Why Google eyes a slice of the machine learning pie

By
Thursday, March 29, 2018

Ubiquitous in many internet services, such as search and email, Google wants to have a bigger slice also of the artificial intelligence cake, leveraging its services and platforms to conquer not only regular users but also enterprise customers.

At the end of February, the company announced it was making its AI and machine learning courses known as Learn with Google AI, which has been taken by more than 18,000 Google engineers, available online and free to the general public in general.

The company also started opening some of its APIs and has programs in place for enterprises to send data engineering for immersions at Google's headquarters in Mountain View. In Latin America, Google's first machine learning customers are mainly in Brazil and Mexico.

BNamericas spoke to Rafa Muñoz, who leads the professional services team (PSO) for Google Cloud in Latin America, about the company's machine learning strategy.

BNamericas: What is Google's approach to leveraging machine learning in Latin America? 

Muñoz: There's a lot of interest in the industry around machine learning and artificial intelligence in general on how to improve businesses productiveness and the lives of users and end-customers.

But the reality is that we are at a very early stage across the board in the industry. There are multiple solutions, for sure, that have a certain degree of complexity and maturity. However, the processes of letting machine learn require not solely massive amounts of information but also a high amount of complexity.

Having launched TensorFlow with other open source technologies enabling companies to start using machine learning and build their modules, we realized that companies' abilities to handle and processing data were not enough.

So in parallel with the TensorFlow and the set of APIs we launched free training modules where they can leverage on Google data services and Google information at scale to augment their applications and to bring machine learning to their applications.

So, for example, we have many customers that connect the audio of their call centers and they use our APIs. They process that audio and generate impact from that, get sentiment, analysis on top of that to have a better understanding of customer satisfaction.

What we released very recently is AutoML, to allow users to create custom modules on the email. It is an intelligent email categorization.

We created it as a way for users to start creating their modules with a just a few datapoints and then allow our machine learning modules to work on top of that. 

What I mentioned around audio processing or image processing are just a few of many potential cases that you can have with machine learning.

Our strategy as a whole is to enable the research community with tools like TensorFlow and APIs and the AutoML generating custom machine learning modules for end-users.

BNamericas: All of this leveraging on Google services and solutions?

Muñoz: Our vision is that it opens up the possibilities for people to start getting familiarized and understand machine learning at a fundamental level without having to learn code or having a massive amount of data in the right format.

BNamericas: Which customers are we talking about, who is Google currently selling these solutions to?

Muñoz: Some of our enterprise customers are leveraging principally on the APIs training modules. We are working on Google Cloud a lot with retail and also with financial services in general. There are also some initial talks with the public sector.

I don't think I can give any local names in particular, but what I can tell you though is that on the top of the options that we're bringing to enable machine learning we have many customers training their data engineering staff to enable their applications. We also have this service what we call the Advanced Solutions Lab, in which a group of five data scientists from enterprise customers go for a one-month training in Mountain View with our engineers in machine learning.

We have customers already in Brazil and in Latin America in general engaged in that.

BNamericas: Where exactly in Latin America aside from Brazil?

Muñoz: Mainly from Brazil and a couple of others from Mexico.

BNamericas: So is the company's strategy addressing first the major markets to then tackle other economies in the region?

Muñoz: Correct. We have coverage in the major markets in Latin America and these options are present to all of them. The first groups that have attendance trainings are from Brazil and some from Mexico.

BNamericas: How does Google look to position itself in this corporate artificial intelligence market considering that brands like IBM, through the Watson platform, are already starting to become dominant there?

Muñoz: The key differentiator in our opinion is that we don't have to bring big infrastructure to deliver these kinds of solutions and even if we consider solutions like the TensorFlow we have cloud email and other things in place for companies to leverage on.

Machine learning requires a massive amount of processing and of storage and we have that. At the end of the day, to teach machines and for machines to really learn they need a massive amount of examples in order to be precise, they need high processing power and storage.

If you have to run this on a traditional IP environment on premise, you're going to have to deal with a very complex infrastructure. But companies can use our managed services. Instead of waiting for several days to train a module using AutoML and Google you take just a few hours.

BNamericas: Does the company also rely on its LatAm's datacenters, in Chile and Brazil, for machine learning?

Muñoz: Yes, we have our GCP cloud region available in São Paulo, for example, but in any case it's possible to move datasets to another region if there's a need, to move managed services across the GCPs with a high degree of performance.


About Rafa Muñoz

Rafa Muñoz joined Google in 2011 as a deployment engineer and led the technical side of strategic projects for the company. Prior to Google, he worked for more than seven years in different GE and GE Capital units in the Americas. Muñoz holds a bachelor's degree in systems engineering from ITESM in Mexico.