The New Zealand Herald
Real risk digital IRD watchdog will cry wolf
If we’re monitored by automated systems we should at least know how they work
The Inland Revenue will turn off some online services today for employers and Thursday for the rest of us, while its new IT systems kick in as part of the tax department’s $1.9 billion transformation project.
We’ll see how that goes, and if the improved online services will be available four days after.
Much of what’s coming up looks good.
EY’s Tori Sullivan raised some alarming points though with the IRD using big data analytics and artificial intelligence to hunt down cash jobs, with computers trawling through vast amounts of information to spot anomalies.
Sullivan gives the example of a cafe making only 80 sales of cups of coffee when the industry norm might be 100, being labelled as a tax cheat by the IRD AI.
That could very well be the case, but the explanation might also be more mundane like shrinkage or poor baristas as Sullivan suggests, but good luck arguing that with the IRD.
It’s not quite as bad as it seems. IRD said that there is no bill before Parliament to extend its data collection powers, but it does have analytical and intelligence tools to “detect any non-compliance” using customer information from third parties.
An IRD spokesperson confirmed that there is no obligation to reveal the sources of information used to detect non-compliance but the taxpayers are in practice often told why they’re being contacted to discuss their tax affairs.
This is encouraging, but the analytical and intelligence tools the IRD uses will for now remain under wraps. So, if the AI thinks you’re a tax cheat, it could be hard to defend yourself because you don’t how the technology came to that conclusion. A cafe making only 80 sales of cups of coffee when the industry norm might be 100 could be labelled a tax cheat by the IRD’s AI.
The danger here is relying on computer algorithms and data processing technologies that are difficult to interpret for anyone apart from computer scientists. Taxpayers or their accountants and lawyers, or the few IRD staffers that are left, are unlikely to understand how the technology works.
You could argue that there’s a positive side-effect in that the AI will punish cafes that pull bad shots of espresso or builders not upping their per-nail earnings ratios.
However, If the outcome of IRD’s AI is to force small businesses into keeping even more detailed records for when they have to row against the machine, it’s time for the tax department to go back to the drawing board and rethink what the point of technology should be.
There’s no stopping AI: it’ll be with us whether we want it or not. For that reason, you should as I suggested give it a spin to understand what AI can, and can’t, do.
The problem is that we struggle to make AI as human-like intelligent and able to understand nuances as people can. Instead, we weaponise AI like Google’s Project Maven and the Palantir surveillance engines.
Also, for now the technology is bad at explaining what it does and why. It’s fun when David Walliams goes “computer says no” but not terribly amusing when an IT system says you’ve done something criminal and nobody quite knows why.
For that reason, it’s in everyone’s interest that the AI systems that’ll judge us are open and transparent and the inevitable errors corrected and algorithmic wonkiness in them ironed out.
Such an awareness and open strategy around the tech it uses might actually be the best way for IRD to boost tax compliance, if everyone knows what they’re dealing with.