The AI Age of Justice
The journey of how innovative applications of AI has rapidly started to integrate itself into the justice space from November '22 to now
On November 20, 2022, OpenAI launched ChatGPT powered by its GPT-3 Large Language Model (LLM). This general-purpose AI model was trained on 175 billion parameters, a large chunk of all the knowledge that exists on the internet. In the passing days, it became apparent to one and all that we had witnessed a giant leap in AI capability.
Up until the launch of ChatGPT, approaches to legal and justice AI had primarily been about painstakingly developing independently trained models to do specific functions in law and justice— create a legal document, understand key components of a legal document, summarise information, answer questions, and so on. Suddenly the general capabilities of GPT-3 upended this paradigm; it showed capabilities to do all of these legal functions with considerable proficiency. One of GPT-3’s first real-world tests was when a group of professors at the University of Minnesota made it attempt the Bar Exam and graded it alongside actual students. GPT-3 passed the test, albeit in the bottom 10% with a C+ score.
Then in March 2023, less than two months later, GPT-4 was released. This time trained on a trillion parameters— essentially almost all of the human knowledge on the internet. This time when put through the Bar Exam, the new model passed in the top 10%, beating out 90% of people who took the exam. Other tests revealed out-of-the-box advanced capabilities directly relevant to law and justice. It was clear that we had arrived at a point when AI could partially or totally fulfil several essential legal and justice needs. Ironically, while the present may be shrouded in uncertainty – how will GPT be applied, what will the early products be, how much human substitution will happen - the future looks more certain than ever. AI will profoundly transform the field of law and justice.
All Hands on Deck
In the wake of the AI developments, leaders from the government and judiciary reiterated calls for AI to be integrated into our core systems of law and justice. On April 14, 2023, the Prime Minister of India sounded the bugle. At a celebratory function of the Guwahati High Court, the PM urged the courts to embrace AI and other innovations to facilitate ‘ease of justice’ for litigants. Even earlier, in January 2022, the Chief Justice of India, DY Chandrachud, at the inauguration of the ‘Justice Clock’ at the Gujarat High Court, urged the Government of Gujarat to adopt AI to improve its ability to predict property disputes before they happen and enable the judiciary to filter out frivolous litigation before they clog the courts. A month later, the Supreme Court had set an example by using an AI solution – TERES, built by a Bangalore-based startup, to transcribe hearings and publish the transcripts.
Innovators are usually closer to the ground and quicker to react than the functionaries of our institutions, and the impact of generative AI has been no exception. Shortly after the launch of GPT-3 in November 2022, around 280 innovators from across India gathered for the Agami Summit, an annual meeting of innovators in the Indian justice ecosystem, in the tranquil hills of Panchgani, Maharashtra. There Dr Vivek Raghavan, arguably India’s foremost AI scientist, the lead scientist at the OpenNyAI mission- co-founded by Agami, EkStep, NLSIU Bangalore and Thoughtworks, and former CTO of Aadhaar- demoed ‘Jugalbandi’, a WhatsApp based chat bot that combined the ChatGPT model, released two weeks earlier, with the Bhashini model— an opensource tool that can translate both text and voice into multiple Indian languages.
The model had been trained on government schemes data from myscheme.gov.in, a government website cataloguing all government schemes nationwide. In the demo, Vivek Raghavan pretended to be a farmer and asked a reasonably complex question about schemes in Hindi. His question was sent to the bot as a voice note. In a few seconds, the bot sends back a voice note, also in Hindi, with a precise answer to Vivek’s question. Not only was the information accurate, but the bot was also able to take a subsequent request to apply for the scheme on behalf of the farmer. The audience was eager to test the bot for themselves, attempting different languages and asking more complex questions. Parts of the demo revealed weaknesses and signs that it was still in its early days. Nevertheless, it was still remarkably efficient and accurate for a demo that was assembled in two weeks. The innovators could see for themselves that not just would AI solve the typical legal functions mentioned earlier, it would also enable any Indian to access critical legal information and guidance at their doorstep at no or low cost. AI could truly solve not just the justice problems but the ‘access’ to justice problems as a whole.
In January 2023, Satya Nadella attended a demo of the same ‘Jugalbandi’ bot. He was blown away not just by the power of ‘combinatory innovation’ - an innovation that’s born from combining multiple technologies, but he was shocked at the pace at which innovation in Silicon Valley had spread across the globe and landed in the hands of an Indian farmer. Over the next few weeks and months, Satya travelled the world, from Mumbai to the World Economic Forum and eventually back to Redmond at Microsoft HQ. At every stop, he recounts this India demo and its profound effect on him.
Everyone recognizes that technology has crossed a chasm. AI presents a huge opportunity for developing countries to leap-frog into the digital world and brings more equity, access and transparency across sectors for their citizens. To make this happen, we need radical collaboration and co-creation across business, society and government to ensure that the AI age can truly empower, and not just serve, every citizen.