Can’t we just buy AI solutions ‘off the shelf’?
By Dr Paul Dalby, Business Development Advisor, Australian Institute for Machine Learning, the ³ÉÈË´óƬ.
This article is an extract from , a report published in partnership with the .
Investing in the development of in-house AI can seem like a daunting task. There are few people with the skills to understand what’s possible, let alone build it. In the absence of investment in the development of AI, off-the-shelf AI solutions can be an alternative.
But it’s important to note that buying AI isn’t like buying traditional software services. The fundamental strength of AI is that AI algorithms are designed for and trained on specific sets of data. Generic AI algorithms and products are built on generic data and might not be optimisable to solve a particular problem. They may also have unknown inbuilt biases.
AI delivers the best results when it’s designed by experts who understand how to develop and adapt AI algorithms for specific datasets, can test data for biases and can then test the final product for performance and bias.
Australia needs homegrown specialist AI solutions for sectors in which we want to maintain sovereign control, such as agriculture. Photo: iStockPhoto, online.
Because AI is a new technology, and we’re still working out what we can expect from it, it’s easy to assume that all AI is largely the same. That isn’t true, and there have been some spectacular failures in the development of AI systems that then produced perverse or meaningless results. Using trusted developers to create a unique AI system for a particular scenario and obtaining third-party testing of performance standards is often a requirement for building trusted AI.
Constructing a machine-learning system that can accurately learn from often messy data to deliver accurate and unbiased predictive tools, computer vision systems or language analysis tools is both an art and a science. There’s an enormous and potentially critical difference between someone who merely downloads an existing algorithm from an online library and runs it over a dataset and a specialist who deeply understands the structure of data, statistics, maths and coding solutions and has a creative ability to find novel solutions to a problem. That’s why the salaries of the best AI engineers are so high.
The difference in performance and trustworthiness between off-the-shelf solutions and highly tuned AI solutions is due to a number of factors, including data quality, but particularly the quality of the data engineers and AI developers. Australia is desperately short of them, as is the rest of the world. It takes many years of training to produce world-class engineers. Demand is already far outstripping supply, and the growth in demand is likely to continue to outpace the growth in supply for some time yet. Australia must ramp up its investment in training highly skilled technical specialists in AI and data science at the VET, undergraduate and postgraduate levels. My back-of-the-envelope calculations suggest that we currently graduate 100 or fewer PhD students in high-end machine learning per year in Australia. I suspect about half of those get jobs overseas, leaving 50 people per year to be shared among the businesses, government agencies and start-ups that need them. In a Covid-19-afflicted world, it’s harder to fill those gaps with international workers. The solution is a massive investment in this high-end specialisation and a national campaign to attract people to this career.
Australia particularly needs homegrown specialist solutions for sectors in which we want to maintain sovereign control. Those sectors include defence and national security, but may also include some of our large agricultural industries, critical minerals and health care (as examples). It would be unacceptable to most Australians to have those industries end up becoming ‘Uberfied’—where the control of the sector is in the hands of foreign technology companies that collect all the data, undertake all the data analysis, and issue instructions to low-paid Australian workers on what to do next.
For defence and national security, there’s a global battle for supremacy in AI. Most of the AI needed will be for cybersecurity, logistics, managing the data deluge and responding to ‘truth-disruption’ of electoral processes and public messaging. Australia should seek to develop and maintain advantage in these areas. It might seem surprising, but US defence agencies have visited Australia to learn about our superior capabilities in AI for security and related areas. We’re eminently capable at holding our own and developing world-class systems.
There’s a global arms race in the development of capabilities in all these areas, and a difference of 5% in performance can have existential consequences. While Australia is unlikely to be able to compete effectively in the development of large weapons and other large machinery for warfare, we can and should develop a security-focused AI sector that’s globally competitive. For example, another middle power, Israel, has successfully developed a world-class capability in cybersecurity by investing in skills development in its army and universities, and then encouraging those staff to spin out cybersecurity companies through grants, start-up funding and contract purchasing. Australia has the capacity to be globally competitive in AI, which would have the double advantage of giving us superior capabilities in our national services and creating a new and globally competitive industry sector.
Buying off-the-shelf AI solutions might seem like a cheaper solution than developing unique systems, but it may also end up being a false economy if it doesn’t work, or isn’t trusted by its users. For industries with sovereign strategic value, it makes even more sense to build our own systems here in Australia. That won’t happen without substantial new investment in high-end skills. With that investment, the result will be better, more trusted AI, but it will also support highly paid jobs and build capability that will sustain Australia’s high standard of living in this new industrial revolution.
This article is an extract from , a report published in partnership with the .