Using artificial intelligence more effectively

Despite its initial promise, AI solutions often fall short in the Indian legal context due to training on non-local data. A hybrid human-AI approach could build more responsive and effective systems.

This article was first published in The Mint. You can read the original at this link.

In 1770, long before computers were invented, Wolfgang von Kempelen built what was widely believed to be the world’s first intelligent machine—a device so smart that it consistently beat some of the best minds of the day at chess. This human sized chess-playing automaton was dressed up to look like a turbaned Turk (because it was believed that chess originated in Turkey) and could physically move the pieces around the board in response to moves made by its opponent.

The Turk was such a success that after its debut at the Schonbrunn Palace in Austria, Kempelen was forced to take it on tour across Europe playing a number of distinguished opponents including Francios-Andre Philidor (the best chess player of the time) Napoleon Bonaparte and Benjamin Franklin. While it did not win every match, the Turk’s performance was so creditable that according to Philidor’s son, it was the most exhausting game of chess his father had ever played.

In 1854, the Turk was destroyed in a fire, taking its secrets with it. It was only much, much later that the contraption was revealed to be an elaborate hoax. Instead of a steampunk computational machine far ahead of its time, Kempelen had built a device that could conceal within its machinery, a human being who observed the moves of the opponent from within the Turk and responded by manipulating the articulated arm of the automaton from inside to move his pieces.

Today’s artificial intelligence is vastly different from what Kempelen was peddling. It has become a part of our life and influences much of what we do. So much so that everyone, no matter what their product or service, is trying to spice up their offerings with a dash of machine learning.

The legal industry has not been spared. I am constantly being offered products that promise insights beyond the ability of ordinary human attorneys to discern with a level of attention to detail that overworked associates cannot be expected to deliver. Some services can pore through the contents of virtual data rooms and reliably extract from all documents and records of an acquisition target, those issues that represent significant risks to the acquirer. Others claim to be able to generate complex contracts to suit the obscure requirements of clients with far more accuracy than a human. Still others claim to be able to sort through hundreds of thousands of precedents and come up with arguments that can be used in court.

Having tried out a number of these services, I have to say my experience has been mixed at best. Artificial intelligence is only as good as the data sets it is trained on and, since most AI products for the Indian legal market have been trained on US and European data, they are less than impressive when it comes to ferreting out Indian legal issues. The risks and precedents that they highlight are often irrelevant in the Indian context and they are frequently unable to identify risks that are unique to the Indian regulatory landscape. To make matters worse, law firms have to deploy these programmes in high-stakes scenarios where misidentification of risk can have dire consequences. Yet, despite these concerns, clients still insist that their law firms should use AI wherever possible to reduce costs.

Most law firms are struggling to balance these concerns. Having made the not inconsiderable investment in AI software, they had expected to recoup their costs by lowering headcount. That now seems unrealistic and unless they want to write off their investments in AI, they will need to adopt to a hybrid approach.

This realisation is by no means unique to the law. Many other industries have reverted to using humans to perform some or all of the services they had previously entrusted to machines. Many automated voice response businesses have begun as an alternative to using voice recognition algorithms to parse human queries and find an appropriate response, to outsource some these services to call centre workers, asking them to keep their responses neutral and voices robotic in line with customer expectations. Companies that use algorithms to process medical scans to determine whether a tissue or blood sample is malignant or not, are getting teams of radiologists to review the output to ensure that the diagnosis that the software came up with is not inaccurate.

Perhaps the most successful demonstration of the power of this hybrid approach can be found in a service that is itself named after Kempelen’s 18th Century invention. The Mechanical Turk is a service curated by Amazon that uses human cognition to perform tasks that AI is unable to perform. It offers a structured platform through which human intelligence can be harnessed and made to fill the gaps that AI cannot. With remarkable success.

Law firms should not be shy to commit to a hybrid approach. AI may not be the magical solution that they were promised but if it can be supervised, its choices studied and corrected so that it learns from its mistakes, it should be possible to build better, more responsive AI systems. This is a costly exercise but, going by the success that other industries have had, one that will eventually bear rich dividends. We will need to educate our clients about this process, encouraging them to invest along with us in training these AI systems — and being patient until we can all see the results.

If we don’t lose faith in its promise, I have no doubt AI will eventually deliver.