Wikibon trip report from IBM’s Watson developer conference: Keep it simple, geniuses


IBM Corp. demonstrated at its first Watson Developer Conference last week that it has successfully hidden enough of the “rocket science” in machine learning to excite a crowd of 1,300 mainstream developers.

That’s right: I said developers, not data scientists. IBM is making a big play to extend the commercial reach of its artificial intelligence-driven computing platform, and indications are that it’s starting to work. A few things we heard at the event in San Francisco:

  • IBM Chief Executive Ginni Rometty (above) announced that its cognitive computing services already reach 200 million people worldwide and would reach 1 billion by the end of 2017.
  • Healthcare and consumers are proving especially ripe targets for Watson.
  • IBM will continue to engage data scientists, but it indicated that increasingly it will turn its attention to reaching developers.
  • It aims to reach those developers by making Watson’s machine learning tools and services as easy-to-use and cost-effective as renting cloud virtual machines.

A third computing transition: cognitive

IBM speaks of a third transition in computing, which can be called the “cognitive transition.” What’s driving this transition is the explosion of digitally accessible information. There’s just too much information to understand for humans to explicitly write the correct, “best” instructions to guide both computers and — more broadly — systems (which include people).

Instead, computers have to learn from all this data and augment humans in making sense out of it. Rometty came back to the theme of augmenting people, not replacing them, several times. IBM believes just about every decision humans make will be assisted by machine learning within five years. And Big Blue wants to lead that process.

How does it plan to lead? IBM’s core differentiation is to take the voodoo out of machine learning. Using a research and development budget that’s among the largest in the industry, IBM has a goal to hand ever more powerful Watson application programming interfaces over to developers and spawn an explosion in applications that’s equal to the explosion in digital data.

Two product lines that target different customers constitute its offerings. First are APIs that mainstream developers without a data science background can use. Second are the semi-custom industry solutions that IBM’s professional services organization has the skills to package for delivery to end customers.

The developer APIs were the focus of this conference, and they were on rich display in a number of impressive showcases. In one such showcase, IBM highlighted a solution that can diagnose melanoma remotely.

Using a smartphone with a special snap-on lens, an untrained individual can take a picture of a lesion on their skin and ship the image to Watson for analysis. Watson does deep analysis in the cloud based on a corpus of images tens of thousands of times greater than any single doctor could study. Watson augments the remote doctor’s ability to diagnose the image with two factors. It returns a confidence level whether the lesion is melanoma as well as which attributes of the image are relevant to its diagnosis. It can also show images of similar lesions in its database.

It’s the Model, Not the Data

The machine learning solutions IBM builds for customers are based on live models with data feedback loops that make the models smarter over time. That’s identical to Wikibon’s notion of the “data feedback loop” capability in our digital business platform research.

IBM sees this process — ingestion, model, refinement — as the true value-add within machine learning. Wikibon agrees. No digital business should have the narrow goal of generating just more data. Rather, IBM’s belief, with which Wikibon concurs, is to generate more models that can complement — and in some cases, supplement — human decision making.

But IBM is going to be very careful with how it leverages customer data. Rometty emphasized that IBM won’t leverage the live models from its customers in order to improve its solutions over time. On this point she drew a clear distinction between IBM and competitors such as Google Inc. (We would add Oracle Corp. to Google’s camp; Oracle’s Datalogix subsidiary, among other Oracle properties, collects large volumes of data on consumers and businesses and openly resells that data). Rometty insisted IBM will not follow that path.

Deep learning today is competitive with humans in recognizing objects in images. Extracting structure and meaning from text just as effectively is within reach. Graphics processing unit-accelerated servers, open source frameworks such as Google’s TensorFlow, Python and its libraries, graph databases such as SNAP and an API to Microsoft’s Bing that can crawl the web are the new building blocks.

Business-to-business marketing firm Demandbase Inc. estimates it would have cost 100 times more to build its solution 10 years ago versus today thanks to these building blocks. The use cases for machine learning, AI and cognitive are extensive, but adoption won’t accelerate if the technology can be understood only by a few priestesses.

Watson is cool technology, no question. But it will enter the pantheon of great technology only if IBM can, in fact, make it both highly functional and as easy-to-use as Amazon Web Services made the cloud. Based on what we saw at the Watson Developer Conference, IBM is off to a fast start.

George Gilbert is big data and analytics analyst for Wikibon.