Jon Bing of the University of Oslo is speaking on the topic of “The Computerisation of Legal Decisions”, though it is indeed a broad canvas, including the history of State data collection in Norway and a real insight into the modern administrative state. He explains the Norwegian systems – e.g. a long-established unique personal identifier and the related recognition of the ‘household’ (address, others domiciled at same address). This is then related to tax authorities (reminding us of course that personal income is published as a matter of course in Norway and other Scandinavian countries). Calculation of benefits and various other functions (even calculating ‘fair housing costs’) is then possible – so taking the housing aid system as an example of ‘automated decisions’ in social welfare systems – a legal decision, even if not a ‘fancy trial decision’. This can be appealed, and the rate was about 10% – so they abolished the appeal system. There are still methods for checking and correcting data. This system has been in place since 1972, written in COBOL. While lawyers are likely to concentrate on hard cases, this is a practical example of computerised decision-making. Bing explained the process of moving from legislation either to lawyers or to computer programs. The understanding of the legal norm is then used to code (i.e. move from natural to computer language). The problems are obvious – for example, the programmer is innocent of legal training. The legislation and program should mean the same thing. Though an interesting twist on this is where in one situation there was a problem with the inconsistent treatment of fractions (they changed the law to accommodate!)

Another case study is the calculation of disability benefit in Germany – based on the average income of the last five years; there was an exception for having served military service, but this was not defined precisely enough. Take for example someone dismissed during service (one day in vs one day from the end?). This shows that, with automation, you really cannot include vague terms – everything must be defined beyond argument. Bing suggests some strict criteria (examples in brackets) – measures (weight), relative measurements (smaller), natural status (sex), legal status (authorised). In natural language, vagueness is usually resolved through textual context. Also, note legal expert judgements (circumstance-based, non-deterministic), which have a certain process but an uncertain outcome. On the other hand, complexity can be a non-problem in computer decision-making; this is often why legal expert is necessary (because there are too many possibilities for a ‘rule’ to be judged) – does this mean that there is a trend towards complex structures based on strict criteria?

The development has been:

The first generation: computerised support, e.g. a case officer facing a client, prepared in the traditional way but has access to a legal information system, internal databases or third party data.
The second generation: paper case replaced by computerised form (collection), with the similar info system and databases – but the system integrates legal knowledge (embedded), e.g. help functions as alternative to legal reference.
The third generation: ‘self-serviced public administration’, client-oriented and interpretation of law is integrated. It is more the execution of public authority than a ‘service’.

Qs from the audience:

– do jurisdictions with strong judicial review have difficulties with this approach, as it’s not so easy to ‘abolish appeals’?
– to what extent could you have an automated, first-stage appeals system?
– does the system itself encourage quite conservative decision-making?