In most countries around the world, the laws that are written as very clear – they are there in black and white, and there’s no room for interpretation. But as human beings, there is a constant need to ensure that things are correct, things are right, and things are as they should be.
This leads to a need for court rooms and judges, trials by jury, and a decision on guilt or innocence being made as an interpretation of the written laws weighed up against the actions of an individual.
In some cases, people have been convicted for crimes that in later years were deemed to not be crimes at all. For example, before California’s Proposition 64, the possession and use of marijuana was illegal in the state. In 2016, it became legal.
This simply meant that for those with prior convictions in relation to marijuana, they could have them struck from the record – an important move, as convictions of that nature could impact the opportunity to get a job.
Entering the Digital Era
The city of San Francisco had been digitizing its files since 1975, so it would seem to be a simple matter to flick a switch and update all the records that were held, quickly removing the convictions. Unfortunately, that’s not how digitization works.
What San Francisco (and many other cities) had stored was effectively a photograph of paper documents, viewable on a computer, but completely unreadable by a computer. To have a conviction removed, an individual must fill in the paperwork to begin the process, and that paperwork may not be simple at all.
San Francisco estimated that around 10,000 people could apply to this process, but by 2018 a total of 23 people had identified themselves. The city took matters into their own hands and spent a lot of time manually going through the documents and found 1230 eligible people. There had to be a better way.
Through the use of optical character recognition (OCR), automation, and artificial intelligence, a program was devised that could process the digitized records and attempt to understand what they said. If a record included mention of a violent crime, the program discarded it as the individual would be ineligible. If there was any doubt about the content of a file, a human was notified for further input. But overall, the software ran itself and produced a result.
Within minutes, it was able to identify another 8000 people who were eligible to have their convictions cleared.
A judge signed them off in one go, and the criminal records of thousands of people were suddenly revised, opening new horizons for each one – and saving the city (and those individuals) a massive amount of work.
How does it work?
OCR has been around for quite a long time, but without the ability to understand, it is mostly useless for tasks of this nature. The ability for a machine to learn, and to thereby understand, gives the opportunity for thousands of documents to be processed by a tireless machine that doesn’t make mistakes in the same way that a human might. The AI industry is prone to calling these algorithms “robots”, as the dictionary definition of a robot includes “…that are able to replicate certain human movements and functions automatically” – by reading and understanding, they are replicating what a human could do.
The artificial intelligence portion of such software allows true understanding and learning. In its simplest form, you could use a keyword search to determine if “violent crime” featured on any record, but the context is important. If the record stated, “this person has never been involved in violent crime”, it shouldn’t be discarded. If that sentence ends with, “until now”, then it should be discarded – machine learning allows these nuances to be understood.
In the California example, if there was serious doubt, a human was called in. It is unknown whether the human interaction allowed the algorithm to be updated – true machine learning would include the ability to make future judgments based on what was learned from how the human reacted.
The future of AI and Law
A precedent has now clearly been set that AI systems can analyze records and find convictions and flag them to be removed, but how far can it go?
Finding a jury that is truly impartial is always difficult, whereas a machine will be impartial by default. Could a program be developed that would be capable of reading court documents, understanding them, and producing a verdict immediately? More importantly, would the public trust such a machine?
It seems as if it is science fiction, but it is much closer to reality than you might think.
For many law practices, the ability to process large amounts of paperwork quickly would be truly beneficial. If a summary of a case could be produced within minutes rather than days, it would make the task so much easier. In fact, there are many services available today that will digitize and process business paperwork using software robots. Large companies that send paper bills to thousands of people (for example, credit card companies) already make use of this technology to process incoming payments – but that’s the tip of the iceberg.
Would a trained law practitioner have a better understanding of a case than a computer? Is there even a possibility that law practices could be replace by AI?
There are automated systems currently in place that issue tickets and fines for vehicles that are traveling above the legal speed limit. In some cases, these tickets have been overturned as the vehicle in question was moving out of the way, preventing an accident from happening – could machine learning help to prevent false positives like this? And if so, would it reduce the need for legal representation in such cases?
The legal system has come a long way in the last few decades, and the move towards digital devices is unstoppable. The question is, how far are you willing to trust them within a law practice, and would you have confidence in a courtroom under the jurisdiction of a robot judge? The technology is improving constantly, and it may only be a matter of time before this science fiction becomes science fact.