The Future of Legal AI - Part 2
This is part 2 of a conversation Tower of Babel had with Horace Wu, Founder and CEO of Syntheia, a new legal AI tool, after a podcast we recorded together over the summer. Find part one of this interview here.
So, we’ve talked a bit about this before – but can you elaborate on the use cases for these newer AI tools in legal? What does Syntheia do, for example?
We expect to see a lot of interesting new technology in the next few years, where people are going to try different ways to improve the work done in the legal profession.
On the podcast, we talked a little about evolutionary changes and revolutionary changes. One of the ways we consider technological change is by categorizing them into one of two buckets of improvements:
Evolutionary changes, where tools we have today will become better, more accurate and faster; and
Revolutionary changes, where things we couldn’t do before become possible.
Evolutionary changes
I’ll address evolutionary changes first. It may seem like the less exciting category on paper, but we have to be conscious that the word “evolution” should not lead us to underestimate the scale and pace of these changes.
Take the use case of document review for due diligence, which has become quite competitive in the last decade. Historically, vendors of due diligence software offered their tools in two parts.
one part comes with pre-trained models that are capable of extracting text from legal documents. Leases are a good example, where the vendor’s team would train their machine learning models on hundreds of examples, and buyers can use these pre-trained models to review the next thousand leases in order to identify the same types of information and extract them into a report; and
the other part of typical due diligence software allows lawyers to train their own models on their own documents. From what we have seen in the market, the current requirements to train a custom model would need anywhere from about 50 example documents to a few hundred examples, depending on the vendor.
Due to the technological constraints and the cost of training machine learning models, these due diligence tools require a centralized training system where the vendor would do the training or multiple firms would contribute their data to a shared model. These due diligence tools are well suited to high volume, low value legal documents.
With newer due diligence tools, we can train a machine learning model to perform with comparable accuracy with, maybe, 5 to 10 example documents. When you change the performance at the scale of a 10x improvement, even though it is an “evolutionary change”, you can fundamentally alter how AI-assisted due diligence is offered.
Concretely, in the past, it made sense to pool training data with an external party, but now, individual lawyers can train their own machine learning models in real-time:
Specialists in niche practices may only have a few example documents, and they can make use of the newer AI technology for document review.
Boutique firms who previously had to rely on vendor-trained machine learning models can configure their own models.
We would put improving user experience and better design of user interfaces into this bucket of evolutionary changes as well.
We think that these types of evolutionary change will bring measurable benefits to lawyers in the short term.
Revolutionary changes
In terms of revolutionary changes, that becomes a bit of a guessing game. I am sure our guesses won’t even cover a fraction of what is possible, but from what we have seen in the market and some of our own experiments, in the short to medium term, AI will understand more and be capable of doing more:
Recommending drafting changes in real time as you are typing, with recommendations that are not based on template precedents or rules;
Summarizing complex legal text like judgments into accurate and succinct reports, and automatically finding relevant primary and secondary sources;
Translating legal text between different languages so that lawyers from different jurisdictions can work on the same set of documents; and
Spotting anomalies in documents based on differences to market trends, and offering market intelligence.
What Syntheia does
Syntheia and Lateral, for example, provide tools that fit in both of those two buckets. Our document review tool is in the evolutionary bucket and enables lawyers to train a model for due diligence from 3 to 5 documents, and our document drafting assistant is in the revolutionary bucket and enables lawyers to get real-time recommendations for drafting contract clauses without needing any prior manual curation.
When it comes to deploying AI technology in practice, our philosophy is to provide tools that are highly adaptable and flexible. All of our applications are designed to solve specific needs our customers have identified.
Are you also seeing changes around how people are using AI and other advanced technologies?
There has been an overall maturing of the market in the last few years, and this has caused a few subtle shifts. For example, people have stopped talking about AI as if it is the core feature and benefit, and instead, people are treating AI as one of the tools for solving identified problems.
We expect to see more integrations with other tools, and the AI element of the software becoming increasingly user-friendly.
We expect to see five trends for AI in the legal profession:
1. First, AI becoming more integrated. Larger vendors are pushing platforms, and smaller vendors and startups are making their technology more connected and integrated. This shift towards more collaboration between vendors is enabling the combination of technologies to solve more complex and higher-valued problems.
2. Second, AI becoming more invisible. AI is taking on more of a supporting role. Your phone is a great example of this - those of you who have an iPhone or an Android have access to lots of AI-powered tools like speech recognition, text translation and facial recognition. Those tools are presented to us as part of what our phones can do, rather than marketed as “AI”. This is a shift to look at how to use AI to solve problems, rather than just offering technology solutions, and instead fitting software solutions into a workflow.
