putilov_denis - stock.adobe.com

Putting artificial intelligence and machine learning workloads in the cloud

We look at the pros and cons of putting artificial intelligence and machine learning applications in the cloud

Artificial intelligence (AI) and machine learning (ML) are some of the most hyped enterprise technologies and have caught the imagination of boards, with the promise of efficiencies and lower costs, and the public, with developments such as self-driving cars and autonomous quadcopter air taxis.

Of course, the reality is rather more prosaic, with firms looking to AI to automate areas such as online product recommendations or spotting defects on production lines. Organisations are using AI in vertical industries, such as financial services, retail and energy, where applications include fraud prevention and analysing business performance for loans, demand prediction for seasonal products and crunching through vast amounts of data to optimise energy grids.

All this falls short of the idea of AI as an intelligent machine along the lines of 2001: A Space Odyssey’s HAL. But it is still a fast-growing market, driven by businesses trying to drive more value from their data, and automate business intelligence and analytics to improve decision-making.

Industry analyst firm Gartner, for example, predicts that the global market for AI software will reach US$62bn this year, with the fastest growth coming from knowledge management. According to the firm, 48% of the CIOs it surveyed have already deployed artificial intelligence and machine learning or plan to do so within the next 12 months.

Much of this growth is being driven by developments in cloud computing, as firms can take advantage of the low initial costs and scalability of cloud infrastructure. Gartner, for example, cites cloud computing as one of five factors driving AI and ML growth, as it allows firms “to experiment and operationalise AI faster with lower complexity”.

In addition, the large public cloud providers are developing their own AI modules, including image recognition, document processing and edge applications to support industrial and distribution processes.

Some of the fastest-growing applications for AI and ML are around e-commerce and advertising, as firms look to analyse spending patterns and make recommendations, and use automation to target advertising. This takes advantage of the growing volume of business data that already resides in the cloud, cutting out the costs and complexity associated with moving data.

The cloud also lets organisations make use of advanced analytics and compute facilities, which are often not cost-effective to build in-house. This includes the use of dedicated, graphics processing units (GPUs) and extremely large storage volumes made possible by cloud storage. 

“Such capabilities are beyond the reach of many organisations’ on-prem offerings, such as GPU processing. This demonstrates the importance of cloud capability in organisations’ digital strategies,” says Lee Howells, head of AI at advisory firm PA Consulting.

Firms are also building up expertise in their use of AI through cloud-based services. One growth area is AIOps, where organisations use artificial intelligence to optimise their IT operations, especially in the cloud.

Another is MLOps, which Gartner says is the operationalisation of multiple AI models, creating “composite AI environments”. This allows firms to build up more comprehensive and functional models from smaller building blocks. These blocks can be hosted on on-premise systems, in-house, or in hybrid environments.

Cloud service providers’ AI offerings

Just as cloud service providers offer the building blocks of IT – compute, storage and networking – so they are building up a range of artificial intelligence and machine learning models. They are also offering AI- and ML-based services which firms, or third-party technology companies, can build into their applications.

These AI offerings do not need to be end-to-end processes, and often they are not. Instead, they provide functionality that would be costly or complex for a firm to provide itself. But they are also functions that can be performed without compromising the firm’s security or regulatory requirements, or that involve large-scale migration of data.

Examples of these AI modules include image processing and image recognition, document processing and analysis, and translation.

“We operate within an ecosystem. We buy bricks from people and then we build houses and other things out of those bricks. Then we deliver those houses to individual customers,” says Mika Vainio-Mattila, CEO at Digital Workforce, a robotic process automation (RPA) company. The firm uses cloud technologies to scale up its delivery of automation services to its customers, including its “robot as a service”, which can run either on Microsoft Azure or a private cloud.

Vainio-Mattila says AI is already an important part of business automation. “The one that is probably the most prevalent is intelligent document processing, which is basically making sense of unstructured documents,” he says.

“The objective is to make those documents meaningful to ‘robots’, or automated digital agents, that then do things with the data in those documents. That is the space where we have seen most use of AI tools and technologies, and where we have applied AI ourselves most.”

