IBM's "Watson" computer made headlines in 2011 when it triumphed in a head-to-head contest against record-breaking human winners of popular US general knowledge quiz, Jeopardy! Now, five years on, a "robot lawyer" has been developed based on the same underlying technology. In a coup for AI enthusiasts, ROSS has apparently been found impressive enough to convince at least one major US law firm to "hire" ROSS to assist attorneys dealing with bankruptcy cases.
From the reported details, ROSS seems to be tailored to analysing plain language. This makes it ideally suited to reading court judgments, which are written in more-or-less normal language, even if it might not always seem that way. Its creators don't see it as replacing lawyers altogether (yet). Rather, its value lies in its use as a tool to assist lawyers in their legal research - carrying out in milliseconds a case law review that might otherwise take hours for a team of paralegals or junior attorneys to complete.
This made me think about whether a similar AI could be of any wider relevance in the world of patent law, beyond the confines of case law research. A judgment is a judgment, after all, and is written in plain(ish) language whether it relates to penury, perjury, or patentability. What interests me in particular is the potential for a ROSS-type AI to have applications in patent searching instead.
Prior art searches are an essential tool for patent attorneys and patent examiners alike when determining the likelihood that an invention is patentable; freedom-to-operate searches are highly advisable before bringing your new blockbuster product to market to make sure that you aren't infringing anyone else's IP. And yet, despite (or perhaps because of) the wealth of knowledge and information now available online and in extensive databases, patent searching remains a laborious and time-consuming effort that still delivers only a limited degree of certainty at best. This is not to cast aspersions on the ability of those carrying out the searches; it's simply the inevitable consequence when a human mind with a limited amount of time and a limited budget is faced with a quasi-Borgesian repository of information.
Unlike court decisions, patent applications are often written in language that can seem obscure, opaque, or obfuscatory to the untrained mind. For every patent application that calls a spade a spade, there is usually another that calls a spade a manually-operated earth relocation implement. As a result, patent searching remains as much of an art as a science, requiring a significant injection of intuition, lateral thinking, background knowledge of patent law and claim interpretation, and understanding of the technical field of the invention in question, in order to define the parameters of the search and to sift through and interpret the results. Even with the assistance of an outsourced searching agency with the time and resources to focus on locating possibly-relevant prior art, ultimately the patent attorney still needs to devote significant time and effort to analysing their findings.
In this context, could an AI research assistant - impressive though these reports are - really find any application in a field that relies as much on Fingerspitzengefühl as anything else? Well, perhaps. The developers of ROSS claim that the more you use it, the more it (he?) learns - and it seems to be context-sensitive. As baffling as the language of patent claims can be, there is normally an example or a detailed embodiment described somewhere in the patent specification which gives some insight as to what the drafter had in mind when writing the document. Over time, it's not inconceivable that a suitable AI might learn to associate particular patterns of patent language with certain types of technology, and become smarter at filtering the data to present only the most relevant results.
At the risk of excessive hype, it does seem that the 2010s could be the decade when AI finally begins to come of age. While AI is still unsettled and sometimes seems closer to Microsoft's infamous "Tay" (the chatbot which reportedly turned racist, misogynist and generally all kinds of abusive within hours of activation) than to Iron Man's JARVIS, the fact remains that Apple's "Siri" has now become almost commonplace for many people and ROSS seems to be undergoing serious consideration by reputable law firms. Ultimately, it's hard to envisage that AI will replace lawyers entirely, so patent attorneys shouldn't fear that ROSS is about to send them on a break any time soon. Law is as much about the very human arts of persuasion and negotiation as it is about the cold, hard facts. But the prospect of being able to ask, "ROSS, fetch me all patents currently in force relating to spades" and have the results delivered near-instantaneously is an enticing one, and when it becomes reality, I, for one, will welcome our new robot paralegals.
"With ROSS, lawyers can focus on advocating for their client and being creative rather than spending hours swimming though hundreds of links, reading through hundreds of pages of cases looking for the passages of law they need to do their job."