Industries / Government, education and community / Government / Courts / Arbitration
Mediation
A mediator is a neutral person who helps the two sides of a dispute talk and reach their own agreement; unlike an arbitrator, the mediator decides nothing.
7
Milestones achieved, from 1947 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1947
A federal mediation service
The US Labor Management Relations Act of 23 June 1947 created the Federal Mediation and Conciliation Service (FMCS), an independent agency that took over the Labor Department's Conciliation Service to mediate disputes between employers and unions. In March 2025, after an executive order of 14 March told agencies to cut staff, FMCS limited its services to large bargaining units and was left with six mediators; a federal court in Rhode Island permanently blocked the order on 21 November 2025, and on 30 December 2025 a federal judge in New York vacated the FMCS policy and ordered the layoffs reversed.
LessonA public mediation service can be cut almost to nothing by one order, and the unions and states that relied on it had to go to court to get it back.
Credit
US Congress; for the 2025 rulings, the US District Courts for Rhode Island and the Southern District of New York
1976
The multi-door courthouse
In April 1976, at a national conference on public dissatisfaction with the courts (the "Pound Conference"), Harvard law professor Frank Sander gave a paper, "Varieties of Dispute Processing", proposing a center where a screening clerk would send each dispute to the process that suits it, such as mediation, arbitration or trial. At the time nearly all US courts offered only trial by judge or jury; the idea came to be called the "multi-door courthouse" and spread court-linked mediation programs.
Credit
Frank E. A. Sander, Harvard Law School
2008
Europe asks for mediation
The EU's Mediation Directive of 21 May 2008 required member states, by 21 May 2011, to make mediation available in cross-border civil and commercial disputes, keep it confidential and make settlements enforceable. A study for the European Parliament in 2014, drawing on up to 816 experts, found mediation was still used in less than 1% of civil and commercial cases in the EU, and called this the "EU Mediation Paradox".
LessonAllowing mediation and praising it did not make people use it. 493042_EN.pdf).
Credit
European Parliament and Council of the EU; Giuseppe De Palo and colleagues for the 2014 study
2011
Italy makes mediation a step
From 21 March 2011 Italy required people in certain kinds of civil case to try mediation before going to court. Official figures quoted in the 2014 European Parliament study count 220,000 mediations from March 2011 to October 2012, more than 12,000 a month by March 2012, and close to half settled when the other side agreed to take part. Italy's Constitutional Court then struck the requirement down because the government had issued it without Parliament's authority, mediations "came to a virtual halt", and a 2013 decree restored it from 20 September 2013.
LessonThe requirement to attend, not the incentives around it, produced the volume. 493042_EN.pdf).
Credit
Italian Government and Parliament; Italian Constitutional Court
2011
China's people's mediation law
China's People's Mediation Law, promulgated on 28 August 2010, took effect on 1 January 2011. It governs free, government-funded people's mediation committees that settle neighbourhood, family and other community disputes. China's Ministry of Justice said 14.94 million disputes were mediated in 2022 and that the country has 3.18 million people's mediators, 412,000 of them full-time.
Credit
Standing Committee of the National People's Congress; Ministry of Justice of China
2020
Settlements enforceable abroad
The UN General Assembly adopted the Singapore Convention on Mediation on 20 December 2018. It lets a party to an international commercial settlement reached through mediation ask courts in other member countries to enforce it directly. 46 states, including China and the United States, signed on the opening day, 7 August 2019, and it entered into force on 12 September 2020. The UN treaty record now lists 60 signatories and 25 parties; China, the United States and the United Kingdom have signed but not ratified.
Credit
UNCITRAL, the UN commission on trade law; Singapore as host
2024
Small claims sent to mediation
In Churchill v Merthyr Tydfil (2023), the Court of Appeal of England and Wales held that courts may pause a case so that the parties try mediation or another form of out-of-court settlement, which overturned a long-held view that courts could not compel it. From 22 May 2024 a pilot running to 21 May 2026 sends defended small money claims of up to £10,000 to a free one-hour telephone mediation with a mediator from HM Courts and Tribunals Service before a hearing. The law firm BCLP estimated that about 92,000 cases a year would be covered.
Credit
Court of Appeal of England and Wales; Civil Procedure Rule Committee; HM Courts and Tribunals Service
Watch for
A treaty body's first settlement
The International Organization for Mediation (IOMed), set up in Hong Kong by a convention signed on 30 May 2025 and in force from 29 August 2025, calls itself the world's first intergovernmental legal organization for mediation and began work on 20 October 2025. Its Secretary-General, Teresa Cheng, announced on 8 May 2026 that it had settled its first case, a confidential maritime dispute between Chinese and Singaporean companies over a chain of ship charters.
