Valuation of Disputes as an Alternative Asset using Artificial Intelligence: No Human Intelligence
Written by No Human Intelligence (Daria Godorozha)
No Human Intelligence is a comprehensive legal analytics pipeline designed to estimate and evaluate litigation versus settlement outcomes. It integrates data collection from sources such as PDFs, dockets, and spend ledgers, with capacity for web/database scraping and synthetic data generation. For text processing, it utilises transformer-based models including Facebook's BART-large-mnli for text annotation and T5 for LLM callibration. It incorporates LightGBM classifiers calibrated with scikit-learn's Calibrated Classifier CV for predicting litigation outcomes. The pipeline also includes policy parsing with regular expressions, sophisticated feature engineering and risk assessment simulations employing Monte Carlo methods. It outputs key financial metrics such as average returns, variability (standard deviation), the probability of superior litigation outcomes, and discounted present values to guide informed decision-making.
In the past I was an intern at Stanford at the Center for the Study of Language and Information and also taught as part of the Stanford SPLASH programme. Being located in the heart of Silicon Valley, Stanford Law school has a significant technological focus, so when I heard about CODEX from Stanford law coming to the UK, I was obviously excited. I also attended the King’s College Cambridge Entrepreneurship Ball a year ago and was familiar with all of the amazing initiatives that they do. Given the high impact of the organisations involved, the Hackathon seemed like a great opportunity. And this assumption proved true.
Much of my background involves working with alternative assets, I have worked in maritime as well as in fine art and I have also been a part of more traditional financial institutions such as the Bank of England and KPMG. Given this, when a challenge came to create a settlement/litigation valuation tool, I immediately seized the opportunity. I approached the Brown Rudnick team, and they were similarly enthusiastic about the project. It was amazing to have such a high level of support, and to work with such passionate people. I'm so grateful to the partner Stephen Palley for his mentorship.
One thing that stood out for me was that there were so many mentors at the event; from leading universities like our hosts Cambridge and their partners, Standford, to mentors from tech companies, such as Google and Groq, and then to those from top law firms. Everyone was very approachable and extremely helpful. One of the best parts of the Hackathon, then, was not just learning about all the new technologies and frameworks but doing so in an environment of engaged and encouraging sponsors, mentors and other students.
Another highlight was when one of the founders of Arm, Herman Hauser, discussed the limitations of current AI processing hardware and so LLM’s, the future potential developments within the hardware space, such as with quantum computing, and also potential new avenues in LLM research with neuro symbolic AI as contrasted with vector based statistical AI.
Once I arrived at the beautiful Judge Business School in Cambridge, we were immediately given an opportunity to attend some of the sponsor talks, followed by lunch and then by the keynote from Herman Hauser, as well as team formation drinks. I was grateful to have had my hotel sponsored by Clifford Chance so that I could attend all three days of the conference. The next day everyone arrived and began to code, this was extremely exciting as we were able to make our ideas come to life! There was so much enthusiasm as everyone was able to use their unique skill set to design a product that would make a difference in the real world.
Litigation is an asset that is investable and sellable. My tool calculates settlement values as well as litigation values and the potential risks in choosing one over the other. While legal data is often readily available it comes from disparate sources and in disparate formats. My tool is able to retrieve data from multiple sources, harmonise it and feed it into two separate pipelines. One that is numeric and forms the primary calculation (involving calculations of previous settlement values as well as judge win rates, opposing counsel win rates and spending amounts) and the second is linguistic and which recalibrates the numerical data using specifically tailored contractual clause extraction tools as well as analysing case briefs and judge sentiment/opinions.
From a technological standpoint it utilises a rigorous pipeline including RAG via FAISS, a policy parser, feature engineering that pre-processes data prior to prediction, a rigorous financial mathematics algorithm that also runs Monte Carlo simulations and, following from that, utilises a T5 LLM for calibration with dynamic updating of the pricing model based on live data. The T5 algorithm utilises encoder-decoder transformers which are able to process both linguistic and numerical data. This can be underpinned by a knowledge graph.
This Hackathon, however, wasn’t only about building products, it was also about inspiring the next generation of lawyers to become literate and entrepreneurial in regards to technology and about encouraging engineers to pursue entrepreneurship in the legal realm. Over the weekend, I met an amazing group of law students and students with business majors, who had never coded before but were able to build an agentic model with a professional front interface within several hours. They told me that they had become inspired to learn how to code longer term.
I am used to doing consulting work and working on various projects, but I have never considered entrepreneurship seriously. Getting to the finals and winning third place, while also winning the sponsor challenge, really inspired me and I’m so grateful for being able to participate. As a top 3 winner I also get to pitch my idea at Legal Tech week which is extremely scary as going out of one's comfort zone always is; however, progress is built on taking steps you never thought were possible and the E-Lab Hackathon is all about inspiring students to do that. Thank you so much to the organisers, particularly Zarja, for all of her hard work and for inspiring next generation talent in the field of legal technology.
Daria Godorozha is graduating in July from her LLM in Law at King’s College London and is an MSc AI and Data Science Student at Queen Mary University of London. In the past she has interned on projects with Stanford CSLI, KPMG, the UN, the Fine Art Group and the Bank of England. She is most interested in symbolic systems as well as the intersection between cognition, mathematics, logic and language.