AI presents significant implications within the legal industry, according to technologist Dr. Khalid Al-Kofahi from Thomson Reuters. We look at how AI will shape the practice of law in the future.
Artificial intelligence is shaking-up the legal industry. This includes some breakthrough AI technology from Thomson Reuters, which will affect how attorneys work with their clients. Examples include changing billable hours, competitive advantages to building cases, and so on.
To find out more about the new AI technology and to understand how new technologies are disrupting law in general, Digital Journal caught up with Dr. Khalid Al-Kofahi from Thomson Reuters.
Digital Journal: How has artificial intelligence advanced in the legal space in recent years?
Dr. Khalid Al-Kofahi: Contrary to common belief, the legal industry has been an early adopter of AI technologies – in particular, natural language processing and machine learning. My own personal journey started 23 years ago (1995) and my first project was to build a system that processed case law documents and extracted court decisions for the matter at hand, as well as any commentary it made regarding other cases outside of its appellate chain.
That was a typical, albeit more ambitious, example of early applications of AI, where the focus was largely on introducing automation and machine-assist to editorial tasks, such as classification, entity extraction and resolution, etc. Then we started to see more AI in areas like legal research, document review, contract analytics, and so on. Going forward, I think AI will continue to drive much of the innovation in legal research – for example question answering technology and document analysis – as well as more advanced forms of contract analysis and management. In the 1990s we saw a mix of statistical and rule-based functions doing certain tasks.
In the 2000s, the industry largely moved to statistical machine learning. At present, and as we are trying to build more ambitious applications like robust natural language question answering, we see ourselves combining advanced machine learnings, including deep learning, with rule-based applications to, for example, capture domain specific knowledge.
DJ: How ‘intelligent’ is AI in the legal industry?
Al-Kofahi: Not to start a philosophical discussion on what ‘intelligent’ means, but whether we call something intelligent or not depends on whether we understand it or not. For example, consider a person who can read a ‘phonebook’ of 1 million entries, including names, addresses and phone numbers, once and then, given a name, she can retrieve the associated address and phone number from memory. We call that person a genius.
If I achieved the same task with a program, we call it a database. We understand how a program does the task, but not the person. This is why for any task, once you are able to formally describe it, it is no longer intelligent. It is just a program. So as someone who has a better-than-average understanding of AI, I don’t think AI is intelligent – not just in the legal industry, but in general.
DJ: What developments with AI do you think are most important for society?
Al-Kofahi: I lived before the age of Internet, when a simple question like “who invented the internal combustion engine?” would require a trip to the local library and a few hours sifting through books to find the answer. AI and machine learning have already transformed and significantly simplified how we travel, how we stay connected, how we work, how we shop and how to stay curious. All of these are extremely important to society, not to mention its application in healthcare and pharmaceutical industry. I don’t have a personal favorite, but I am especially fond of AI applications that aim to democratize access to knowledge and healthcare.
DJ: How is AI impacting on the legal profession?
Al-Kofahi:In knowledge work, the desired impact is often getting more work done in less time, while maintaining or improving quality. I think this is exactly what AI has been able to achieve for the legal profession. Just imagine if we didn’t have modern legal citators, the simple task of figuring out whether a decision was still ‘good law’ would be daunting. Or imagine the amount of work a large-scale document review would require.
When we released Westlaw Next in 2010, we did a study where we asked attorneys to complete a series of tasks with Westlaw Classic and Westlaw Next, and we measured an average of about 30 percent improvement with Westlaw Nest. Significant opportunities remain. For example, in 2014 the ABA released a study where they concluded that about 80 percent of demand for legal services in civil matters in the US goes unmet. That is a significant access to justice problem and, while it is not a problem that can be solved by technology alone, I think it could benefit from broader and different adaptation of AI.
DJ: Are there any possibilities for errors?
Al-Kofahi:Yes. All machine learning solutions make mistakes, hence the term error-rate. But not all tasks are created equal and not all errors are created equal. For example, for critical tasks, parole decisions for instance, the ‘machine’ should be viewed as simply making recommendations based on transparent criteria, while the human is responsible for making decisions. For other tasks – for example, a search engine – the researcher is responsible for finding relevant material, so all the machine is doing is reducing the amount of leg work.
In the latter case, a higher error-rate, while not desired, is tolerated. The other dimension is that some errors are ‘understandable’ while others can be embarrassingly wrong. A solution designer should minimize – as much as possible – embarrassing errors because they cause users to distrust the system.
DJ: What types of AI technology has Thomson Reuters developed?
Al-Kofahi:Consider what we do at Thomson Reuters, from a data and content perspective, across our industries. First, we author and collect content from variety of sources, and we enhance, normalize, connect and mine this content. Second, we make it available through push, pull and navigate, and third, we build vertical applications designed for specific use cases. Over the years, we developed AI and machine learning capabilities across this entire workflow.
For example, on the publishing side, some of the technologies include large-scale text-classification solutions, named entity extraction and resolution, web-scale concordance, document summarization, relation extraction, polarity detection, event extraction, language generation and so on. On the content delivery side – we started with horizontal search engines, in fact we developed the first commercially available search engine that deploys probabilistic rank retrieval back in 1993. After horizontal search engines, we moved to vertical search engines – these are engines that are tuned to the use cases and meta-data of the domain. We primarily focused on learning-to rank-approaches.
Then we designed a number of recommender systems, and most recently we developed domain-specific but open-ended natural language question answering capabilities. Finally, on the vertical applications front, we created numerous AI-enabled applications including those for document drafting; abnormality and risk mining, including reputational, financial and physical risks; due diligence for “know-your-customers” applications; e-discovery; debunking rumors on social media; and so on. From a technology perspective, we are heavy on Natural Language Processing and machine learning, both supervised and unsupervised, including deep learning.
DJ: How does this technology differ from your competitors?
Al-Kofahi:I don’t have deep insights on the technologies that are used by our competitors – but successful applications in our verticals require three ingredients: comprehensive, accurate and enhanced content, subject matter expertise to ensure we build the right solutions, and technology, including how to apply AI on the nuances of the domain. Some of the new and upcoming competitors don’t have all three ingredients and I think they will find it challenging to succeed at scale. As for more established competitors – I believe we have a stronger mix of the above three ingredients and this is why we are leading the industry in AI-enabled applications.
DJ: Where can AI for the legal sector go next?
Al-Kofahi:These questions are difficult because humans use a linear model that relies heavily on existing challenges and opportunities as the basis for our predictions. Often, this is not the case in reality. With that in mind, I expect to see more utilization of AI in the consumer space, such as online dispute resolution systems, in jurisdictions that allows it. Transactional law is another area that I expect will be transformed through more clever applications of AI – not just in contract review and analytics. On the content side – I think AI will allow publishers to create ‘smart’ analytical content, that is tailor-made for particular use cases and users.
Within 5-7 years, I expect to see a personal digital assistant for lawyers, so a personalized suite of applications that have a sense of what an attorney does and how she likes to do things, and that are able to ‘remember’ one’s past and experiences and proactively bring them into the right context. For this to happen, one would need to overcome several challenges, including, for instance, who owns attorney experiences, the attorney or her employer?
The current prevailing wisdom seems to distinguish between physical work-product, which would be owned by the firm, and experiences, that we keep in our heads. But for an assistant to work, these experiences would need to be encoded in some digital form. This would make them a form of work-product, which should result in very interesting debates at some point in the future.
This article originally appeared in Digital Journal