AI Predictions and the Case for Neutral Evaluation

Picture a mediation in which both parties arrive with an AI-generated assessment of the case, and each assessment says that its side will win. Neither report moves the other side an inch. As prediction tools become cheaper and more widely used, that scene is likely to become familiar, and the disagreement will simply have acquired two confident new voices. pinqDR considers why private predictions rarely settle disputes, and what does.

Two predictions, one dispute

The spread of prediction tools is broadly to be welcomed. Parties can test their position earlier, prepare more efficiently and spend less on work that goes nowhere.

Each party, however, feeds its tool the documents it holds and, naturally, the documents that help it. The claimant uploads the contract, its invoices and its own account of the correspondence; the respondent uploads its version of events. Each tool reasons faithfully from what it was given, and two partial records produce two coherent, well-reasoned and incompatible predictions.

Neither prediction is dishonest, but both are incomplete, and put side by side each becomes a talking point rather than a test. Lawyers will recognise the pattern. Each side has always taken counsel's opinion on its own papers, and opinions on the same dispute have often pointed in opposite directions. AI does not create that problem, but it can make it faster, cheaper and more persuasive.

Why disputes persist

Parties rarely disagree about the law in the abstract. They disagree about the facts, about which documents matter and about how a tribunal will see them. A private prediction built on one side's record cannot resolve that disagreement, and it is more likely to harden it.

It can also lend the disagreement a veneer of objectivity. "Even the AI agrees with us" is a powerful line in without prejudice negotiations, and a misleading one when the AI has seen only half the file. The position each party brings to the table becomes more entrenched and settlement, which depends on each side revising its expectations, becomes harder to reach.

The value of a neutral view

Settlement tends to come when both sides accept a view that they did not produce themselves. That is the logic of early neutral evaluation, which long predates AI: an independent, non-binding assessment of the material the parties have put forward, delivered early enough to shape the negotiation. The courts recognise its value. In Lomax v Lomax [2019] EWCA Civ 1467, the Court of Appeal held that the court may order an early neutral evaluation hearing under its case management powers even where a party does not consent. Its value lies as much in its neutrality as in its accuracy. A party can dismiss its opponent's prediction; it is far harder to dismiss a reasoned view formed on the material from both sides.

The same principle applies to AI evaluation. The important question is not which tool is the most sophisticated, but what the tool was given. An evaluation run on the full record, including the other side's pre-action correspondence, documents and stated position, is a different thing from one run on a single party's file, because it tests a case against its strongest opponent rather than its most convenient one.

That has a practical consequence for anyone using AI on a live dispute: give it the other side's case. The letter of response, the key correspondence and the documents on which the opponent relies all belong in the file. An evaluation that has not seen them is a rehearsal rather than a test.

From evaluation to decision

A pinqPredict early neutral evaluation is only as complete as the documents it is given, which is why the file matters as much as the tool. Used on the full record, it gives a party a candid view of how its case looks against the other side's best points, early enough to act on it. For a respondent or defendant the discipline is the same: the claimant's best points belong in the file.

Where evaluation does not settle the matter, pinqDR provides the neutral endpoint. Each party puts its case to an experienced arbitrator, who decides on the whole of the material and issues a final and binding award in 6 to 8 weeks, enforceable, with the leave of the court, in the same manner as a judgment. At that point there is no competing prediction to argue about, only a decision.

When everyone has a forecast

As prediction tools spread, their power to persuade will fall. When every party can produce a favourable forecast, favourable forecasts will carry little weight across the table. What will retain its value is neutrality, both in evaluations that engage with both sides' material and in tribunals whose jurisdiction both sides have accepted.

At pinqDR, we see AI and fast arbitration as working together. AI makes an early, reasoned view affordable, and a binding award in weeks makes the final answer quick. Between them, they shorten the period in which two confident predictions keep a dispute alive.

To find out more about pinqPredict and pinqDR, get in touch at hello@pinqDR.com.

Useful links

Lomax v Lomax [2019] EWCA Civ 1467

https://caselaw.nationalarchives.gov.uk/ewca/civ/2019/1467

pinqDR, Introducing pinqPredict

https://pinqdr.com/blog/introducing-pinqpredict

pinqDR, Rules

https://pinqdr.com/resources/rules