That said, fresh and future law graduates should definitely keep the tabs on the latest machine learning trends. There are many jobs in the legal sector that machines already do better than by humans, and the capability of deep learning algorithms in the field is growing.
You can break down the jobs performed in the legal environment into three main types of actions. The first is text analysis, the second is argument building, and the third is judgment. All three are being supported with increasing efficiency by algorithms, but on a different scale and with different accuracy levels.
McKinsey Global Institute estimates that around 22 percent of lawyer's jobs and 35 percent of legal clerk's jobs can be automated with the current technology. Meanwhile, according to Deloitte, 100,000 legal roles will have become automated by 2036.
This is a lot of jobs, mostly at the lowest tier of the industry.
carry out legal research,
manage and review contracts,
perform due diligence,
automate tasks (such as divorce),
predict legal outcomes.
That's enough to make most graduates redundant.
Let's take a look at how machine learning algorithms can improve the legal sector's performance, and at the same time reduce the number of people working in the sector.
Deep learning text analysis in the legal sector
Thanks to deep neural networks, machines are getting better at understanding an unstructured text, that is, content without metadata. If they can “read” emails, instant messages, social media posts, they can also process legal documents. In fact, they do it much better than humans, especially if the machine learning model can learn from big sets of data.
You must have seen the typical scene in legal movies and TV shows: young ambitious lawyers spend the whole night in the library looking for traces of some legal precedents to win a crucial case. Finally, they succeed thanks to their talent, knowledge, hard work, and determination.
Document discovery is simple for machines. A deep learning algorithm can easily go through a text in any form and bring the exact result to the courtroom in seconds. No need for a creative claim from an attorney to postpone the hearing.
Back in early 2017, Contact Intelligence (COIN), a program used by JPMorgan, did the job that used to take 360,000 hours for legal aides in a matter of seconds. It is much faster now.
CaseMine, an Indian legal company developing document discovery software CaseIQ, declares a reduction legal research project times of approximately 50 percent. It does the legal "data mining" jobs traditionally performed by young interns. CaseIQ helps judges in making the right decision by providing the full background of precedents and lines of thought related to a case.
The market is highly competitive. This summer, two e-discovery companies Hanzo (a pioneer of contextual collection and dynamic web content software) and Zapproved (a discovery software provider for managing corporate litigation readiness) decided to offer a better product and increase their customer base.
This kind of software uses natural language processing (NLP) which helps machines understand the legal language even in very old documents.
Tasks more difficult than research and data discovery
The expertise and excellent skills of lawyers using their knowledge and intuition to get through vast volumes of not-yet-digitized books, briefs, notes are getting irrelevant as the legal system is getting increasingly transparent thanks to digitization. In the future, everyone will be able to search for files in a matter of seconds.
A due diligence review is more challenging for the AI, but AI is gradually getting the hang of it. Due diligence consists in analyzing contracts and looking for similarities and anomalies in language and terminology. It takes a skilled lawyer days or weeks.
An artificial neural network can do it within seconds, recognizing patterns and providing suggestions that can be then accepted by a human supervisor. It entails massive savings for legal companies and huge profits for legal services providers who will embrace the technology first.
Kira Systems, a Toronto-based startup offering machine learning contract search, review, and analysis has recently raised $50m in the first external funding round. The company boasts that since early 2017, five of the top 10 firms ranked by the Vault (a directory ranking the most prestigious law firms in the US) have signed up to use the service.
The most prominent competitors of Kira Systems are LawGeex and eBrevia. The market is growing very fast – machine learning solutions can already sort contracts much quicker and with higher accuracy. The price businesses are willing to pay for such services has remained stable, so far.
Al winning a case with a dazzling closing speech?
Argument building is a much more challenging task for computers. Nonetheless, machine learning engineers and scientists are already training AI systems to argue. For now, AI systems are arguing with one another.
It will take some time before you could hire a robot to represent you in a complex court case, but some rudimentary applications have already been developed. A London based startup DoNotPay has been developing a legal bot that initially helped drivers dispute parking tickets. But the bot's capabilities have grown, and now it helps people claim asylum in the US, the UK, and Canada.
Divorce is another type of legal process currently being automated. Companies such as Wevorce in the US or Neota Logic in Australia use technology to help couples agree on divorce terms faster and cheaper. Partners can define their “optimal outcomes” and the AI system helps them reach a consensus. The process is overseen by human legal experts that intervene whenever necessary.
Obstacles to machine learning development in the legal industry
The main problem of ML in the industry is the lack of accessible data. Data is the fuel for machine learning. In the legal environment, many valuable documents are kept in private repositories. This is a barrier for innovative disruptors, making the market less competitive and more profitable for the established players, as developers need to strike expensive deals in order to gain access to the data.
There are initiatives aiming at removing this main obstacle. Harvard University's CaseLaw Access Project has already achieved its milestone by scanning 39,796 volumes and 38.6 million pages of legal material covering 334 years of American case law. The data are being digitized and prepared for machine learning, which is a great step towards democratization of AI services in the legal domain.
Moral concerns about robot judges
The third type of activity central to the legal process, and probably the most important of them all, is the judgment. Robotization works here, but it raises high ethical concerns. The famous case of an algorithm predicting the verdicts of the European Court of Human Rights with 79 percent accuracy is already two years old. Which is very old in terms of machine learning. Now it works much better for sure.
The case is similar to any other aspect of automation: who will bear the responsibility for the decision? We can trust an AI driver, pilot, doctor, or even a judge. Probably in most cases, they will all act faster and make fewer errors. Taking a personal accountability for a mistake is what they cannot do as they will never have their skin in the game. On the other hand, no machine learning engineer will take the responsibility for all the verdicts an AI judge will make.This is a hard obstacle for AI development in the professions that involve decisions where the life or death of real humans is at stake.
Drifting in the direction of an "AI augmentation"
Just like in the case of the most specialized ("humanized") jobs, such as cybersecurity, machine learning in the legal environment is developing in the direction of solutions that jointly engage humans and machines. Algorithms will help the most knowledgeable and skilled lawyers do their work faster and with better results. The rich will get richer, while the less talented or more disadvantaged young professionals will probably be pushed out of business.
The probability of every lawyer's job being automated is low, while the probability that some legal jobs disappear seems to be very high.
At the same time, the businesses and customers are likely to receive a more equal, more transparent, and cheaper legal services in the future. Hopefully, the law will become more egalitarian thanks to the AI.