Great article Roger. You are spot on when you mention these models can now automatically get smarter the more information we feed them. This is the key difference to how software teams have worked for the past 20 years. Until recently we solely created coded accounting and SMSF rules that remained static until someone, either the accountant or the software developer invested time to update the rules.
As a software company we still produce and invest in a lot of compliance based static rules. particularly when the ATO introduce a tsunami of compliance changes such as CGT Reset ,1.6 Caps and TBAR. This is slowly beginning to change. Recently we released a new AI Assistant to help accountants and administrators detect and locate possible exceptions and outliers, in either human entered or automated data in the Pre Audit phase of the Accounts preparation process. This pre audit detection process is evolving and learning as the data set grows and as a result automatically produces more certainty in the predictions and detection of possible errors. We have more recently progressed to using a specialised field of AI called Deep Learning to memorise transaction narrations, amounts, dates, chart of account categories and reliably know when they encounter them again. This categorisation and prediction is no different to what many of us did as younger accountants. We learned the basic anatomy of an SMSF, and applied categorisation patterns to common transaction types.

Of course AI is not all plain sailing, there are many challenges. Our team spent the best part of 2 years researching and implementing AI deep learning. The core weakness of AI rests in its complexity. This complexity can result in AI being applied to the wrong areas which will result in lost opportunities or wasted time. Currently, AI systems are incredibly narrow in what they can do. Deep Learning remains brittle and requires a lot of preparation by humans, just ask the BGL Big Data Team how much time they invest in data preparation. AI also requires highly trained software engineers, special-purpose coding, special-purpose sets of training data, and a custom learning structure for each new problem. This all takes money and time to develop and learn. I encourage anyone investing in AI in the SMSF space to make sure they not only educate themselves on the above weaknesses , but also involve the accountant in any evaluation. AI is all about applying the right technology to the right processes and no one knows the SMSF processes better than the Accountant doing them day in day out. Hope this helps those in the industry to better understand AI and how it can be applied in their business.