By making enterprise-grade computing available to entrepreneurs at peppercorn rates, the rise of cloud computing fueled a huge rush of innovation. In a similar vein, open-source tools, cloud storage and automated machine learning platforms mean businesses can begin to dabble in artificial intelligence without prohibitive costs — and risk.
Extending far beyond academic institutions and research labs, AI services delivered by cloud providers and paid for on a per-use basis are removing some of the biggest barriers to entry. These dynamics may go some way to explaining why a 62-year-old field of computer science is currently the source of equal parts hype, promise and job-destroying anxiety.
While the opportunity is real, practitioners Which-50 spoke to cautioned that success still depended on a high level of digital literacy, project governance, and forward-looking mindset.
A recent report from McKinsey Global Institute argued the AI revolution is no longer in its infancy, but the main economic impact of artificial intelligence hasn’t yet arrived.
The consultants predict companies that embrace AI will double their cash flow by 2030, while those that don’t could lose 20 per cent of their revenue.
Olivier Klein, head of emerging technologies AWS, says given the low barriers to entry, every company should make use of AI and machine learning services to future-proof their business.
“Before cloud, building AI tools required many specialists, multiple iterations, vast compute power and large budgets, placing it out of reach for most,” Klein told Which-50.
Released last year, Amazon SageMaker now provides modules that enable everyday developers to quickly build, train, and deploy machine learning models at any scale. Klein says Amazon’s cloud computing clients frequently use the pay-as-you go model to quickly test and validate ideas and concepts in the AI space, without large up-front costs.
“This iterative approach removes the question of whether an AI project should start big or small because it’s now easy for companies to experiment without risks of large-scale failure. Cloud technology means infrastructure is available whenever a small project is ready to grow.”
According to Klein, most businesses will soon be entering the “golden age of AI” and it’s important to not get left behind.
“Now that AI services can be consumed as needed, the risk from ignoring the movement is more profound than the risk of exploring its benefits,” Klein said.
“AI will revolutionise almost all aspects of technology. It will reduce the time taken for tasks such as product fulfillment, logistics, personalisation, and language understanding; while introducing shifts that will completely change business models, such as the arrival of connected cars.”
Sean Alexander, principal director of engineering, Microsoft AI, says it’s now table stakes for businesses to store their data in the cloud to take advantage of AI tools to recognise patterns in businesses systems of records.
“Machine learning has been around for many generations, but what has changed most recently over the past four years is the massive amounts of compute that are now available to look for those patterns in the data and yield new insights,” Alexander told Which-50 during a recent visit to Australia.
For example, Microsoft CFO Amy Hood has a system that she calls ‘Modern Finance’ which depends upon deep learning taking both micro- and macroeconomic data to build out a quarterly forecast for the company, “which is significantly more accurate and delivered faster than throwing human capital at that problem,” Alexander said.
According to Alexander, using traditional methods for machine learning used to take 24 hours to run a single simulation. “Using things like Azure Batch AI, I can spin up hundreds of thousands of CPUs or GPUs and literally run decades worth of simulations overnight and only have to pay for the compute time in the cloud that I am actually using to inference a given model.”
“That pace at which things are changing at business, we are really starting to see take off over the past four years.”
Alexander said as AI solutions mature the costs are expected to continue to come down.
Short termism dilemma
McKinsey’s research into AI has highlighted another phenomenon; companies and countries that don’t embrace AI now will find it increasingly difficult to catch up, with early adopters displaying considerable advantages over their rivals.
“Their ability to reinvest these gains and pull even further ahead of competitors may create an insurmountable advantage, and increases the importance of all companies to consider how automation and AI could affect their businesses,” the report states.
According to an analysis released last week by Data61, Australia’s digital R&D spend is declining while the OECD average is increasing.
“Our companies are being forced to focus on short-termism. They are also coming under increasing margin pressure from global competition and they are viewing that R&D investment in longer-term digital projects as being discretionary. They are effectively mortgaging their future to focus on the here and now,” Adrian Turner, CEO of Data 61, told Which-50.
“There is no doubt that as every industry becomes data-driven, the global platform companies are thinking long-term in the way that they are investing. Amazon, for example, thinks in seven-year horizons. Companies need to capture the short-term gains but absolutely need to be thinking longer term,” Turner said.
“The short-termism is a dilemma.”
Sri Annaswamy, founder and director of Swamy and Associates, argues Australia has a critical role to play both as a consumer of AI but also as one of the handful of global AI hubs, alongside Silicon Valley, China, India, Singapore and Israel.
He says there’s no dichotomy between demonstrating short-term value and long-term investment to build relevant skill-set.
“My overall view on AI and indeed, ‘emerging technologies’ is very simple — they can be demonstrated to work in POCs and pilots but the real challenges lie in scaling them up across business and functional groups in a cost and time-effective manner,” Annaswamy said.
Annaswamy argues business should focus on executing proof of concepts that demonstrate short-term value, and can be scaled up rapidly to avoid being caught in a perpetual cycle of piloting. Neither of those things require spending a fortune on technology or salaries.
“Scaling AI today does not require huge budgets and can be done even by a small organisation on a shoestring budget due to open-source AI tools, cloud storage, onshore/offshore centres of excellence and machine learning automation platforms,” he said.
Annaswamy points the finger at vendors and consulting firms for creating a “wide-spread fear of ridiculously heavy technology and consulting investments for AI projects and AI strategies”.
The danger of innovation labs
Annaswamy argues AI and machine learning projects should be judged by the same ROI measures as other technology investments.
“I am firmly of the ‘prove value and scale rapidly’ school of hard knocks, if you will. Otherwise, you will simply build organisational white elephants with tonnes of PhDs and expensive technology and very little value coming out it — what is often called the ‘perpetual pilot trap’.”
He warns, “The worst you can do is create an ‘Innovation Labs’ type white elephant full of Phds and technologists that just keeps chewing up investment dollars without anything tangible by way of demonstrated value!”
Vectore founder Shamima Sultana, an AI specialist with a deep practice knowledge built by working in some of Australia’s most successful financial services companies, has a similar warning about the dangers of AI labs.
“Something that has originated in a lab, always stays in the lab in a way,” Sultana told Which-50.
The danger is that even a successful proof of concepts isn’t always converted into organisational capability.
“Without a proper champion, without a proper strategy, it never gets adopted within the organisation.”
Sultana recommends the proof of concept team sits with the business unit. Being in the same physical location not only builds the relationship with the end-users of technology (which increases the chance of success) but means the proof of concept will be enterprise-ready, cutting down the time and red tape required for a technology to meet business criteria.
She notes that it is necessary to have governance in place at the proof of concept stage to identify what success looks like, what commercial value it will generate and, if successful, are you willing to roll it out to the organisation?
But before embarking on a proof of concept, Sultana also recommends clients explore AI before seeking out a business problem to apply it to.
“It is quite a valuable tactic to begin with, it allows you to understand what you are dealing with before defining a problem and what kind of business problem you could solve with it,” she said.
For corporates exploring AI, Sultana recommends reaching out to start-ups, to learn about what they are working on and then find an area where it could be valuable inside their own organisation.