AI gets involved with the law

IF YOU are hoping for parole, you better hope the judge has just eaten. A recent study of the behaviour of parole board judges revealed this interesting human trait. The law isn’t supposed to put the rumbling of stomachs above fact and reason, but that’s the problem when those charged with administering it are human.

Computers may be able to help. Software tools are already important in the legal world, especially for big cases like company mergers, where algorithms help people comb through vast piles of documents.

But the application of artificial intelligence to the law promises to go beyond document mining. It aims to let automated systems handle arguments where the logic is not clear.

The first signs of this are starting to appear. Last year, Tom Gordon of the Fraunhofer Institute in Munich partnered with German company Init to start developing an AI application called Elterngeld – German for “parent’s money”. It is designed to make automatic decisions on child benefit claims to the country’s Federal Employment Agency (FEA), probably with some human auditing of its decisions behind the scenes, Gordon says.

Elterngeld is based on the open source Carneades software, designed by Gordon, which takes human claims like “I require government assistance to support my 5-year-old child” and determines whether the statements put forward to support the claim are justified based on the tenets of the law. Each statement is broken down and coded in a machine-readable format which the system then compares with elements of the law, using this to score the claim.

Gordon will present a web-based version of Carneades at the International Conference on Artificial Intelligence and Law in Rome, Italy, this June.

The developers are now in talks with the FEA about how to deploy the system, although it is not yet ready to replace humans. That’s partly because it still needs the text of each law to be broken down into a structured, machine-readable format – a painstaking process that at present must be done by hand. Gordon hopes that one day, new laws will be drafted with machines in mind from the start, so that each is built as a structured database containing all of the law’s concepts, and information on how the concepts relate to one another. This would allow artificially intelligent software to implement legislation on a wide scale.

“If new laws were drafted with machines in mind, AI could implement legislation on a wide scale”

Using a method similar to Gordon’s approach, Anna Ronkainen, co-founder of legal tech firm Onomatics in Helsinki, Finland, is building a tool called TrademarkNow. This measures the similarity of new trademarks to those already in existence, to help avoid potential legal issues. It works by mining the databases of the US and European trademark registers, and looking for similarities.

These are the simpler legal issues – AI software isn’t yet sophisticated enough to take on the mess of abstruse language and sometimes contradictory logic that tends to plague legal systems. But legal artificial intelligence is poised to take off, driven by the cost savings offered by machine judgement, says Ronkainen. “There is a huge movement brewing,” she says. “Corporations are no longer happy to pay huge legal bills.”

This article appeared in print under the headline “AI rules from the bench”