For any cloud to be taken seriously, it has to meet an ever rising bar of features. Machine learning seems to be on that list, as all the major cloud providers now feature it.
But how they go about doing it is another story. Aside from the "curated API vs. open-ended algorithm marketplace" models, there are the "everything and then some vs. just enough" variants. Here's how the four big cloud providers -- IBM, Microsoft, Google, and Amazon -- stack up next to each other in machine learning.
IBM: Turning the ship with, Watson at the helm
When IBM first announced it would turn its Watson AI system into a consumable service, the questions piled up. What would it look like? How would it be consumed? But most important, how much support would it lend IBM's effort to reinvent itself as a cloud giant?
Two years and change later, IBM has rolled out an array of machine learning-powered services on its Bluemix PaaS: weather prediction, for instance, or systems for analyzing language, image recognition, language translation, sentiment and tone analysis, and so on.
Of all the companies offering machine intelligence of some kind in the cloud, IBM's roster has the most ambition behind it. More significant, IBM has also been working to complement the pie-in-the-sky tools with more down-to-earth ones, mainly involving analytics and reporting. (Growing Watson also seems to be the motive force behind many of IBM's strategic acquisitions across multiple fields: weather, health care, and so on.)
The question isn't even whether Watson services will find practical uptake -- in time, they're likely to do so -- it's whether they'll do so at a scope and at a pace IBM is comfortable betting the farm on. Most of these offerings need time to find use cases, whereas IBM's well-established Spark services (or its new AWS Lambda-like service, OpenWhisk) are easier to monetize immediately.
Some organizations have started to put Watson's analytics services to work in creative ways. But while IBM has said it believes it can grow all this into a a $10-billion-a-year business, the uptake for Watson so far hasn't been nearly enough to offset IBM's slumping revenues.
Microsoft: Built for you and by you
Where IBM has Watson, Microsoft offers Project Oxford, a set of curated high-level APIs to cover machine vision, speech recognition, and language analysis. The list of APIs isn't as broad as Watson's (and let's face it, How-old.net wasn't quite as fun as Watson playing "Jeopardy"), but Microsoft's intentions are much the same: a proprietary set of curated APIs that leverage machine learning.
Azure Machine Learning Studio is possibly the more important half of Microsoft's machine learning ambitions. There, people can bring their own data, train machine-learning models on it, then reshare the resulting model as an API via a REST interface. IBM has something similar in its Predictive Analytics service on Bluemix, but Microsoft's Studio has been around for longer and has a more general-purpose feel to it.
Both IBM and Microsoft are attempting to create two different flavors of machine learning services. One's been created behind closed doors, as it were, with a curated data set and tuned behaviors (the Watson APIs, Project Oxford). The other is a platform upon which new kinds of machine learning services can be built, shared, and even monetized (Azure Machine Learning Studio, Predictive Analytics).
But the biggest difference between Microsoft and IBM isn't in the services, but the motivations. Microsoft's attempts at future-proofing itself by pivoting to the cloud have been aided by its other successful business sectors -- gaming, for instance -- so it hasn't felt existential pressure of the same degree that IBM has. But that doesn't mean Microsoft can't sense which way things must go.
Amazon and Google, the minimalists
If Google and especially Amazon have any one guiding tenet to their cloud approaches, it's "less is more." Maybe better to say "just enough is more," which includes the way both companies offer cloud-based machine learning services.
In Google's case, Google Cloud Platform currently offers only two services akin to the others profiled here: Google Translate (an API supporting Google's existing machine translation engine), and Google Prediction API. The former is a proprietary API maintained exclusively by Google. The latter, despite the unassuming name, is a broadly inclusive service that allows users to upload data and train models in the manner of of Microsoft Azure Machine Learning Studio. (Data can be derived from Google services like Google BigQuery.)
Amazon Machine Learning is similar to Google Prediction API in that models can be trained against data and used to make predictions. It's a deliberately simplified service, either for the sake of appealing to developers who only want to solve a specific, narrow problem or because Amazon wanted to test the market waters first.
In both Amazon and Google's cases, their targets are developers both with narrowly defined needs and with data already on those clouds -- the "just enough" model. IBM and Microsoft are aiming for far broader territory, and while IBM strives to have the most to offer, it also has the most to lose.