Google LLC remains a powerhouse in artificial intelligence, which is a fundamental frontier in cloud, edge and enterprise computing.
Hyperscaling is a principle that drives everything Google does in the AI market. A year ago on theCUBE at Big Data SV, a Google executive discussed AI best practices that the company executes comprehensively within its approach to market, with scalability central to them all:
- Establish an agile business culture that puts AI-first approaches at the heart of digital transformation.
- Assemble a global AI team with the skills and vision to drive the necessary innovations.
- Deploy a high-performance AI hardware infrastructure all the way to the edge.
- Provide developers with the tools to accelerate and automate building of innovative AI applications.
- Ensure that there is high-quality curated big data for building and training all AI apps.
Google rigorously applies these hyperscaling principles to accentuate its AI industry leadership in the following areas:
- AI developer community engagement: Google is the principal developer of the TensorFlow framework, which is the most widely adopted toolkit for deep learning worldwide. The company open-sourced TensorFlow a few years ago and continues to make deep investments in the framework.
- AI R&D: Google continues to explore AI frontiers from research conducted in its own labsand by Alphabet Inc. company DeepMind Technologies Ltd.
- AI in the cloud: For enterprise customers, Google Cloud Platform is the heart of the vendor’s AI and data management go-to-market strategy, though AI technologies are also central to its mobile and Web product units, such as Android.
As befits a company whose very business model relies on heavy doses of AI, Google this past year created a new business unit dedicated to integrating AI across its product portfolio. Google Cloud Platform is the centerpiece of that strategic push, with specific emphasis on the following emerging requirements for AI:
- Automating customers’ AI DevOps pipeline to the max: Google Cloud is rapidly enhancing its cloud’s support for automation of data ingestion, preparation, modeling, hyperparameter optimization, training and other highly technical data science pipeline workloads in its public cloud. Google’s Cloud AutoML service, still in beta, is a technical pacesetter in this arena. Furthermore, Google has been the primary developer behind Kubeflow, an industry open-source project that’s creating a framework-agnostic pipeline for putting AI microservices into production across a multiframework, multicloud, cloud-native ecosystem.
- Accelerating time to value on customer AI application development initiatives: Google Cloud has steadily expanded its one-stop shop portfolio of developer-friendly tooling, prepackaged solutions and pretrained models for simplifying access to AI capabilities in its public cloud. In this regard, the company now provides such offerings as Cloud Tensor Processing Units, Machine Learning Engine, Job Discovery, Dialogflow, Natural Language, Text-to-Speech, Translation API and Video Intelligence. It has also integrated AI into its tools for mobile and Web application development as an enabler for low-code programming. And it has rolled out the first instances in a growing portfolio of AI-enabled prepackaged cloud services for specific business applications, such as recommendation engines, contact center and talent management.
- Optimizing the end-to-end customer data platform with and for AI: Google Cloud continues to build more AI-driven IT optimization, cybersecurity and compliance features into its service portfolio, including storage and big data analytics offerings such as BigTable, Cloud Storage, Datastore, Persistent Disk, MemoryStore, BigQuery, Dataflow, Dataproc, Composer, Datalab, Dataprep, Pub/Sub and Data Studio. The company is evolving these cloud solutions to handle the most demanding AI workloads, while partnering with companies such as NetApp to provide best-of-breed cloud file storage for AI applications. For example, in the past year, it released Google Cloud Spanner, which is a horizontally scalable relational database service suited for processing both online transaction processing workloads and resource-intensive AI workloads. Spanner has been engineered to meet Google’s operational needs for global scalability and efficiency. But it’s not well-suited to all enterprise workloads, having been engineered more for strong global consistency than availability. It has functional limitations in online analytical processing, resource access controls, migration tooling, backup integration, view functionality and data modeling.
When we consider the interviews that took place on theCUBE at Next ’18, we can find clues to how Google plans to incorporate AI into its cloud data hyperscaling priorities in coming years:
- AI in cloud data warehouse hyperscaling: Sudhir Hasbe, director of product management, Google Cloud: “When we were talking to our customers, one of the biggest challenges they were facing with machine learning in general [was that] every time you want to do machine learning, you have to take data from your core data warehouse, like in BigQuery you have petabytes of scaled data sets, terabytes of data sets. Now, if you want to do machine learning on any portion of it you take it out of BigQuery, move it into some machine learning engine, then you realize, ‘Oh, I missed some of the data that I needed.’ I go back then again take the data, move it and you have to go back and forth too much time…. Eighty percent of the time the data scientists say they’re spending on the moving of data, wrangling data and all of that…. Why do people have to move data to the machine learning engines? Why can’t I take the machine learning capability, move it inside where the data is, so bring the machine learning closer to data rather than data closer to machine learning. So, that’s what BigQuery ML is, it’s an ability to run regression-like models inside the data warehouse itself in BigQuery so that you can do that.”
- AI in cloud cybersecurity hyperscaling: Suzanne Frey, director of security, trust and compliance and privacy at Google Cloud: “We talk about spam and phishing protection and things like that and we get billions of signals every single day about malicious information or malware, ransomware, those sorts of things. So we have a very low-latency view into what’s happening at the next minute around the world in that respect. And that gives us a competitive edge in terms of really thinking about what’s the next thing that’s going to happen. We certainly know that machine learning, whether it’s Smart Compose and Smart Reply, or it’s actually based in security, an anomaly detection. What’s an anomaly to one company is not necessarily an anomaly to another, depending on what business you’re in and the like. So we’re investing in machine learning and understanding how to be that security guardian for our customers in an automated fashion, so the people don’t have to worry about security. We’ve taken care of it for them. That’s the holy grail and that’s what we’re investing in right now.”
As Wikibon looks ahead to Next ’19, Google has its work cut out for it on the cloud AI front. The past year has seen significant advances by its chief public cloud rivals to ramp up their tooling, services and apps for driving deep learning, machine learning and other AI capabilities. Just as important, Amazon Web Services Inc., Microsoft Corp. IBM Corp. and other cloud players have brought innovative AI into their multicloud and hybrid cloud offerings in innovative directions that challenge Google to respond if it hopes to remain relevant. At the very least, Wikibon recommends that Google pursue the following cloud AI priorities going forward:
- Grow its range of out-of-the-box, line-of-business and industry-specific services that incorporate AI to speed customer time to value and position Google Cloud as a more credible SaaS player.
- Ramp up the productization of its AI pipeline automation tooling, based on commercialization of the rich TensorFlow stack and architecting it as a team workbench on Kubeflow for cloud-native DevOps, thereby fielding a powerful alternative to AWS SageMaker, Microsoft Azure Machine Learning and IBM Watson Studio.
- Automate customers’ labor-intensive workflow of labeling of AI training data through a cloud-based and crowdsourcing-centric alternative to the recently launched general availability of Amazon SageMaker Ground Truth.
A key step in enterprise cloud AI and data management journeys is to attend Google Cloud Next 2019, which is taking place April 9-11 in San Francisco. And theCUBE will do live interviews with Google executives, developers, partners and customers during the conference.
