The Machines Have Won: AI Lessons for E-Discovery from the World of Chess

Artificial intelligence isn’t new to the legal field. Machine learning applications like predictive coding have been around for years and are well established. But compared to chess, legal AI is very young indeed. So for legal professionals concerned about AI as a threat to humans, or uncertain of how best to use AI, the world of chess has a number of lessons to teach.

AI achieves superior results than humans.

Not at everything, of course, but AIs trained (or self-taught) specific tasks perform better than humans. In 1997, Deep Blue defeated Gary Kasparov, then world champion in chess, to mark a passing of the torch from humans to our silicon progeny. The gap between man and machine has only widened since, as there are numerous iPhone and Android chess apps (such as Stockfish) capable of beating the best human players in the world.

While no serious chess player would dispute that AI is simply better at chess than human players, many legal professionals still prefer “human eyes” to AI review of ESI. In a recent Exterro poll, 47% of respondents would rather use attorneys than AI to review documents for responsiveness. But that view is outdated, in a 2011 academic study on technology-assisted review (TAR), Maura Grossman and Gordon Cormack established that “technology-assisted processes, while indeed more efficient, can also yield results superior to those of exhaustive manual review, as measured by recall and precision.” (Emphasis added for purpose of clarity.)

Modern AI is radically different than its ancestors.

In 1997, when Deep Blue defeated Kasparov, it relied on both brute force, scanning hundreds of millions of moves per second, as well as programming and training by grandmaster-level chess experts. The processing power of AI algorithms today is orders of magnitude greater than it was twenty years ago. However, modern AIs learn in completely different ways. DeepMind’s AlphaZero AI learns to play games not based on training by human experts, but with the rules of the game itself as the only input. AlphaZero taught itself to play chess in under four hours, and then proceeded to beat the world’s leading chess AI Stockfish 8 in a 100-game match—without losing a single game while evaluating a “mere” 80,000 positions per second.

In e-discovery, older machine learning technologies like predictive coding evaluate documents for responsiveness based on human-coded seed sets. While the time- and cost-savings of predictive coding are compelling, new deep learning AIs (like AlphaZero) in e-discovery learn on their own, in an unsupervised fashion, by observing human reviewers and making progressively more accurate predictions about document responsiveness in iterative cycles. Where predictive coding is limited by its human trainers, deep learning AIs never stop improving at their tasks.

Human-AI collaboration is stronger than either human intelligence or AI on its own.

Once Deep Blue defeated Kasparov, the battle for chess supremacy between AIs and humans ended. The machines had won, and further competition had no value. However, human contributions to chess did not end; freestyle tournaments played by teams including both human and AI players began—and the winners often were neither the best human nor the best machine players. No less a luminary than Kasparov himself explained in a New York Review of Books article, “Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.”

What does this say about e-discovery? The machines aren’t coming; they’re already here. But they’re not coming to replace humans. Instead, they are coming to augment them. Given the wealth of ESI pertinent in even mid-tier civil litigation, there is simply too much data for human review to be a viable standard. However, even the most intelligent machines on their own cannot produce results superior to those produced by people who collaborate effectively with the technology at their disposal—especially under real-world conditions, with the deadlines and complex decisions that occur throughout the litigation process.

2017 certainly was—and 2018 is shaping up to be—another big year for artificial intelligence in the legal industry, and we’re looking forward to exploring its impact, where it’s heading, and practical use cases for in-house legal teams.

Make sure you're getting the most out of your investment in AI with the practical tips in Exterro's infographic What AI Is--And Isn't--In E-Discovery.