Story image

IT infrastructure key to AI success

26 Nov 2018
Twitter
Facebook

Article by Seagate Technology vice president Asia Pacific Sales, Robert Yang.

It’s well known that data is the building block that drives Artificial Intelligence which means as a result that the heart of AI's success will depend on a solid data storage strategy.  

Yet according to a study from Seagate, Data Pulse: Maximising the Potential of Artificial Intelligence, one in five IT decision-makers across Asia Pacific are not currently ready to handle the volume and velocity of data that is coming their way.

In Australia, where 75% of respondents said their organisations have already implemented AI in one form or another, one fifth believe they are currently not ready to handle the ever increasing stream of data.

Just under 90% of Australian respondents said their organisations think they should invest in IT infrastructure in order to handle that growing workload from AI solutions both processing and generating an increasing amounts of data.

A further 89% said they believe they require even more robust data storage solutions to enhance and support AI applications and workloads. 

An AI-enabled infrastructure is key to successful AI deployment 

The lack of a clear strategy for AI adoption within the organisation was cited as one of the two biggest hurdles to successful implementation of AI. The other big challenge is having an IT infrastructure sufficiently robust for AI deployment, especially as businesses look to harness insights and information for real-time data processing

According to Airtasker senior data scientist Adrian Letchford, an Australian company that provides an online and mobile marketplace enabling users to outsource everyday tasks, many companies forget the fundamentals of infrastructure requirements when they bring AI into their operations and client service delivery.

“If you look at any project that involves AI, the size of the project may be quite large but the AI component is actually quite small,” he says. 

“The rest of it is infrastructure, your data warehouse, shunting data around, connecting things together, processing requirements etc.”

Airtasker, which has placed AI at the heart of its operations, did so to support its rapid growth when it was finding it difficult to hire enough people to take on some of its key tasks.

“To give you an example, we have insurers covering our workers and the jobs they undertake. To prepare a quarterly insurance report used to take six weeks. After a little bit of research and bringing in some AI algorithms, that came down to about 15 seconds.” 

Letchford has spoken about the need for organisations to set up the infrastructure to support data collection which in turn allows data scientists and AI experts to dig into the data and start building models and enable machine learning. 

“When it comes to AI implementation, you also need to consider the accessibility and processing requirements. You may have applications that are used to hitting a database for the data they need to function. That is not going to be enough to support AI applications, as it doesn’t scale,” he says.

“The idea is that while the R&D phase is happening, the organisation builds up the infrastructure side of things so that by the time the machine learning experts have developed something, the business will have infrastructure and scalable resources set up to implement it,” recommends Letchford.     This machine thinks

According to IDC, the amount of data subject to analysis globally will experience 50x growth to 5.2 billion, and 100x growth of AI-related data expanding to 1.4 billion by 2025. Remember, data is not merely created on our computers, but also generated by myriad devices in the real world – in the office, smart homes, and smart city infrastructure. 

The internet of things implies that much of what AI needs - data and more data - is generated and collected from multiple sources, without human involvement, and increasing at unprecedented speeds. To generate insights and make decisions in real-time via AI, this staggering amount of data will need to be integrated, managed, and processed end-to-end. 

It’s daunting then to learn that almost 20% of the respondents from Australia in the Data Pulse Study believe that their organisations have yet to invest sufficient man-hours and budgets in AI development and implementation.

More than 60% struggle to know where and how to start their organisations’ development and implementation of AI and 92% call for more investment to be made in understanding the value of AI to their organistions. 

The success of AI depends on many things. However, organisations need to ensure that their IT infrastructure – so often simply regarded as basic technology that “keeps the lights on” – needs to be AI-enabled to provide a backend (storage, access, compute, transfer) sophisticated enough to create optimised environments that ensure AI deployments are well-supported. 

That makes the difference between an AI that can serve up a solution with a ‘here you go!’ and one that constantly asks: ‘Can you please repeat the question?’