He sees a growing push from the large public cloud companies to provide AI tools and models. Initially, that is to third-party software suppliers or service providers such as his company, but he expects the cloud solution providers (CSPs) to offer more AI technology directly to user businesses too.

“It’s an interesting space because the big cloud providers – spearheaded by Google obviously, but very closely followed by Microsoft and Amazon, and others, IBM as well – have implemented services around ML- and AI-based services for deciphering unstructured information. That includes recognising or classifying photographs or, or translation.”

These are “general-purpose” technologies designed so that others can reuse them. The business applications are frequently very use-case specific and need experts to tailor them to a company’s business needs. And the focus is more on back-office operations than applications such as driverless cars.

Cloud providers also offer “domain-specific” modules, according to PA Consulting’s Howells. These have already evolved in financial services, manufacturing and healthcare, he says.

In fact, the range of AI services offered in the cloud is wide, and growing. “The big [cloud] players now have models that everyone can take and run,” says Tim Bowes, associate director for data engineering at consultancy Dufrain. “Two to three years ago, it was all third-party technology, but they are now building proprietary tools.”

Azure, for example, offers Azure AI, with vision, speech, language and decision-making AI models that users can access via AI calls. Microsoft breaks its offerings down into Applied AI Services, Cognitive Services, machine learning and AI infrastructure.

Google offers AI infrastructure, Vertex AI, an ML platform, data science services, media translation and speech to text, to name a few. Its Cloud Inference API lets firms work with large datasets stored in Google’s cloud. The firm, unsurprisingly, provides cloud GPUs.

Amazon Web Services (AWS) also provides a wide range of AI-based services, including image recognition and video analysis, translation, conversational AI for chatbots, natural language processing, and a suite of services aimed at developers. AWS also promotes its health and industrial modules.

The large enterprise software and software-as-a-service (SaaS) providers also have their own AI offerings. These include Salesforce (ML and predictive analytics), Oracle (ML tools including pre-trained models, computer vision and NLP) and IBM (Watson Studio and Watson Services). IBM has even developed a specific set of AI-based tools to help organisations understand their environmental risks.

Specialist firms include H2O.ai, UIPath, Blue Prism and Snaplogic, although the latter three could be better described as intelligent automation or RPA companies than pure-play AI providers.

It is, however, a fine line. According to Jeremiah Stone, chief technology officer (CTO) at Snaplogic, enterprises are often turning to AI on an experimental basis, even where more mature technology can be more appropriate.

“Probably 60% or 70% of the efforts I’ve seen are, at least initially, starting out exploring AI and ML as a way to solve problems that may be better solved with more well-understood approaches,” he says. “But that is forgivable because, as people, we continually have extreme optimism for what software and technology can do for us – if we didn’t, we wouldn’t move forward.”

Experimentation with AI will, he says, bring longer-term benefits.

Cloud-based AI’s limits and prospects

There are other limitations to AI in the cloud. First and foremost, cloud-based services are best suited to generic data or generic processes. This allows organisations to overcome the security, privacy and regulatory hurdles involved in sharing data with third parties.

AI tools counter this by not moving data – they stay in the local business application or database. And security in the cloud is improving, to the point where more businesses are willing to make use of it.

“Some organisations prefer to keep their most sensitive data on-prem. However, with cloud providers offering industry-leading security capabilities, the reason for doing this is rapidly reducing,” says PA Consulting’s Howells.

Nonetheless, some firms prefer to build their own AI models and do their own training, despite the cost. If AI is the product – and driverless cars are a prime example – the business will want to own the intellectual property in the models.

But even then, organisations stand to benefit from areas where they can use generic data and models. The weather is one example, image recognition is potentially another.

Even firms with very specific demands for their AI systems might benefit from the expansive data resources in the cloud for model training. Potentially, they might also want to use cloud providers’ synthetic data, which allows model training without the security and privacy concerns of data sharing.

And few in the industry would bet against those services coming, first and foremost, from the cloud service providers.

Read more about cloud AI

Read more on Artificial intelligence, automation and robotics

Data Center
Data Management