Achieved whenIOMed publishes its caseload, or a dispute between states is mediated there and the states confirm the outcome.
Claimed by
IOMed, May 2026
AI mediators calm disputes
James Hale and Jonathan Gratch had 164 crowd workers play a seller in an online-purchase dispute against a buyer played by GPT-4o, with an AI mediator built on OpenAI's GPT-5 models, an untrained crowd worker as mediator, or no mediator. They report that the AI cut negative emotion in the sellers' messages significantly more than the human mediators did. 47.3% of AI-mediated disputes ended in deadlock, against 58.0% with a human mediator and 54.3% with none, a difference they found was not statistically significant.
Achieved whena trial with people on both sides and trained mediators finds a significant difference in the share of disputes settled.
Claimed by
James Hale (University of Southern California) and Jonathan Gratch, in a preprint not yet peer reviewed, September 2026
Open
01
AI mediators with real people
A study in which people, not software, play the two sides of fictional disputes, with some sessions led by an AI mediator and some by trained human mediators, measuring how often each reaches agreement, how far the parties move toward each other, and how the parties rate the process.
The briefCredit · unclaimed02
Comparable counts of mediations
An open table of how many mediations take place each year by country and by kind (court-linked or private; family, labour, commercial or community), taken only from official published statistics and coded with the Council of Europe's baseline categories, with every gap marked as "not available" or "not applicable" in the way that template does.
The briefCredit · unclaimed
Open 01 · the brief
AI mediators with real people
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A study in which people, not software, play the two sides of fictional disputes, with some sessions led by an AI mediator and some by trained human mediators, measuring how often each reaches agreement, how far the parties move toward each other, and how the parties rate the process.
Why it is still open
The main test beds for AI mediators run their disputes between AI-simulated parties: Microsoft's ProMediate (arXiv 2510.25224) and SoCRATES (arXiv 2606.05563), in which the strongest of eight models closed only about a third of the gap between the parties. A 2026 study found that AI-simulated disputants behave differently from people with the same personality traits. The trials with people stop short of a full comparison: Hale and Gratch (2026) put one human against a GPT-4o opponent and used untrained crowd workers as the human mediators, and Jamie Bergen and Sarit Kraus (2026) compared an AI with professional mediators only in the preparation stage before a joint session, on the parties' own ratings. Dyspute.ai, which sells an AI mediator, had published no outcome data as of January 2026.
Built with what exists
Fictional cases written for the study, each with a private brief for each side; ProMediate's published measures of consensus change and intervention timing; an AI mediator built on a current model, including an open-weight one such as Mistral Large 4 (see the Note "A trillion open parameters"); video calls recorded with consent and transcribed with ElevenLabs Scribe v3 (see the Note "Fewer word errors"); volunteer role-players from negotiation courses, with a university ethics review.
First step
Write three fictional cases with private briefs and run six pilot sessions with classmates, three with each kind of mediator.
Done looks like
60 sessions on 10 fictional cases, with people playing both sides in every session, 30 led by the AI mediator and 30 by accredited human mediators, with agreement rates, consensus change and party ratings published for each, set beside the same cases run with simulated parties.
Open 02 · the brief
Comparable counts of mediations
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of how many mediations take place each year by country and by kind (court-linked or private; family, labour, commercial or community), taken only from official published statistics and coded with the Council of Europe's baseline categories, with every gap marked as "not available" or "not applicable" in the way that template does.
Why it is still open
The Council of Europe's justice-efficiency commission (CEPEJ) compiled one table, of civil mediations in 2016 across its member states, and its mediation working group wrote that "in most cases" the figures were expert estimates by national correspondents, not official statistics. The same group set a target that by 2025 disputes settled by mediation should equal at least 25% of those decided in court, which nobody can check without yearly counts. CEPEJ wrote that "in some Member States such statistics are not collected at all" and published a template rather than a table, and the European Parliament's 2014 figure of less than 1% also came from expert estimates. 493042_EN.pdf)
Built with what exists
The CEPEJ baseline grid; national statistics such as Italy's Ministry of Justice mediation figures, China's Ministry of Justice yearly totals, FMCS annual reports and HM Courts and Tribunals Service data; Python and a spreadsheet; a language model to pull figures out of PDF reports, with every figure checked by hand against its page.
First step
Fill in the grid for five places with known official figures (Italy, England and Wales, China, US federal labour mediation and Singapore) and publish it for comment.
Done looks like
At least 25 jurisdictions, with yearly counts from 2015 to 2025 wherever one is published, every cell linked to its source, released as a CSV with a list of countries that publish no count.