Sam, Josef's COO, recently sat in on a lecture by Harry Surden, Associate Professor of Law at the University of Colorado, on the past, present and future of AI and the Law. Here's what he learnt.
When I’m talking to others about Josef, I am often asked some variant of the question:
“Does Josef have AI?”
The answer is “Yes”. AI is in our DNA: Tom, our CEO, studied legal data analytics at Columbia University in NYC and Kirill, our CTO, is a machine-learning expert.
But that is often not the answer I give. Instead, I have to first unpack the question with the person who asked it. This is because AI is a heavy concept, weighed down by exaggeration and hype in popular culture, excitable media articles and – I hate to say it – marketing materials from tech companies. This noise has given rise to countless misconceptions. And, now, dealing with them is a hazard of the job.
Recently, I attended a lecture by Harry Surden – Associate Professor of Law at the University of Colorado – on AI, hosted by the CodeX Institute at Stanford University. At the lecture, Surden explained what AI is, where it came from in relation to the law and where it is headed.
For anyone who wants to cut through the noise, too, below is a summary of his talk.
What is AI?
AI is notoriously hard to define. In fact, many dislike the term because it is inherently misleading. During the lecture, Surden proposed the following definition:
AI is technology that does things that, when done by humans, requires some intelligence or cognitive processes.
On this definition, a driverless car uses AI because it requires spatial reasoning. But, Surden was quick to point out that there is little similarity in the way this intelligence is realised. “Computers solve these problems very differently to humans. Computers drive cars very differently to humans.”
Surden spent more time discussing what AI wasn’t, than what it was. In particular, he stressed that we’re not close to “strong AI”. This is the AI of the media, the AI that functions with equivalence to a human mind. In Surden’s view, we’re not even close to this. At the moment, a 2-year-old child has more advanced cognitive skills than AI that currently exists.
What does AI look like today?
But that is not to say that AI doesn’t offer significant opportunities for society, even today. AI is very powerful in limited, specific situations, particularly when it involves patterns. AI can already, for example, play chess better than humans.
Broadly, there are two forms of AI. The first is rules-based AI, which is what Josef uses. Surden called this “knowledge representation”, by which we map real world processes or systems using logical rules.
Surden took as his example Turbotax in the US, which is used by over 60 million people and is often deferred to by the IRS. All that was required to create Turbotax was to translate legal rules into logical rules.
The second category is machine-learning. This has become very popular over the last 25 years. Broadly, machine-learning involves algorithms that examine data, detect patterns in that data and then use that to achieve different things. “When a car drives itself, this mostly involves machine-learning,” said Surden. “When Netflix suggests a movie, that involves ML. When email tells you that an incoming email from Nigeria is likely spam, that’s ML.”
Surden was careful to make clear that this form of AI does not “learn” in the sense that we usually use that term. Rather, it recognises patterns formed by “good” and “bad” examples in large datasets.
History of AI and the Law
Surden spoke about the history of AI and the law, which he had traced back to the lawyer and mathematician, Leibniz, in the 1600s. Leibniz, the co-inventor of calculus, suggested we could use mathematics or algorithms to make the law more useful or predictable or understandable or manageable.
Surden connects Leibniz’s project to that of the legal practitioners and scholars in the modern era, from the 1960s onwards, who began to use artificial intelligence to try to solve legal problems. The first wave occurred from the 1960s to the 1980s. These academics, including Richard Susskind, used rules-based systems to digitise legal knowledge. Most of this work occurred in the US, Netherlands, the UK and Germany.
From the 2000s, we have seen more use of machine-learning in this space, including through institutions such as CodeX. Though inspiring, these projects remain somewhat hindered by a lack of accessible data in the legal industry.
AI and the legal industry
Surden counts three primary users of legal AI. The first is judges and the courts, such as the sentencing and bail reports which are produced in the US by machine-learning algorithms to show the risk of re-offending.
The second is practitioners. Practitioners are using more and more technology, such as Josef, to supplement their day to day work. For example, technology assisted document review has now largely replaced purely manual document review. Surden recalled being an associate sitting in a dark room going through boxes. Today, we use machine-learning algorithms which are better and faster at this initial review. “Humans are still in the loop”, noted Surden. “We’re just shrinking the haystack”.
And the final category is consumers of legal services, which Surden calls out as being served by bots and chatbots, like DoNotPay.
Surden was careful to point out that there is a lot of uncertainty about where AI is headed next. However, one thing he was quite confident about was that AI would remain, at least for the foreseeable future, a tool for humans to use, rather than anything to be scared of.
AI enhances humans, rather than replaces them. It provides them with the ability to make better decisions. For example, AI can now beat humans at chess. But, if you combine a human with an AI chess-player, the human can beat the computers.