On Thursday, IBM announced that it is adding Google’s TensorFlow deep learning framework
to its PowerAI toolkit, a collection of software resources for building artificial intelligence applications on the POWER architecture. I did not previously know that PowerAI was a thing, and I don’t sense too many people are building new applications using POWER8 processors (much less on IBM servers). And even still, I believe PowerAI seems much more likely to drive meaningful AI revenues at IBM than does the overhyped Watson platform.
Don’t get me wrong: Watson was a great ambassador for AI well before deep learning came along, but it’s the “well before” that’s the problem. A lot of work went into tuning the original Jeopardy!-playing system to do its thing, and IBM’s Watson lineup of services is still fairly focused on analyzing text, speech and the like. IBM has no doubt re-architected Watson to tackle more general-purpose tasks, but it doesn’t strike me a general-purpose solution. And the buzz around around things like sentiment analysis must be winding down.
Plus, today, there is plenty of competition for API services around text, speech and computer vision, with the most obvious examples coming from mega cloud providers such as Amazon Web Services and Google. Not only are those the places where many companies want to deploy new workloads, but Amazon and Google also own the consumer platforms (Alexa and Home) with which many developers will want to integrate their own AI services. Amazon, Google and even Microsoft also just seem more like experts in that field than does IBM, given that their voice assistants and NLP algorithms are already inside our homes and pockets.
However, where those providers don’t want to go is inside of large enterprises to help them build out AI systems that will be useful for the specific needs and specific data of each company. That’s harder, more resource-intensive from a vendor perspective, but it’s also a prime opportunity if you can figure out the right business model. Even in a world of open source software (and AI is not immune to that
) large companies will still pay for software licenses and enterprise editions if there’s value to be had.
“Everything in the enterprise space is ROI driven. They don’t know that the newest deep learning paper just came out from Google. They’re not going to clone some random GitHub repository and try it out, and just try to put it in production. They don’t do that. They want to understand ROI. They work a job, they have a goal, and they have a budget. They need to figure out what to do with that budget as it relates to their job at their company.”
Forget about speech and text and all of that. If you can go into an enterprise and use deep learning to squeeze a few extra dollars out of their data, there’s probably a business to be had. Skymind is trying it, H20
(which on Wednesday announced a new deep learning product
for enterprises) is trying it. IBM should try it, too.