Computerworld

Regulation as a platform: How Data61 is turning text into APIs to serve the burgeoning RegTech industry

The cost of compliance is billions of dollars high and rising. Now scientists are applying some logic to the situation

As one Australian Securities and Investments Commission employee put it at a RegTech event in Sydney earlier this week: “If fintech start-ups are tech rich and the banks are data rich then I think it’s fair to say that ASIC is text rich. Our regulatory guidance is all text based.”

And there’s lots of it. ASIC’s guidance around, for example, employee incentive schemes – a relatively straightforward area of regulation – runs to more than 21,000 words over 59 pages.

Digesting, comprehending and acting on this text is a considerable burden on Australian industry and its financial institutions.

Deloitte Access Economics estimates that federal, state and local government rules and regulations cost $27 billion a year to administer, and $67 billion a year to comply with. According to KPMG, the big Australian banks spent between $350 million and $450 million on regulation and compliance each year.

In an effort to reduce this burden, the government is backing proof-of-concept projects by CSIRO’s Data61 group to transform government rules into public APIs.

Paragraph by paragraph, subsection by subsection, volumes of legal text are being converted into machine readable logic, resulting in what researchers have dubbed ‘regulation as a platform’ or RaaP.

Think satnav for regulation,” says Data61’s Leif Hanlen. “So in the same way that satnav is going to tell you how to get through the city without bothering to stretch the roads out and make them all neat and straight, that's what we're trying to do for regulation. We're not trying to reform regulation but rather to make it easy to navigate on top of it.”

From page to platform

The process of turning regulation as text into APIs works something like this:

To begin with you need regulation that is prescriptive in nature; hard and fast rules that an organisation is either in compliance with or not.

“Things that are prescriptive can be converted into a form of logic that, if true then this is obligated, if false then that is permitted,” Hanlen explains.

“About 50 to 60 per cent of legislation you can imagine can be made prescriptive. And that element of most legislation is what you'd call the boring stuff and it’s fairly trivial from the point of your understanding, it is just hard because you have to read 50 pages.”

To that body of text Data61 applies its Defeasible Deontic Logic technology, the result of a decade of research at the organisation, which was formed out of a fusion of the CSIRO’s digital productivity group and NICTA in 2015.

Aided by natural language processing, the regulation text can be scanned in a matter of seconds and the logic suggested. When the regulation is not prescriptive – such as featuring ‘reasonable person’ rules or best practice guidelines – these parts are flagged.

Then, researchers join with regulators to check through the results, working at a rate of around 10 pages per person per hour.

“The entire process is the computer provides suggested logic where reasonable and then the regulator policy owner looks through that and says yes, no, maybe,” Hanlen says.

Finally the logic is coded into a single API which will eventually cover all government regulation and legislation.

Page Break

RaaP game

Data61 has been working with a number of agencies on RaaP pilots, as part of the federal government’s National Innovation and Science Agenda: Platforms for Open Data framework.

With the Australian Taxation Office (ATO), a ‘new business concierge’ tool called PermitMe has been launched. Work is ongoing with the Australian Transaction Reports and Analysis Centre (AUSTRAC) and the Department of Finance.

As well as financial and business related regulation, Data61 is also running a two year project with the Australian Building Codes Board (ABCB) to code elements of the National Construction Code which sets out the minimum requirements for the design, construction and performance of buildings.

ASIC is likely to get involved soon too. Speaking at the InnovationAus RegTech event in Sydney, ASIC senior executive leader, strategic intelligence, and coordinator of the commission’s innovation hub, Mark Adams said: “The concept of software-enabled codification of laws, is something that’s being discussed around the world. And we plan to talk to Data61 in thinking about the opportunity in that space."

Each agency's regulation logic will ultimately feed into the same API, which will sit, open source, at data.gov.au.

“Our aim is to enable third party to develop services and business applications, interfacing with regulation quickly and more efficiently,” Hanlen adds.

In addition, Data61's ‘defeasible logic reasoner’ – Spindle – is also available on Github for free for anyone to build upon.

Progress is being made in “small bites” (“'Hey we're going to automate the entire world of regulation’, that would be a bit insane,” says Hanlen). In each case a minimum viable product is being made; enough to be useful, and enough for developers to build upon.

“You hit different pieces of different acts, and organically build up that computation,” Hanlen says.

Code first, text later

Although similar work is going on elsewhere in the world – such as the present RegTech drive from the UK’s Financial Conduct Authority – Australia is “leading in this space” says Data61 research director Liming Zhu.

The end game is not simply that all regulation be available via API, however.

“In the future, the long term vision is to see whether we can start a new regulation from a computable form, then turn those into text later,” Zhu says.

Rather than translating text to code, regulators would set regulation as computable logic, which could be put into text if needed.

“At the moment what we're doing is essentially backward compatibility, but the end state is print to text if you feel like it but the business rule is the thing that's important it's not the actual words written down on a page,” says Hanlen. "We're trying to work at the creation stage of legislation rather than let's reverse engineer the logic out of the legislation.”