Thursday, June 30, 2016

Chatbot Wins 160,000 Legal Cases - the "Future is Now"

From Yes Virginia, There is a Santa Clause (1974) came one of the great movie lines of all time: "have your people call my people, we'll do lunch." It has reappeared in various other movies including The Muppets Take Manhattan, Full House, JAG, and more. It has become iconic. A fellow traveller (a hat tip for the tip) emailed me a link to an article from the Mirror this week. As I read it, this quote occurred to me, but I struggled to put it into context. 

The article World's first 'robot lawyer' overturns 160,000 parking tickets, appeared in the Mirror. And the age of technology hits a little closer to home. I cannot feign surprise. I have touched on technology a few times, such as" How will Attorneys Adapt and Attorneys Obsolete. Technology is changing our worlds and we cannot feign surprise. It is not yet foolproof, fortunately. We may still have time to adapt (or run). Coincidentally we learned in June of Promobot, a Russian artificial intelligence robot., escaping its creators and making a run for freedom, twice. Now, exceeding its creators' control, Promobot may be destroyed.

The legal world itself is in  a state of change. Law schools across the continent have produced ever-larger crops of new attorneys in the last twenty years, and the market is flooded with attorneys. Despite this expanding supply, the attorneys want to override the basics of economics and feel they should nonetheless be paid ever-increasing fees. Supply and demand dictates that as supply increases, price should fall. If that applied to the practice of law, some argue that the lawyers should be paying their clients by now.  

Another challenge in the legal market is the consistent criticism of insularity. There are disputes about "what are legal services," and who can provide them. The WaPo recently reported Lawyers remain deeply skeptical of non-lawyers investing in law firms. There are those who feel that such diversity of ownership and some deregulation of the profession would mean more diversity and more jobs for attorneys, others fear competition and argue such deregulation would "destroy the practice." In 2015, the WaPo reported on a move in Washington state to empower paralegals to perform a broader range of services in Who says you need a law degree to practice law? There are similar initiatives under debate and development in other professions, such as medicine. The protectionism of professional privilege is under attack from non-professionals, para-professionals and now technology.

Last spring, I received an email from a colleague about a new legal research tool called Ross. It is an artificial intelligence application that provides intuitive and interpretational help to lawyers performing legal research. I wrote about it in Ross, AI, and the new Paradigm Coming. Ross is built on a similar platform as the International Business Machine program Watson. Some may recall watching Watson defeat two champions simultaneously on Jeopardy?

The import of Ross is that we have progressed from the days when lawyers wrote complex "search string" inquiries for research databases. Those strings were words, associated with each other by "connectors" such as "and," "or," and "but not." We evolved to restricting these searches later with such things as "w/2" and "/s" and "/p." The language was known as boolian alegebra (I was so pleased in the 1980s to find an alebra I understood). The "connectors" were merely logic statements for the computer.

To search for a case in which comparative negligence in an automobile accident was an issue, one might use the search string: Automobile w/5 "comparative negligence." The "w/5" told the computer that you wanted cases in which the word "automobile" appeared within 5 words of the phrase "comparative negligence." If that specificity did not yield results, the lawyer might change it to "w/s" meaning in the "same sentence, or "w/p" within the same paragraph. 

The computer was powerful, its knowledge base deep and broad, but it needed us. The lawyers wrote the query and that produced specific results. Written too broadly, the query returned thousands of responses; too narrowly and there were none. We had to specify, analyze, and refine repeatedly. The senior partner in my first law firm could not comprehend or appreciate the technology. He would send me his ideas for search strings in printed memos and I would do his research for him. He was convinced computers were for "secretaries," and that my interest in technology would doom my professional future. Unfortunately, it was not the only time he was wrong. 

But, we lawyers evolved from Boolean algebra strings to "natural language." These were programs that could take a plain English sentence and without our help turn it into a Boolean query. These "thinking" programs evolved rapidly, and have been quite common for the last decade. I taught a legal writing class last year and found that today's students do not even know what Boolean algebra is. They, as a generation of students, have grown up with the "natural language" format that has become the norm in legal research and the Internet generally. 

My analysis of Ross last spring has made me think a great deal. Can computers be programmed not only to respond with factual data, but to learn and think? It seems the answer is clearly yes. It is not coming, it is here in Ross and programs like it. For those of us that remember the Boolean algebra process and how long it ruled, its replacement by "natural language" and the current threat of artificial intelligence (AI), the pace of change appears to be increasing. We are assuaged by the little setbacks like Promobot, because they suggest our relevance remains, for now.

