Editor’s Note: The author of this post is a data scientist at Paul Hastings, and author of a column called “Matters of Opinion.”
By Thomas Barnett, Special Counsel, eDiscovery and Data Scientist at Paul Hastings
One-hundred years ago, doctors didn’t specialize like they do today. They were doctors. Not interventional thoracic radiologists, just doctors. And they didn’t have office managers, billing companies, or nurse practitioners.
And most lawyers were just lawyers. Not cross border structured finance and securitization tax specialists. They didn’t have document processing departments, paralegals or IT experts.
Twenty-five years ago first-year associates at large firms could sometimes be found standing in front of a copy machine feeding in paper for alarmingly extended periods of time — at alarmingly high billable rates. This led some to question their career choices.
Fifteen years ago, junior associates at large firms might spend hundreds or even thousands of hours reviewing and coding discovery documents — again, at full billable rates.
Those days are long gone. Sadly for some, thankfully for others, depending on whether you were sending the bill or paying it.
This evolving division of labor is called disaggregation. It has occurred in many fields for many years, breaking up jobs into component parts and distributing them in order to create economic efficiency. Sounds like a major improvement for lawyers, right? Maybe so… if you assume your own job is safe.
A hot button issue in the presidential campaign this year is jobs moving to places with lower wages, also known as labor arbitrage. But this is also nothing new. Manufacturing jobs in the United States have been leaving for lower wage climes for decades. The migration of office jobs is a somewhat newer phenomenon but has been accelerating rapidly for at least a decade. The US Bureau of Labor Statistics reports that between 2001 and 2013, 3.2 million jobs were lost to China, 800,000 of which were office jobs, 2.4 million of which were manufacturing jobs. And that doesn’t even include jobs lost to other countries around the world.
In the world of large law firms, labor arbitrage also has resulted in the outsourcing of many so-called back-office functions. Often this involves separate companies performing tasks domestically or overseas that used to be done by full-time firm employees. And if you go back far enough, many of these tasks were performed by lawyers themselves.
Certain time-consuming and onerous tasks requiring a law degree such as reviewing discovery documents have largely been outsourced to attorneys in far off lands or to those simply willing to work for far lower wages in the US. This is classic labor arbitrage. While this has affected the bottom line of law firms that previously used vast cadres of junior associates to do this work, most associates probably don’t pine for those good old days. Interestingly, in the years following the Great Recession, with full-time law firm jobs becoming increasingly scarce, the number of U.S. contract attorneys willing to do document review for far less than their law firm counterparts has increased significantly. Put another way, the hourly rate differential which originally led to the arbitrage opportunity overseas has lessened significantly, making review by U.S.-based contract attorneys a viable economic option whether in New York, Chicago, or Wheeling, West Virginia.
Not to say there is no fear of technology on the part of many lawyers. But they haven’t yet confronted the potential loss or replacement of their most important and seemingly irreplaceable asset: their ability to think, reason and render legal judgment.
Perhaps the most frightening form of disaggregation involves actually replacing human beings with machines. For better or worse, automation has been a driving force in industrialization worldwide for centuries — though never popular with those facing displacement. The term Luddite and the early 19th century movement associated with it arose from what some believe to be an apocryphal story of a young 18th-century weaver, Ned Ludd, smashing his weaving equipment in possibly the first recorded case of rage against the machine. The term Luddite has survived over the centuries and has come to be associated with people who generally dislike, or are unwilling to adapt to, technology.
Until very recently however, the kinds of jobs displaced by machines have not been a threat to experienced attorneys with expensive law degrees and coveted pedigrees. There haven’t been reports of 21st century law firm associate Luddites smashing their laptops and personal devices — though I imagine the thought has occurred to some. That is not to say there is no fear of technology on the part of many lawyers. But they haven’t yet confronted the potential loss or replacement of their most important and seemingly irreplaceable asset: their ability to think, reason and render legal judgment.
But, alas, ye legal Luddites, there is a new scourge upon the legal landscape: machine learning, artificial intelligence, natural language processing. It goes by many names — computer programs that are on the brink of being able to analyze, synthesize, categorize, and classify documents and contract provisions — at a level that could make some lawyers quake in their designer shoes.
This is a new paradigm in the legal industry. The economic imperatives are unavoidable. High billable rate associates can be greatly assisted, if not eventually replaced, by computer programs that learn by modeling human decisions. The unsettling thing for some lawyers not previously threatened by outsourcing is that the type of tasks that are starting to be automated are far more complex and impinge on their expertise and decision-making ability.
Supervised machine learning, by definition, looks at decisions made by human beings and attempts to replicate them. In other industries this has been used to perform tasks that require significant amounts of knowledge, experience and judgment. Some examples include medical diagnoses, loan application approval, financial and investment decisions and various types of pure research. This doesn’t mean that humans are completely cut out of the picture; far from it. But what it does mean is that the work of many individuals for many hours can be dramatically reduced. This leaves higher level judgment and decision-making to attorneys, which, at least for now, can’t be replicated. In a large law firm, if a well-trained computer program can analyze at the level expected of a 4th or 5th year associate, that computer work product can then be assessed by a senior lawyer and modified as necessary, potentially eliminating the work of large numbers of such mid-level lawyers. The traditional pyramid law firm business model of a small number of partners above large numbers of high billing associates is directly challenged by this possibility. And the driving force for change is formidable: The large firm client base is acutely and intensely interested in exploiting every possible cost-saving measure.
Surprisingly perhaps, attempts to get computer programs to perform legal reasoning go back decades, with several books and academic papers on the topic dating back to the late 1970’s and early 80’s. This is well before computers were in wide public use or even present in the public consciousness and may tell us more about how people felt about lawyers than how they felt about computers. But with advances in recent years in computing power, statistical engineering and machine learning technology, automating legal analysis is a lot more plausible now than in the era of the main frame computer and the floppy disk.
Ultimately, the question is not whether computers have to learn to think like humans to solve human problems. The real question is whether human thinking is required at all for certain tasks.
We now have computers that can beat the best chess player in the world and that can dominate Jeopardy tournaments. Just this month, a computer was able to beat the world’s best Go player, the ancient Chinese game, using techniques that allowed it to model 30 million moves of human players as well as play thousands of games against itself to improve its performance. Computers programmed to play these games rely on the ability to process and compare many millions of options in fractions of a second and pick the best one. A world class chess player can think 12 to 14 moves ahead. The best chess program can process millions of moves per second. It’s the same problem but with a very different solution. Computers don’t approach these highly specialized problems in the same way humans do.
Does that mean computer programs can think or reason? That depends more on how you define those words than on some ultimate truth about human physiology and computer programing. And ultimately, the question is not whether computers have to learn to think like humans to solve human problems. The real question is whether human thinking is required at all for certain tasks. That is, can sophisticated legal analysis be successfully measured and modeled in such a way that it can be replicated using a defined set of steps or instructions, aka, an algorithm?
Based on current trends, the answer appears to be, yes. Using machine learning to perform more advanced legal decision-making is on its way. There will no doubt be strong resistance to this new division of labor between people and machines in the law firm. Because, as Neil Sedaka’s 1962 classic hit reminds us: breaking up is hard to do.