by Franziska Jandl
Currently, artificial intelligence only takes on highly formalizable and standardisable processes in legal technology. But that does not have to stay that way.
IT systems with artificial intelligence (AI) make use of the machine learning method and big data to communicate with humans and to learn constantly.
AI has been particularly successful in areas with a high degree of digitalization. The demanding acquisition of data and the complexity of legal language and terminology are still obstructing its use by legal experts.
The law firm CMS is planning to use the AI software Kira to save both time and costs for due diligence audits.
Now is the time to develop strategies for the use of AI throughout the entire company.
An Overview on AI
Expert Opinions on Theory and Practice
Artificial intelligence is primarily based on the currently popular deep learning algorithms which build giant neural networks, thereby mimicking the functions of a human brain. This way, these systems are able to learn. The increasing degree of digitalization and computer performance lead to more and more data being available. Therefore, the learning curve for speech and text processing or image recognition is quite steep.
For example, the quality of the translation provided by Google Translate has increased significantly during the past few years. Current technology has evolved to a point where it is often difficult to say whether you are communicating with a machine or a human being.
A scientific experiment which involved chatbots talking about random topics for 25 minutes achieved extremely high ratings of those self-learning dialogue systems.
What is Fact and What is Fiction?
AI is based on huge amounts of data. Therefore, it is most useful in areas with huge amounts of digital information, such as image recognition in social networks. However, the field of medicine also quite successfully integrates AI in its processes: Due to its access to medical databases all around the world, IBM Supercomputer Watson was able to identify causalities in cancer therapy that had been previously unknown.
However, legal evaluation of facts is not based on huge amounts of text data, but requires a profound knowledge of legal language and terminology which differs from one country to another. Jurisdiction and laws are subject to constant change. Therefore, acquiring data is far more difficult, also because some documents are confidential and won’t be obtainable at all. So, we see, AI is far from replacing legal experts any time soon.
Moreover, AI lacks emotional intelligence and ethical comprehension. For example, chatbots will advocate quite radical views once they have been fed biased data. Ethical concerns won’t even compute for the machine.
Im Gegensatz dazu beruht die juristische Bewertung von Sachverhalten nicht auf massenhaften Textdaten, sondern erfordert ein tiefes Wissen in der Rechtssprache, die in jedem Land variiert. Rechtsprechung und Gesetzgebung ändern sich ständig. Die Datenakquise ist kompliziert, weil nicht alle Unterlagen frei verfügbar sind. Noch kann KI den Juristen also nicht ersetzen.
Darüber hinaus mangelt es auch an emotionaler Intelligenz und ethischem Verständnis. Chatbots vertreten beispielsweise sehr radikale Meinungen, wenn sie mit einer tendenziösen Datenbasis „gefüttert“ worden sind. Ethische Grundsätze hat die Maschine nicht.
Consequences for In-House Counsels
Now is the time for companies and their legal departments to develop strategies for using artificial intelligence. You see, soon software will be able to do more for us than just taking on tedious repetitive tasks. Many researchers are now focusing on individual disciplines, such as law. Huge Internet corporations are very interested in tackling the business models of lawyers due to the high remuneration involved.
“AI is one of the pivotal technologies to allow robot cars to independently interpret data and predict the behavior of other road users. The society has to start discussing possible reactions of autonomous cars when accidents with personal injuries are inevitable.”
Dr. Dirk Hecker
for legal tech software with artificial intelligence
Drooms: This AI software employs new image recognition methods to analyze scans and images within a data room which had previously been impossible to analyze.
IBM Ross: Ross replaces research and legwork by counsels. It knows the relevant legal provisions. However, this faculty only came with IBM Supercomputer Watson.
Leverton: The contracts governing property-related transactions often consist of hundreds of pages. This software points out relevant clauses to portfolio managers and recognizes, amongst others, whether or not the tenant of a commercial property has been granted special termination rights.
