Is Google Cloud Machine Learning enterprise-ready?

Google's announcement of refreshes to its cloud machine learning service provides a good midcourse indicator of just how ready Google Cloud services are for the enterprise, not to mention, machine learning in general.
Written by Tony Baer (dbInsight), Contributor

From the get-go, the core value proposition of the Google Cloud Platform (GCP) has been granting enterprises access to the same infrastructure and advanced software that Google uses to run its own business.

Paging back in history, that also was the original value prop of Amazon Web Services. Since Amazon runs a globally distributed transaction business, why not let enterprise clients come along for the ride? A decade in, we now know how well that worked out.

Google's secret sauce has been the high automation of its compute infrastructure and the artificial intelligence that's been baked into its operations. As consumers, we see the fruits of that as we consume traffic data on Google Maps, get stories generated from sets of photos taken on our Android phones, and, of course, take advantage of Google search. As IT, we debate the merits of Google's NoOps vision for lights-out management of IT.

Given Google's heritage as a mass consumer service, it's not surprising that the bulk of the deliverables for which Google has become known have been more consumer- than enterprise-focused.

And given the next-generation NoOps focus of Google's IT infrastructure, it also shouldn't be surprising that the bulk of Google Compute Cloud's early enterprise references have largely been moonshot, rather than keep-the-lights-on projects. For instance, Coca Cola looked to GCP because it had the infrastructure that could scale for a global marketing campaign for consumers to get its photos stitched into a Happiness Flag presented at the 2014 World Cup. Not exactly the stuff of keeping your back office SAP system running.

And if you look at the machine learning (ML) APIs that it's released, the mix still very much reflects Google's consumer heritage, such as vision that is geared toward identifying friends within objects, their sentiments or expressions, or landmarks like the Eiffel Tower.

With this wave of releases, the pure consumer business focus is beginning to change -- emphasis on the beginning. Google realizes ML as a business won't grow with a "build it and they will come" attitude. So now it is creating an actual products group for ML, getting its toes wet in professional services to help its enterprise clients figure out how to use ML and start adopting more of a product mindset.

The core offering, Google Cloud Machine Learning, has just entered general release this week after a beta period of mostly bug fixes. Google is entering a market where other cloud providers, like Amazon and Microsoft (plus niche providers like Databricks), already offer ML services -- and IBM just announced one.

Befitting the early stage of the ML market, the cloud services from each of the players are hardly comparable. For instance, while TensorFlow, the heart of Google Cloud Machine Learning, targets deep learning (which simulates human thought and decision-making), Amazon's ML service tackles less ambitious tasks such as classification and regression.

Beyond Google Cloud Machine Learning, aimed at data scientists, the brunt of Google's ML cloud services for now target developers, with the Cloud translation, speech, vision, and natural language processing APIs.

An early result is the Jobs API that was announced as beta with this release. Not surprisingly, it grew from internal need for Google to improve recruiting with a smarter, data-ware way of matching jobs and prospects. It's the type of vertical application that Google -- and others -- will need more of to get ML out to a more mainstream enterprise market. Ultimately, demand for more vertical ML apps will be limited only by the imagination.

While university programs are turning out more data engineers and data scientists, there will likely never be enough of them to go around to make them as commonplace as SQL developers.

Other examples include a new premium edition of the Translate API that is geared toward the types of high volume jobs, such as translating thousands of emails or documents, which will be more useful for enterprises that need to adopt this technology at scale. And for the vision API, there are new business-focused capabilities that can detect entities such as corporate logos.

As a consumer, ML is practically all around us, as you experience when Amazon or eBay shows you next-best offers or when Netflix makes program recommendations. Google's challenges in cracking the nut to sell ML and AI to the enterprise reveal the challenges that are faced when starting with practically a blank slate.

Editorial standards