Artificial Intelligence

Hold on, we’re entering the age of the AI-accelerator

Technology accelerators speed things up, obviously. But then so does extra processing power, additional server space and drinking too much coffee - we are talking about the more considered use of IT accelerators in the new world of AI and ML.


Technology accelerators speed things up, obviously. But then so does extra processing power, additional server space, GPU-charged super boosting and drinking too much coffee.

Simply pumping more juice into an enterprise IT system is not regarded to be an accelerator per se. We are talking about the more considered use of IT accelerators in the new and far more algorithmically advanced world of Artificial Intelligence (AI) and Machine Learning (ML).

Industry accelerators have actually been around for most of our post-millennial existence. SAP has championed their use in its various platform guises as a means of getting customers running with live production systems faster. Through the use of templates, pre-architected application and data services design, customers can start with what clearly is rather more than a blank first sheet of paper.

Sometimes using obfuscated and anonymised datasets to run at system test stage, accelerators can get organisations to market faster, but only if used prudently, as not necessarily as some sort of blanket deployment panacea.

The weight of a thousand clouds

Among the firms now tabling accelerator-flavoured enrichment is Accenture. As part of an extended relationship with AWS, the IT services and consulting gurus at Accenture claim to have had experience working on ‘thousands of cloud projects’ in recent times. This, the company says, gives it the ability to understand the ‘human and business dimensions’ of cloud change at scale with greater speed and certainty

This human-business duality is an (arguably) refreshing way of expressing cloud computing deployment challenges. In live production environments we know that all too many of them suffer from poor integration, clunky alignment and misconfiguration headaches, the latter in particular being one of the problems that key players like Qualys are working to address through the use of Infrastructure-as-Code technologies. 

Over the next five years, Accenture will develop a range of new accelerators to address the biggest challenges in cloud migrations with a goal of enabling AWS innovations to be adopted at what is promised (at this stage) to be up to 50% faster.

To date, Accenture and AWS have co-created nearly 40 solutions for 16 industries with proven use-case relevance in order to ‘jumpstart’ client value. In today’s hyper-competitive era of compressed transformation, organisations must implement change under tremendous time pressure.

“We’ve learned through experience customers success often hinges on two key factors: speed and the ability to adopt new ways of working,” said Matt Garman, senior VP of sales and marketing at AWS. “Our investment to create reusable accelerators and mechanisms with Accenture is intended to make cloud transformation and change programs easier and more predictable — so customers can continuously innovate and lead their industries.”

For Generali Vitality, a subsidiary of one of the largest global insurance and asset management providers, their wellness business group needed to scale quickly to reach new customers. Working with Accenture and AWS to tap into cloud-native technology, Generali Vitality is now able to roll out new features at the push of a button to continuously improve their products and engage customers.

So what really makes an accelerator work at the software code level? How much do accelerator technologies depend on AI & ML algorithmic strength? When is an IT accelerator not an IT accelerator… and where do accelerators go next?

More cores, does not equal smart acceleration

CEO at cloud workload management software company YellowDog Simon Ponsford says that his firm has witnessed the growth of data analytics in the shadow of ever-increasing iterations of high performance servers, each becomes faster than the last.

“The number of cores increases and network throughput is ever improving. Nevertheless, enterprise systems are typically used for running the same applications as they were 20 years ago, only faster... and with more data. This works for some processes, but in many cases, once results data has been generated it requires humans to interpret it,” he said.

 At YellowDog, Ponsford and team say they have come across many organisations that accelerate processes only to find it then takes them, days, weeks or even months for their experts to interpret the results. This is where ML and AI can play a key part, learning from the data and giving viable results in minutes to truly accelerate discovery and provide a shortcut to so-called ‘great business outcomes’, as the marketing people like to say.

An enterprise applications and data platform specialist with a history of working for firms including ServiceNow, Cognizant and the NYSE, Chris Pope agrees with the ‘acceleration does not equal intelligence’ sentiment. He says that software accelerators are only appropriately and effectively deployed if there is a distinct benefit to the business or use problem identified being solved.

“The practical aspects of software accelerator implementation need some care. Should we be asking ourselves whether humans also need to accelerate if we are being given faster results and insights – and, if so, at what point do we accept that we need to hand over some of that decision making to machine intelligence? There are questions to be asked here,” said Pope.

It’s a point well made. As we start to apply more of these super-charged, pre-templated, auto-tuned accelerator controls to the coalface of business, we should be looking introspectively at the engineering DNA we are feeding into these self-leaning engines. Having seen a wide variety of customer use cases run in every industry vertical imaginable, Pope heeds some cautionary advice and questions whether badly built systems will start to look for problems that aren’t actual business problems.

“Crafting human ingenuity, empathy and problem solving scope into the new software constructs being built to accelerate our future operations is not a one-click task. The old adage of ‘trust and verify’ still applies if we are to create acceleration controls that really work… especially as we look 10-years into the future and start to accelerate our accelerators,” said Pope.

Our accelerated future

We may not quite get to ‘push of a button’ simplicity in every deployment scenario, but the drive to apply IT accelerators to the coalface of business is on.

Undeniably, the advancements that have come forwards in algorithmic logic and power are a big part of the way these technologies can now be applied. We are also at a far more advanced stage in terms of our understanding of the ‘shape’ of data, so much so that so-called ‘data exchange’ platforms now exist where a business can actually trade their anonymised datasets with other firms inside of their business verticals and specialisms.

Like any physical automobile accelerator, nobody should be driving with a heavy right foot that stamps too hard on the gas, so prudence and patience is still needed for enterprises that wish to accelerate inside the IT stacks and drive at speed.

We’re building IT accelerators, but we’re not necessarily building a particularly sophisticated braking system or any formalised version of a reverse gear, so please look both ways.