RAVN: This AI software is able to search (inter alia) the documentation of any given property transaction for specific terms faster, more efficiently, and at lesser costs.
Practical Test of AI Software Kira at CMS
The initial situation:
Due diligence audits for M&A transactions will always involve huge piles of documents to be scrutinized by young professionals for any provisions governing the change of control. Moreover, the intense competition forces law firms to look for new ways to work more efficiently and lower the costs at the same time. As for now, legal technology is merely capable of searching terms such as “change of control”. However, these words do not necessarily appear within the corresponding clauses. Relevant provisions may be hiding behind a different wording.
“The important question is not whether or not a machine can outperform a human being, but rather if it enables us to perform our work more efficiently at less costs.”
Dr. Frederik Leenen, Counsel und Head of Legal Tech, CMS Deutschland
Kira is supposed to change this for she will be able recognize not only simple terms but entire causalities thanks to its AI. The use of this software, which has been developed by Kira Systems from Canada, is part of CMS’ strategy for digitalization. Artificial intelligence will support better and faster handling of extensive amounts of data, thereby improving consulting services.
Instead of going through a pile of documents manually, the lawyers can now focus on their actual task: legal review. Kira’s learning methods are specifically designed for transactions. The first field test will be in due diligence settings.
“Kira’s artificial intelligence enables her to recognize causalities on the basis of probability forecasts: The position within the document, the surrounding wording, and hundreds of additional criteria support the likely assumption of the contract containing provisions governing the change of control.”
Dr. Frederik Leenen
At the turn of the year, CMS signed the agreement to implement the machine learning software. Knowledge Management currently trains Kira in German contents. For now, the software only recognizes specific clauses in English data sets, it still has to learn the German language. The more sets of data Kira knows, the better it gets. Presumably as of the middle of the year, CMS’ counsels will be able to conduct first due diligence audits with the assistance of the software. The corresponding AI trainings are currently being organized and assigned to the respective participants.
Training the software for its future tasks brings dependency from the provider. Therefore, CMS places high value on properly documenting the process to be able to repeat it without investing too much additional resources, if required.
However, Kira’s results still need to be checked by a human being. CMS does not expect the virtual colleague to replace the real partner in any law firm any time soon. The decision to use it is based on efficiency: Artificial intelligence will enable us to find and resolve risk-bearing provisions much faster and more reliable. It will give us a competitive edge, because human beings still waste huge amounts of time on this.
Kira Systems was founded in 2011 by Noah Weisberg (CEO) and Dr. Alexander Hudek (CTO) and is based in Toronto. Weisberg previously worked for Weil, Gotshal & Manges in New York, Hudek wrote his doctoral thesis in computer sciences at the University of Waterloo (Ontario).
The company develops software with the capability to automatically identify and extract relevant clauses from contracts of almost any format. According to the company’s own statements, summaries can be created within seconds and analyses that would take weeks of preparation under normal circumstances can now be started within a fraction of that time.
When it was launched six years ago, the focus was on English M&A transactions only. Since then, Kira software has been employed by numerous law firms in different languages. According to the company, CMS will be the first European direct customer working primarily with a language other than English.
Most legal AI service providers started out with the English language, however, “Artificial Lawyer” suggests that, in the near future, AI systems will have to prove whether or not they are able to work across the language barrier.
Preparing for the new virtual colleague
Proceed step by step and first identify the most time-consuming, yet simple processes: Where can I employ machine learning software most simply to achieve the most? The Pareto principle will help us here: Achieving 80 per cent by merely using 20 per cent.
In-house counsels will need expertise in hardware, software, and at least one programming language for training and employing virtual colleagues as well as for resolving legal issues regarding their company’s new digital business models.
More than 600 lawyers and tax advisers in Germany, 65 branch offices all over the world in 38 countries. Upon turn of the years 2016/ 2017, CMS signed an agreement on the use of the machine learning software Kira.
Photos: © iStockphoto/Devrimb