The Administrative State and Computability
I would like to record some thoughts in light of the recent news regarding Robo-debt and government surcharging.
The key issue is that there is a clash between two different forces in administrative law:
- The legal system (legislation, regulation, court rulings etc.) appears to routinely create obligations on organisations (we'll focus on government for now) that are not capable of being implemented in a computerised manner.
- The administration of public services, given the scale on which it operates, is physically incapable of performing its obligations other than through a digital/computerised
As a result, governments are responding through some combination of the following behaviours:
- Fail Open: Giving up on assessing claims and yielding in favour of any counterparty
- Fail Closed: Non-compliance with the law, frustrating counterparties
- Retroactive Determination: Retroactive immunity for historically illegal conduct
Legal obligations that are non-computerisable and therefore infeasible to implement
We have recently seen governments in Australia confront the following problems:
- Re-assessing robodebt claims without using income averaging cannot be done in bulk by a computer
- NSW government recently discovered that it has been illegally surcharging consumers for many years but cannot determine who needs to be refunded
- The Commonwealth government, upon being prompted by the news in NSW, has similarly discovered that it has been illegally surcharging consumers based on payment method, in contravention of Cth law
A few quotes to illustrate the point.
On the infamous "income averaging" of Robodebt:
Even where there is enough information to reassess the true value of the debt, these take an average of five hours to process. A conservative estimate suggests there are at least half a million such debts that could need recalculating.
On the Commonwealth Government illegally surcharging consumers:
However, government sources say there are no plans to refund people charged extra for using their cards, due to the challenges of auditing tiny merchant fees across government services over so many years.
Or a fun one from the US, with a contrasting example:
In 2014, California passed Prop 47, a referendum that reclassified all burglaries of property under $950 from felonies to misdemeanors, and applied retroactively. But rap sheets do not include a code for whether the value of the stolen property was under $950, meaning people trying to change their criminal records would have to dig through paper files and hope they could find enough evidence for a judge to conclude the theft qualified. As Pahlka says, “the way the law was written makes automated expungement impossible.”
Two years later, California passed Prop 64, which expunged marijuana-related felonies and was relatively simple to implement in an automated fashion. The Code for America team developed an algorithm to scan electronic court files and identify eligible records. This tool has already found 144,000 convictions eligible for expungement. “Instead of decades of work,” Pahlka writes, “the software program had taken about thirty seconds to run.”
The nature of infeasibility
A few key points:
- Because implementation occurs on a sufficiently wide scale, it requires administration through computerised means, which are presently the only way of administering a system at a large scale.
- The correspnding limits on implementation are actually defined by the limits on computerised systems. In terms of practical limits, these are usually limits on what is observable, stored or recognisable by a computer, in an economically feasible way. (The recent developments on AI are therefore very interesting, because they expand the scope of what can be observed and interpreted in bulk by computer systems.)
- The limitations of computability exist nowhere in traditional legal systems. As the legal system creates more and more obligations without regard to computerizability, this situation arises more frequently.
Responding to the problem
Government's respond to such problems in one of the following ways:
- Fail Open: Giving up on assessing claims and yielding in favour of any counterparty
- Fail Closed: Non-compliance with the law, frustrating counterparties
- Retroactive Determination: Retroactive immunity for historically illegal conduct
Fail Open
One possibility is just to yield to the counterparty in an automatic fashion. From the earlier article on possible resolutions to Robodebt:
The debts themselves average about $500, which might be a lot of money to a person but is not worth the effort for government to re-argue. Waiving debts en masse, however, raises the prospect of having to pay back hundreds of millions of dollars, if not more, in unlawfully raised overpayments.
Arguably, a similar thing is happening with asylum claims in the US - with asylum seekers getting the right to stay in the country until their case can be determined, a process that often includes waiting years for an initial hearing.
Fail Closed
Another possibility is to simply fail by refusing the relevant action. Arguably Robodebt continues to partially fail closed as it has not actually refunded all of the relevant parties.
Retroactive determination
An even more fantastical solution is just to retroactively legalise your conduct.
On the Commonwealth Government collecting illegal surcharges:
Laws will be introduced on Monday to retrospectively make these surcharges legal.
Extended edition: Private sector also included
I will note that a relevant story is that AMP once called me up, several years after I'd stopped being a customer. This was post-Royal Commission, and they were calling to tell me that they had miscalculated my interest and needed to know where to send me a cheque for ~$0.60. I told them to forget about it. Certainly, the cost of paying someone to make that call would have been more than the debt but something tells me that they were beyond rational cost-benefit analysis at that point.