KnowYourRights: The Hackathon experience
Written by KnowYourRights
KnowYourRights aims to educate ordinary citizens about their rights under consumer protection law. It aims to correct the imbalance between consumers on the one hand, and traders on the other hand.
Imagine you buy a dodgy IPhone charger. You want to return it. The store owner denies your request on the basis of “company policy” – he claims that electronic goods cannot be returned. If you're not a lawyer, chances are, you will accept this as the truth. Even if you are a lawyer, you still might not know what you are entitled to. It’s five pounds. It’s not worth engaging a solicitor. You grumble. And you leave.
Company policies or store polices are frequently invoked to obscure protected legal rights, yet it remains virtually impossible for a “non-lawyer” to navigate the Consumer Rights Act 2015, a lengthy 194-page document. To work to correct this, we developed an application that responds to the problem: KnowYourRights. Our team comprised three Machine Learning MPhil students, and one Masters of Law student.
Our application seeks to improve legal education among ordinary citizens concerning the rights protected under the Consumer Rights Act. The application works by first requiring the user to input information. A summary is then generated and parsed through a safety pre-processor. At this point, implicit biases are filtered out. We found, for example, that if the user used emotive language, then they were 7.5% less likely to receive a refund. The cleaned information is then subject to a tree traversal process. This compartmentalises the legal reasoning process into multiple discrete steps, instead of the LLM “one-shotting it”. Initial headers are scanned for relevance and then the agent drills into particular Parts of the legislation to identify relevant provisions. It jumps out of the Part if need be. This tree traversal ensures that the agent accurately mimics how a human lawyer actually reasons. Human lawyers do not read legislation sequentially, section by section, but chaotically jump around a piece of legislation, and only focus on single provisions if necessary. There are two outputs from the tree traversal process. The first is a practical legal solution for the customer (e.g., “You are entitled to a refund”). The second is a history logbook which accurately and meticulously records each step taken by the agent.
Reflecting more broadly on the Hackathon, we believe the key take-away has been the immense importance of inter-disciplinary collaboration. Three examples instantiate this point.
The first was in a lab-room. It was hot. We were tired. The STEM team-members kept asking me, the law student, to write down the steps of how I would read the Consumer Rights Act. I tried. I didn’t really know what the steps were. I could not explain the process of legislative interpretation as a formulaic series of steps that you follow in every case. Then another member of the team asked me to pull up the legislation online. We projected it onto the big screen. He asked me to show him how I would reason. How would I actually think through a problem? As I jumped around the legislation, this represented the first break-through in our product’s development: the process of reading legislation is chaotic, and this chaos could only be captured through a tree traversal. The law should not simply be reduced to a flowchart, or a series of binary questions. This insight was only achieved through genuine collaboration, and the process of tapping into a lawyer’s mind.
The second example comes for our experience with pitch preparation. As the dedicated pitcher, I presented to the team for feedback. This was incredibly informative. The first issue I had was that, in the process of becoming a lawyer, I had begun to speak with unnecessarily complex words. This had become engrained into my style of public speaking. It was a disease! Another difficulty I had was in explaining the solutions stage. I robotically explained the mechanics of the technology without succinctly explaining the ‘highlights’ of our product. We switched to a Q&A style in the presentation, where I asked and answered the questions that a hypothetical audience member might have. This went a long way in ensuring our pitch was delivered effectively.
The third example was in the preparation of the slide deck. In my legal career, I had actively and successfully avoided putting together pretty power-points. I didn’t really see the point. I naively said to our team, “why don’t we just use a white background with some black dot-points? We can put it in Times New Roman if you really want”. I just couldn’t be bothered. A computer scientist on our team, who had a background in user experience, politely informed me that I was being silly. He insisted that I used “Figma” to create the presentation. I immediately panicked – something new to learn at 8pm! I had never heard of this before. But I realised that it was very user-friendly. And there was virtually nothing to learn at all. This was a learning moment for me – the packaging does matter, and it’s not always going to be difficult to achieve.
At each of these moments, we all learnt something from one another: collaboration is absolutely essential. And, not only that but, it is genuinely so fun to learn new things from a completely different perspective. The hackathon felt a bit like an optical illusion, the kind where you see the ballerina spinning one way, and then the other, or you think you’re looking at a rabbit and then suddenly it’s a duck. Without collaboration, you never really get the opportunity to see all the possibilities with clarity.
Laksshini Sundaramoorthy is a Masters student in Law at the University of Cambridge. Prior to this, she practised as a lawyer at the Supreme Court of New South Wales and at Allens, a top-tier commercial law firm. She is particularly interested in private law, which governs how individuals interact with one another.
James Xu is an MPhil Machine Learning student and Part III Mathematics graduate from Trinity College, Cambridge, leveraging advanced mathematical and statistical techniques to address real-world challenges. With experience as a Machine Learning Engineer at multiple startups, he specialises in using theory-inspired probabilistic methods for interpretative AI algorithms to drive innovation.
Hidde Heijnen is an experienced developer and aspiring entrepreneur who, during his computer science bachelor, ran a successful development agency. He is now doing an MPhil in Machine Learning at Cambridge.
Leonhard Vulpius is a Masters Student in Machine Learning and Machine Intelligence at the University of Cambridge where he also completed his first Masters in Theoretical Physics. He further holds two Bachelors degrees in Mathematics and Physics from the University of Göttingen.