MACHINE LEARNING: WHEN COMPUTERS GO TO LAW SCHOOL In the 1950s, while working at IBM, Arthur Samuel developed a program that played checkers. To train it, he made it play against itself. Since the machine learned autonomously, Samuel called this method "machine learning". In the expert systems we studied in the previous lesson, a programmer writes the steps a software must follow to achieve some goal. In machine learning, it's the computer that discovers the steps by searching for patterns in datasets. Deep Blue's victory over Kasparov was a big event in the history of computer science, but it was not unexpected. Since the early days of artificial intelligence, it was thought that someday a machine would beat the world champion of chess. However, it was unclear that a machine would someday beat the world champion of Go, a much more complex game. In 2015, AlphaGo, a program developed by Google with a type of artificial intelligence known as "deep learning", defeated Go champion Lee Sedol. It was a spectacular show of the advances in machine learning. Machine learning is at the core of many products we use every day such as GPS navigation, online translations, virtual assistants, content recommendation engines and facial recognition. Much of a lawyer's job is about pattern recognition. And machine learning can recognize patterns much faster and much more efficiently than humans. This is giving birth to revolutionary products in the legal industry. A first type of application is contract analysis. Noah Weisberg was a corporate lawyer based in New York. In his job, he had to read thousands of pages to identify anomalies in M&A contracts. So he decided to develop some software to automate the process. That idea gave birth to Kira Systems in 2011. The user uploads a contract on a platform and the program automatically identifies risks. According to the company, this cuts the review time between 20 and 90 percent. Another firm in the area of contract review is the Israeli Law Geex. In February 2018, the company organized a competition between its algorithm and 20 American lawyers. The goal was to detect problematic clauses in five confidentiality agreements. The average accuracy of the attorneys was 85 percent. That of the algorithm was 94 percent. On average, it took lawyers 92 minutes to analyze the five contracts. The program did it in just 26 seconds. Another key application of machine learning is in the field of legal analytics. Much of a lawyer's job involves predicting the outcome of cases. For example, how likely is it that a client will end up being compensated for a traffic accident? What is the probability that a case over a contract breach will have a favorable decision in court? Lawyers study the facts, analyze strategies and determine odds of success. For this, they must know the results of similar cases in the past and also try to predict what the other party's strategy might be. Legal analytics leverages data science to help in these decisions. A pioneering company in this field was Lex Machina, born at Stanford University in 2006. It offers a database which assists users in evaluating court performance. For example, how did the judge rule in similar cases in the past? What were the arguments he considered most convincing? Algorithms help estimate the odds of success of different claims. Lex Machina started with intellectual property cases. After being acquired by LexisNexis, it expanded its reach to antitrust, finance, commercial, labor and bankruptcy law. Premonition is another pioneering company in the field of legal analytics. Before accepting a case, a lawyer can use this service to answer a number of key questions: What is the typical duration of a case such as this one? What proportion are won by the defendant? In what percentage is a settlement reached? The other party's lawyer, tends to be conciliatory or tough in negotiations? With these tools, lawyers can make better estimates of the potential gain and cost of different cases. These estimates are also transforming litigation funding. If we can predict the likely outcome of a trial, then we could estimate how profitable it may be to fund the parties. A company called Legalist raised a $100 million investment round with this business model. They use analytics to predict case results and they fund the party that is most likely to win. Machine learning has a great potential to transform the legal industry. But some are concerned with ethical aspects. Could it be an undue influence on the behavior of judges? Could it result in higher inequality between parties that have access to these tools and those that don't? Because of concerns like these, the French government introduced restrictions on the use of analytics in court. In the 19th century, jurist Oliver Wendell Holmes predicted that, someday, courts would base their decisions on statistics. Today, his prediction seems right on spot. If we combine expert systems with machine learning, we can start to imagine what the legal system of the future will probably look like.