“I started working on legal AI three years ago because we understood there is a gap here,” says Dr Ghosh during his conversation with INDIAai’s Jibu Elias. Dr Saptarshi Ghosh is an Assistant Professor at the Indian Institute of Technology Kharagpur.
“Most of the work in this field is done in Europe, China or US, but to my knowledge, very few people in India work on it. I had one distinct advantage, though, which is that IIT KGP is possibly the only top engineering Institute with a law school inside. Interestingly, it’s just a five-minutes walk from my Computer Science department,” he says.
Generally speaking, the application of AI to any particular field requires concerted effort from AI experts and SMEs. “When working in these interdisciplinary topics, I usually think that most problems come from the domain experts. They tell us what is interesting, what is challenging and so on. So we have been talking with the law school’s senior students and faculty members over the last few years, and they suggested some problems.”
Navigating the complex legal landscape with algorithms
He relates to us some of the pertinent problems in the legal spehere that he has been designing AI solutions for.
“One of the problems was this: in India, the bigger case documents are really long – often they can go up to hundreds of pages and unlike in other countries, they are not written in a very organised way. If you look at, say a document from the United Kingdom or some other country, they have nice paragraph headings which make reading them easier. However, in India, it’s just a plain text – paragraph after paragraph of text. So even understanding which part is what is an issue.”
Now the legal practitioners don’t want to read the whole case document. The only parts of relevance and interest are the issues discussed, the facts of the case and the judgement. That’s where the automated reading of case docscame in.
Another project that Dr. Ghosh’s team is working on is the automatic summarisation of legal documentss. “There are lots of techniques already about legal documents summarisation, but when we applied them to Indian documents, we found that that do not work so well.” This is because of a lack of structure in these documents. This work has recently been published at another conference.
“Let’s say we take a summary written by an expert and the summary generated by an algorithm, and then we see how closely they match. That gives a score in the range of 0-1. One means exact match and zero means no match at all. The most efficient system would give the score as 1 but we are far from state-of-the-art as of now. The methods that we have tried give the score of the range of 0.5-0.6.”
Another problem pertains to finding related cases. “When a lawyer has to argue one case, he or she would like to find out previous cases which were similar to this. So how to find out similar cases?”
All this work on legal AI is done in collaboration with the Rajiv Gandhi School of Intellectual Property Law. “Our algorithms will tell something, but there is someone who needs to judge whether what they are saying is correct. Or maybe the summary that the algorithm is giving – how good it. So we actually show it to the experts and they tell us how to improve, what is good, what is bad and so on.”
How much is too much?
The Indian judicial landscape presents some very interesting problems to solve using AI. Pendency is a big issue in the Indian legal system. “Can AI be used to handle pendency of cases in the judiciary? We are thinking of quickly starting something on this.”
But the situation it poses a complex challenge. “Even how to tackle it using AI is an issue because law practitioners don’t like AI to intervene at every stage, which is quite understandable. Would we like to hear an algorithm is judging us? Maybe not.”
But there are some undisputable benefits that AI can bring to the legal field. “One advantage is speed. The things take a week for a human to do may be done by AI five to ten minutes. So if the more mundane jobs can be pushed to AI, then the law experts can actually focus on legal questions.”
“So briefly, what I have understood is that the law practitioners are fine with AI helping them, like summarising a document, making it easy to read, translating it from one language to another etc. But they’re still not okay with AI trying to replace them, which is perfectly fine. For example, there is the problem of legal judgment prediction. It refers to actually trying to predict the judgment of a case. Personally, I have not worked on that and I don’t intend to work on that right now because how much acceptable it will be is an issue. In legal AI, there are a lot of ethical issues.”
Some of the hesitancy surrounding legal AI may be tackled by bringing in the element of explainability. “Can an AI explain why it took a decision or why it said that someone’s bail should be denied? There is a lot of research going on about explainability. The critical question is what kind of explanations would make sense to a law practitioner. Because the explanation that would make sense to an AI person may not make sense to a law person. That is something which we need to figure out. If AI model can give those kinds of explanations, maybe it becomes a bit more acceptable.” So these may be some ways of making AI more acceptable in the legal sphere.
Source: indiaai.gov.in
