Google’s CMMA keynote was an ML sales pitch – but a good one

google ML as a service
Would you buy an ML service from this man?: Richard Harshman, managing director of Asia-Pacific at Google Cloud

Google Cloud’s CommunicAsia keynote was a blatant ad for its ML-as-a-service to enterprises. But as sales pitches go, at least it was a pretty good one.

It seems to be a tradition at CommunicAsia (now part of ConnecTechAsia) that the designated Major Digital Company keynote be delivered in the form of a blatant self-promotional sales pitch. And Google Cloud did not disappoint.

Richard Harshman, managing director of Asia-Pacific at Google Cloud, spent his opening keynote at the ConnecTechAsia plenary conference advertising the company’s machine learning tech and its strategy to offer ML-as-a-service to enterprises. But as sales pitches go, it was an instructive one.

Unsurprisingly, Google is already integrating ML into practically every product it has – from obvious things like search and YouTube to relatively banal services like Gmail, which makes use of ML for features like “smart reply”.

But Google also sees an opportunity to bring ML to the masses because – not to put too fine a point on it – ML is hard. Harshman explained that companies need three things to use ML successfully: a model, massive amounts of computing power (either in the cloud or on-premises) and large amounts of quality data.

The third one is the most important – partly for obvious reasons (ML won’t work if it has no data to learn from), but mainly because data analysis is the easiest part to get wrong.

“We work with many organizations who are trying to get a handle on their data strategy, and they may be incorrectly capturing the data, or the data lives across different systems and silos, especially if you’re in a large organization,” he said. “So it’s very hard to be able to get insights out of data, and then create a machine learning model.”

Meanwhile, he added, CEOs are under constant pressure to address constantly shifting consumer habits, the speed of technological change and a distinct lack of people with the right skillsets to drive the necessary changes the company needs to undergo. “So you’re going to need to be able to bring a new skillsets, partner with new companies, and explore new initiatives – that’s a lot for an individual company to be able to take on.”

As it happens, Google has developed pre trained APIs designed to help enterprises get started on ML by inserting them into their workflows. “You can start using them today – go to our website, and you can start training your models immediately.”

Then there’s Cloud AutoML, a tool that automatically create a machine learning model from whatever data you feed into it in minutes – as opposed to going through all the steps of building your own model, which could take weeks if you’re not an ML expert.

And that’s really what Google is addressing here, Harshman says  – the fact that most people aren’t experts, to include Google’s own developer community.

“Of the [21 million] developers that we currently work with, only 1 million of them are data scientists, and only a few thousand of them globally have the right training and deep learning models,” he said. “So innovating on the tech is only one piece of it. Getting it out there and [making it] usable is quite another story. And at Google, we want to change that.”

Google isn’t the only one. Microsoft has already been offering ML-as-a-service to enterprises for some time, while Amazon launched its own ML service late last year. IBM and HPE are also players, and there are a host of start-ups competing for a piece of that pie.

As for the size of that pie, it’s early days – both for MLaaS and ML in general – but Orbis Research expects the global MLaaS market to grow from $932 million last year to $8.3 billion by 2023 (that’s a CAGR of roughly 43%). Stratistics MRC is forecasting similar growth – from $679.3 million in 2016 to reach $7.6 billion by 2023 (a CAGR of 41.2%.).

Be the first to comment

What do you think?

This site uses Akismet to reduce spam. Learn how your comment data is processed.