A case for using AI to fix judicial delays: Experts discuss pros, cons

The Kerala High Court recently mandated that, starting November 1, witness depositions in state district courts will use Adalat.AI, a speech-to-text translation artificial intelligence tool. In a vast and complex legal system beset with delays, the decision is a pioneering use of technology to reduce procedural challenges.

 

In most courts, witness statements are recorded in writing — either by the presiding judge or a designated court typist working on the judge’s instructions. 

“Automating the speech-to-text process will free judicial time, enable quicker listings, and shorten case calendars, as long as accuracy and workflow integration are maintained. For lawyers, the immediate practical effect will be a shift of effort from mechanical notetaking to analysis, strategy, and rapid verification,” said Suvigya Awasthy, a partner at law firm PSL Advocates and Solicitors. 

The AI tool is also likely to help judges, lawyers and court administrative staff by speeding up preparation between hearings, making searches for previous statements easier, improving cross-referencing of contradictions, and reducing long-term costs, Awasthy said. 

Witness protection 

There are other challenges in the witness and statement-recording process in India. The lack of a proper witness protection programme adds to the complexity involved in the deposition process. In a detailed judgment in 2022, a two-judge bench of now-retired Supreme Court Justices A K Sikri and A Abdul Nazeer had said that the condition of witnesses in the Indian legal system was “pathetic”. 

“Many times they are made to appear long after the incident of the alleged crime, which significantly hampers their ability to recall necessary details at the time of actual crime. They are not even suitably remunerated for the loss of time and the expenditure towards conveyance,” the judgment said. 

Other reasons for delays, particularly in witness deposition, include the intimidating nature and appearance of courtrooms, repeated summoning of witnesses to courts, and improper treatment by the court’s administrative staff, according to the judgment. 

That’s not the end of problems. In the same judgment, the Supreme Court noted that a major main reason for witnesses turning hostile is intimidation by the accused person or persons in criminal cases. 

Though Adalat.AI promises to reduce such risks by automating the recording and translation of witness statements, experts believe diligent care is needed in the tool’s use. 

“Lawyers will need to vigilantly review the output for mistakes, omissions, or distortions, especially in critical passages. Witness testimonies may include sensitive or privileged content. Ensuring that the AI system, Cloud storage, data transmission, and access controls maintain confidentiality is crucial,” said Shiv Sapra, a partner at Kochhar & Co, another law firm. 

An AI system is also unlikely to capture the depth of emotions of a witness the way a human stenographer can, Sapra said.

Transcripts must be accurate as court statements are often made in local dialects, Awasthy said. “Accuracy remains critical, as speech-to-text systems struggle with overlapping voices, accents, legal jargon, and multilingual testimony, making rigorous testing essential. Procedural clarity is needed to define whether AI transcripts are official records and to provide standardised correction mechanisms,” he said. 

Transcript accuracy 

The success of shifting to an AI-based speech-to-text translation system will also depend on transparency, so that lawyers and litigants do not face a system whose errors they cannot review or remediate, Awasthy said. 

“It is precisely for this reason that before the transcript is entered in the official court records, it must be reviewed and corrected. Lawyers and witnesses alike will need to be conscious of and adapt to the new environment where their every word, stumble, or whisper is recorded,” said Indranil D Deshmukh, partner and the head of disputes practice at law firm Cyril Amarchand Mangaldas. 

As AI systems require massive amounts of energy, courts will need to ensure that their underlying networks match the demands of the tools deployed, he said. 

“District court infrastructure in many states is underfunded and needs an overhaul. Once the IT infrastructure is of the required standard, only then technology like live transcription can be safely implemented," Deshmukh said. 

A new AI tool for witness deposition introduces additional workflow burdens, such as requiring each transcript to pass a verification check, Awasthy said. 

“The automated record may not be flawless. Transcription errors, misheard words, poor punctuation, dropped clauses, and misattribution of speakers will fall to lawyers to identify. If these errors remain in the official record, they can impact cross-examination, misstate admissions, or be used on appeal,” he said. 

Parties in a case may challenge whether an AI-generated transcript is reliable. The transcription can be considered a faithful record if AI’s “black box” nature is open to cross-examination,” Sapra said. 

“There may be procedural appeals or objections about the use of AI transcription as orthodox depositions. Precedents will need to develop,” he said. 

The promise of AI in judicial processes must be balanced with robust safeguards, said Kazim Rizvi, founder of The Dialogue, a tech-policy advocacy body. 

“Witness statements often contain deeply sensitive data that must be handled with the highest degree of confidentiality. Human oversight and strong cybersecurity measures are therefore indispensable,” Rizvi said.