Why LUCI

Imagine having a personal lawyer, ready and available to read any amount of legal material for you, to answer any research question you have. Imagine that this lawyer is available 24x7, and most importantly, can read more than 1000x faster than a human lawyer. Thats LUCI.

Say you have a research query, "what is an authors copyright over cinematographic use of his work?"

Pre-LUCI, the process would look something like this:

  • Look up all applicable acts, especially copyright act.
  • Look for provisions which mention anything that includes authors work (such as literary work)
  • Look for provisions which also speak about cinematographic rights
  • Look for cases harmonising these provisions

Efficiently done, this process would still take at least 30 minutes. In contrast, LUCI provides you with the exact, correct answer in under 5 seconds:

LUCI is a legal research tool by lawyers, for lawyers. We can break down the LUCI AI Question Answering process into three important concepts.

How it Works

Real Understanding

Attention Models are the latest breakthrough in Natural Language Understanding, invented in late 2018. Very simply, such a model computes an attention value to every surrounding word for each word it is interpreting. By computing such attention gradients for words during large scale "pre-training" on vast corpuses of english text (>1 Billion words), the AI refines its understanding of the contextual meaning of every word, phrase, and sentence.

Beyond Keywords

Even when the ideal answer contains few or none of the keywords mentioned in your query, LUCI can identify them, unlike keyword based search engines. LUCI does not limit your search by merely counting the keywords in your query. Emphasizing keyword matches is an outdated and inferior way to rank search results because it often misses the point of your question.

For example, the answer below does not have the words "laid down", and the case name is also not an exact match, yet LUCI automatically found the answer below, which completely and accurately answers the users question:

LUCI is specially trained on legal documents to intelligently direct its attention to relevant parts of the paras it is reading, like a human. LUCI intelligently finds the entire position of law applicable to your query, where possible.


Inverted Reading

Normally, after entering a query, you are required to read through a number of full length search results to find what you’re looking for. Things become even more complicated if the answer is scattered across multiple different documents or sources.

LUCI inverts the whole process by reading all these documents for you in seconds and differentiating between sentences which are merely relevant to your query and sentences that actually answer your query.

Your research will now begin starting with the exact answer and reading as much context around it as you need, rather than starting by irrelevant contexts looking for a relevant answer.

Your research will now begin starting with the exact answer and reading as much context around it as you need, rather than starting by irrelevant contexts looking for a relevant answer.

Judgements and Legislations used intelligently

Where one provision from one act or case can answer your question, LUCI will intelligently find it:

In cases where multiple provisions influence the answer, not only does LUCI find all the applicable provisions; where possible, it looks for a judgement to harmonise the two:


Constant Learning

The Feedback Loop

LUCI’s proprietary reading technique instantly finds paras that keyword search platforms miss. But that’s not enough. We designed LUCI to constantly keep learning from user feedback and growing its knowledge base, so that it never makes the same mistake twice.

The feedback loop has three steps:

  • User Query + Feedback
  • Quality Control
  • Retraining

At the first step, we ask our users to select the answer they find most relevant whenever they have a query. Through a simple, user-friendly tick, users can indicate which answer they found most useful, completing a feedback loop that allows the machine to iteratively learn.

At the second step, each users' data is subject to rigourous quality checking to ensure our model only learns from legally correct data. After quality control, the additional data is fed into the model every night.

The end result of this process is not only has the LUCI learnt the answer to the said question (and most sematic variations) once and for all, it has also stregenthened the pattern of reasoning it took to arrive at that answer. Through multiple cycles of such iteration we expect LUCI to develop a legal intuition matching that of the very best lawyers.

We believe a dynamic, organic learning process from real-world users is the most democratic and powerful way to teach an AI legal reasoning.

This also means that if LUCI makes a mistake, your best bet is to give it feedback and ask again the next day! Chances are LUCI would have learnt, and thus it would never make the same mistake with anyone else.

For example in the question below, while answer #1 is correct, answer #2 is more comprehensive and thus better. All users shall have the option of upvoting answers to continually refine the results.

Why Machine Learning?

Supervised learning is a machine learning technique that uses examples to teach a computer through a process called training. During the training process, LUCI finds the patterns in the training data that yield the best answers.

LUCI leverages advanced attention models to read the same paras with a differential focus depending on the query it is trying to answer - much like a human would. With usage, it continually refinforces the neural connections that lead to the best answers.