At ROSS, we’re always talking to lawyers about how artificial intelligence is changing their practice. Our discussions are usually focused on AI and legal research. But sometimes the conversation turns to applications that might have seemed far-fetched a couple of years ago but now don’t seem quite so crazy.
A recent conversation with plaintiff’s personal injury lawyers turned toward the application of prediction markets to settlement negotiations. The gist of the idea was that crowd-sourced predictions on settlement amounts could move unreasonable parties into reasonable settlement range. Although the theory that prediction markets could assist in litigation settlement has been around for awhile, it hasn’t gained much practical traction. We discussed whether or not artificial intelligence could play a role in changing the status quo.
A prediction market is a speculative exchange created for the purpose of predicting the outcome of a particular event. Prediction markets attach a monetary value to every possible outcome of a certain event, with each outcome being valued as through it were a stock traded on a stock exchange. The price of each outcome is interpreted as the market’s collective estimate of the likelihood of that outcome occurring. Group predictions are designed to replace the biases that normally influence decision-makers and provide a financial incentive to make for accurate predictions.
A prediction market focused on presidential elections is a good example. A contract can be structured to pay $1 if the predicted outcome materializes and $0 if it doesn’t. If a futures contract stating that Candidate X will become president trades at 70 cents that means, according to the market, there’s a 70 percent chance that Candidate X will win the elections. If Candidate Y’s contract trades at 30 cents, the market puts their chances at 30 percent. If the contract that you bought supports the eventual outcome, you win the bet. If the outcome goes the other way, you lose. As time goes by and more and more people buy the contracts, the prices will fluctuate depending on market conditions and the combined information held by market participants.
In the litigation settlement context, a mediator could propose a prediction market contract to try to close the gap between the plaintiff’s $25 million demand and defendant’s $25 offer. The mediator would issue a contract that pays $1 if a case settles between 5 and 7 million dollars. If the contract is issued at the commencement of trial, market participants could review the evidence. If the contract will resolve pre-trial, market participants could review motions for summary judgment and other information as a basis for their prediction. If the case settles, the settlement amount would be reported and the correct predictions would be paid. If no settlement is reached, the investments would be returned. The changing contract price would indicate the perceived likelihood of a certain settlement level. As the contract price moves in one direction or the other, the parties might be influenced by the collective estimates of the market participants of the value of the case.
Litigation settlement prediction markets have never taken off for a variety of reasons. Parties are reluctant to report much information about settlements in the vast majority of cases that never go to trial, prediction markets are cumbersome to set up for events that might not attract the interest of a presidential election, and reporting the final results can be cumbersome.
Interest in settlement prediction could be revived by blockchain prediction markets. Consider Augur. Augur is billed as the first decentralized blockchain, oracle-based decision market platform. As in any prediction market, Augur users can bet on the outcomes of future events. Augur claims that its blockchain platform will enable people to trade in prediction markets at very low cost. In theory, a blockchain platform like Augur is low cost because the platforms do not rely on a central market maker. Anyone can open a market, meaning that every mediator could, with the consent of the litigants, enlist prediction market participants to assist in the resolution of a case.
The most important byproduct of legal prediction markets would be data. Once enough markets were opened and resolved, significant testing and training data would be available to machine learning engineers working to create accurate prediction algorithms correlating evocative facts, predicted market results and actual reported settlements. As algorithms generated more and more accurate predictions, they could be used to advise (and bully) more and more parties in early case evaluations, mediation and arbitration.
The most significant drawback to the application of prediction markets to alternative dispute resolution--and the largest barrier to adoption of the concept--is that prediction markets have not historically been very good at making predictions that are dependent on an individual's idiosyncratic choice. Since most litigation settlement decisions ultimately turn on the decision of a single person or small group, the perceived whims of individuals could dissuade bettors from taking part. However, as the use of prediction markets became more widespread, it would become more difficult for individuals to justify edge settlement positions. And those individuals who insisted on outlying positions would be unlikely to participate in a prediction market to being with.
Another significant barrier to adoption is the reluctance of litigants to disclose settlement amounts. For the most part their concerns are misplaced. It should be possible to anonymize some information without lessening the value of the information to market participants. In any event, the aversion to public disclosure could weaken over time as the prediction markets matured. Once lawyers and litigants saw the benefit of obtaining guidance that could settle intractable cases (always confident that the other guy is being unreasonable) they might be willing to make some public disclosure in exchange.
The primary regulatory challenge would be controlling insider trading, particularly in low liquidity markets. But it’s unlikely that a lawyer or client would steer toward a suboptimal settlement in order to game a prediction market.
Ultimately, machine learning might reveal what many lawyers suspect: very few disputes are truly unique and enormous sums of money are sometimes spent discovering facts that don’t really change either party’s view of the value of the case. As machine learning becomes more accurate, parties could start to view the algorithm as a trusted guide to settlement value.
Charlie von Simson is a legal subject matter expert at ROSS. He practiced law for twenty years before running away to join a startup.
Last week I attended the ABA Section of Labor & Employment Law National Symposium on Technology in Chicago. Since Labor & Employment law is a focus here at ROSS, I was particularly interested in a panel discussion on the evolving technology-related jurisprudence of the Trump-era National Labor Relations Board...
In order to understand how ROSS’s A.I. is changing legal research, it’s important to understand machine learning, natural language processing, and other A.I. basics. From that basis, we can more meaningfully discuss how those systems are being deployed...
As artificial intelligence underpins more and more public and private decision-making, critics are justifiably concerned about the potential lack of accountability for how those decisions are made. In our view, transparency is the key to ensuring that decisions that are aided and augmented by algorithmic data analysis are ethically sound...
Join the leading minds in legal tech and get the latest news and updates on the law, legal technology, artificial intelligence, and much more.