather than getting the design right. So we think of it as a big deal. And we’ve been doing it for a long time. We’ve had all kinds of user-centered initiatives at IBM for many years, but we felt it was time to bring it up to a much higher level of focus. That’s what Phil Gilbert is doing. It’s more than just being an evangelist or enthusiast for this, it’s also knowing what the team on the other side thinks. Phil knows how the development teams think. He knows that his young design people need to figure out how to connect to developers so developers know how to change their orientation towards design first rather than function and scale first. It’s a design-led model and it’s cultural.
And that’s based on a chip?