3. Third, software development becoming more iterative. AI software is becoming intertwined with design thinking and problem solving, and fitted to human needs. That is, vendors are starting to allow more customization of their software, and deploying multiple versions over time which are adapted to the specific needs of their customers.
4. Fourth, and we think this one is only just beginning, vendors providing more visibility of the functions in their technology. Vendors are starting to become more open with buyers when talking about how they use technology to solve problems, and showing the layers in how their technology works. Making APIs available to their customers is one of the indicators of this increased openness.
5. Fifth, AI becoming more “self-service”. This is tied to our earlier point about how there is an evolution of improvements where lawyers can train their own industry-specific machine learning models with fewer data points. With these evolutionary improvements, it becomes possible to decentralize components of the technology, allowing law firms and legal departments to be nimbler and self-sufficient.
The combination of those trends and putting the technology into the hands of lawyers who can directly influence the outcomes are helping to garner more trust in AI technology, so lawyers can feel assured that the results are reliable, fair and explainable.
How disruptive is this going to be to the current landscape?
We think the disruption will be to categories of tasks that are performed (but not automating away jobs).
In the short term, we think disruptions will continue to be slow and minor. COVID has accelerated technology adoption for lawyers, but it has also changed the buyer behavior of law firms and companies. We are seeing strong adoption of foundational technologies in the last few months - better communication software is one example of this.
In the medium to long term, we also think that disruption will be on two fronts - to the providers of legal services and to the consumers of legal services.
For consumers of legal services, they will have better access to some slices of legal services. Commoditized legal services, high-volume document review work, and relatively repetitive low-value work can be (at least partially) automated. Consumers will also have access to better chatbots and other user-friendly ways to get answers to their questions. We don’t think this will replace lawyers, but it will help triage the need for legal help.
For providers of legal services, like law firms, we think AI is going to change how law is practised and delivered. But, let’s qualify what we mean by “disruptive” - we think that AI won’t take away legal jobs, but AI will improve both the speed and quality of work. The transformations will let lawyers cover more ground more quickly, and catch mistakes they might have made.
We are already at the point where technology is available which will let lawyers:
Manage and make sense of large datasets of documents and information;
Train their own machine learning models;
Rapidly adjust and retrain machine learning models; and
Have transparency on how their machines are working.
These changes to the technology could change how law firms and their staff manage technology and its interactions with their proprietary and valuable organizational knowledge.
Are you suggesting that KM could become the curators of AI technology in this new world? Why would this be the case specifically in relation to AI?
Yes, depending on the organizational structure of the law firm or the company.
A key aspect of legal practice is using expert knowledge to solve client problems. Philosophically, better curation and application of knowledge can lead to improved results in the practice of law. With the way some law firms are structured today, the KM teams hold the position as the custodians of knowledge.
We view AI technology as a “force multiplier” for knowledge managers and lawyers to better capture and utilize legal knowledge. With the new generation of AI-enabled software, where you need fewer data points and you have greater transparency and control of the processes, it becomes possible for experts at law firms and companies, like knowledge managers, to play a central role in the curation and application of knowledge.
Do you have any tips for firms or corporate departments who want to invest in AI technology?
In the past, developers built monolithic software to solve a problem, and then they productionized the solution. In that framework, there is a hesitancy to do a major rebuild because of the costs involved.
In newer software, we can easily swap out components and upgrade when it makes sense. This is a world where technology improves every month. This field is moving so quickly that any technology you adopt should be extensible and upgradeable.
In the new AI world, we think there are some questions law firms and legal departments can ask to assess whether an AI solution is a good fit for them.
What problem or pain point is the software trying to solve? Is the AI software helping you, as the buyer, with a problem you want to solve?
How is the software solving your particular needs?
Does the vendor offer self-service trials or pilots? Do they only give guided demos and videos? How confident are they with letting you validate their software for yourself?
What does the vendor offer in terms of adoption and implementation? What efforts will the vendor make to understand your people and process?
How comfortable is the vendor with explaining their tech stack to you? How much impact can you have on the different layers of machine learning based on your needs?
How is the vendor of the software dealing with rapid advances in technology? Will they upgrade, how easy is it to upgrade, and how often will they upgrade?
Many thanks to Horace Wu of Syntheia and Martin Karlsson from Lateral.io for their insights.