Now, researchers at the University of Buffalo claim that their computer can detect deception with increasing accuracy. An older article from MIT attests that this research and progress is not as recent as you might think. Computers are increasingly able to discover deception, effectively, efficiently, and increasingly accurately (hold on, that's my job; this is getting personal). 

Returning to the Mirror, we have news that the World's first 'robot lawyer' overturns 160,000 parking tickets in London and New York. This success is not a lawyer's. This success belongs to a "free online chatbot lawyer." A chatbot is an artificial intelligence. They interact with humans and other chatbots, and they learn, they adapt, they develop. Recently there have been some interesting chat bot stories in the news. They are not yet infallible. Microsoft's chatbot Tay learned to become a racist in no time. Some even question whether chatbots are just a fad.


Well, the "Robot Lawyer" chatbot has overturned "a staggering 160,000 parking tickets in London and New York City, saving users an estimated £2.9 million." This is a recent development; it has been operating for less than two years. The "world's first robot lawyer" is housed on the web at a site called "DoNotPay." The chatbot guides "users through a range of basic questions to establish if an appeal on their parking ticket is possible." In effect, it asks the questions that a lawyer might ask, in seeking to evaluate someone's chances in a legal setting.

Well, with such success, my first perception was some attorney programmed a computer to mirror her or his legal advice thought process. But no. It was not written by some legendary barrister of "Superlawyer." The chatbot is the brainchild of 19-year-old British student and self-taught coder Joshua Browder." He concluded that many parking tickets are illegally issued, and difficult to dispute. So, he wrote DoNotPay "by scanning thousands of documents released under the Freedom of Information Act." With some "guidance" from an attorney, he built the foundation of this litigation chat bot. I say foundation, because it is critical in this context to remember artificial intelligence is dynamic; it learns and evolves from success and failure just as we do. 

DoNotPay has helped a staggering 250,000 "clients" in its short (less than two year) career. It has succeeded in 160,000 of those cases, an amazing 64% success rate. No lawyer in her or his life will handle 250,000 cases. And a success rate of that magnitude, for a computer program, is simply amazing. Mr. Browder feels that DoNotPay is fighting injustice, protecting "the most vulnerable in society." Well, perhaps not as vulnerable as those who are accused of crimes that might result in jail or worse. But, admittedly, there are some towns and villages that may exploit the convenience of a parking ticket more for revenue than justice (some might even deploy their own cowardly bots to snap photos of accused violators in parking and beyond?)

Those cases will have to wait for now. In the foreseeable future, Mr. Browder sees the technology helping "people claiming flight delay compensation, help HIV Positive people to understand their rights, and provide support for refugees struggling to navigate foreign legal systems." And any such deployments may impact attorneys. As technology evolves and these kinds of programs are mated with ever more capable computer platforms like Watson and those that follow, it may be eve more difficult for attorneys to keep up, just as those Jeopardy champions were bested, the most capable attorneys may find themselves losing to chatbots

More importantly, at least for me, is the potential that such programs could be mated with the deception detection technology described above. That hybrid computer/software (digital judge, or "DJ") might measure the relative weight and persuasiveness of two other chatbot's arguments (digital attorneys), and assess the credibility of the humans involved regarding facts. Then the DJ computer could itself make a decision. The first artificial intelligence (AI) judge. And I find myself replaced by a DJ. 

Of course, we might still disagree with the DJ outcome. Could we have the decision re-analyzed by three more AI computers (we could call them the appellate chatbots or AC). They could review the analysis of the chatbot DJ and then agree (affirm) or not (reverse). In the process, the collaboration of those three AI ACs would "teach" the chatbot DJ, which would evolve and develop and learn. And, as importantly, as we learned on Jeopardy, the computers would all learn and respond faster than we could. An appeal could be as simple as pushing another button on the computer, and the AC response might take seconds. 

It is an interesting time. But, it may yet become infinitely more interesting. As the professions argue about dividing lines between professionals, paraprofessionals and lay persons, the bots may be taking over. It may be that we are fiddling while Rome burns. 

Oh, and now I realize why Yes Virginia, There is a Santa Clause is in my thoughts; "have your computer call my computer" and let me know what it thought of all this. Because we may not have or need "people" much longer. The Jetsons are here.