Human-centered evaluation of explainable AI applications: a systematic review

Artificial Intelligence And The Rule Of Law - UNESCO

The judiciary plays an important role in the governance of Artificial Intelligence by applying international human rights standards to the ethical concerns related to bias, discrimination, privacy, and transparency, while also leveraging AI systems to strengthen access to justice and enhance the efficiency of judicial administrations.

Operating in over 160 countries, UNESCO's Judges Initiative offers comprehensive and practical training tools to members of the judiciary, in order to strengthen knowledge and capacities on regional and international standards on Artificial Intelligence and the Rule of Law, Freedom of Expression, and Access to Information.

Institutional Capacity Building

Based on UNESCO's Global Toolkit on AI & the Rule of Law, both in-person and virtual training workshops, including training of trainers, are organized worldwide. The Toolkit provides a curriculum that can be adapted by national judicial training institutions, universities and other legal education organisations to offer trainings.

Network of Experts

UNESCO has established a Global Network of Experts on AI & the Rule of Law, that provides technical assistance and trainings to judiciaries worldwide. The diverse group of experts bring an inter-disciplinary perspective, both as academics and practitioners to support the responsible adoption and governance of AI technologies in judicial systems globally.

The results of the UNESCO survey will form the basis of the publication of the UNESCO Guidelines for the Use of AI Systems in Courts and Tribunals.

AI and the Rule of Law: Capacity Building for Judicial Systems

How The JAMS Artificial Intelligence Rules Will Improve Dispute ResolutionJAMS - JDSupra

With deep experience in both AI and dispute resolution, we are the co-creators of the JAMS Artificial Intelligence Disputes Clause, Rules and Protective Order (AI Rules). We created these rules to address some of the challenges posed by AI-related disputes. Tailored rules can result in fairer outcomes that are faster and cheaper. JAMS is already seeing a growing number of AI-related disputes, and this will continue as both AI capabilities and industry adoption grow.

AI development and use can involve complex ecosystems and numerous stakeholders, such as parties building AI models (including open-source models), training models (or generating, collecting or curating training data), integrating models into broader systems or with sophisticated hardware, or using AI systems directly (as individuals or enterprises) or licensing platforms. Problems can arise at all stages between different types of stakeholders, and complex systems can break in complex ways. For example, a dispute may arise between a generative AI system provider and an individual user if the user is sued for copyright infringement by a third party based on AI-generated content. Or a platform provider may be sued by an enterprise licensee for providing a system that allegedly violates privacy laws and regulations such as the EU AI Act.

AI systems have the potential to result in massive social benefits and value, but they also involve significant risks. When problems happen—and they will happen—it is important to be able to resolve disputes in a fair, efficient and just manner. Unfortunately, conventional litigation may not achieve that type of outcome and may involve extensive delays, high discovery costs and burdens, and the loss of confidentiality.

The AI Rules are designed to further improve the alternative dispute resolution (ADR) process in the context of these specialized cases. The rules are not intended to address the use of AI in the process of dispute resolution itself (e.G., neutral and advocate use of generative AI systems); JAMS has developed separate internal guidelines for the use of AI in ADR. The AI Rules also include a broad and functional definition of AI, which is important because there are many different types of AI and damages caused by AI do necessarily depend on specific features, such as how an AI is structured (e.G., machine learning vs. Expert systems), how much data the system is trained on and how much computing power is applied in training. It is up to the parties to decide whether their relationship is one that could benefit from the AI Rules.

Experienced and Knowledgeable Neutrals

Rule 15(b) of the AI Rules provides that "JAMS shall propose, subject to availability, only panelists approved by JAMS for evaluating disputes involving technical subject matter with appropriate background and experience. JAMS shall also provide each Party with a brief description of the background and experience of each Arbitrator candidate."

There is a long-standing debate in ADR circles about whether it is better to pick a neutral who is a generalist or a subject matter expert. But the need for a subject matter expert is more pressing when disputes involve technologically specialized subject matter. Using a JAMS prequalified neutral makes it more likely the arbitrator will quickly understand the issues, which in turns saves the parties time and expense, and it increases the likelihood that an arbitrator will get to the right answer.

Built-in Confidentiality and Protections

Rule 16.1(a) provides that, unless the parties agree to another form of protective order, the AI Disputes Protective Order automatically applies to protect confidential information.

It is important to have a protective order in place given that AI-related disputes often involve confidential information. Using a predetermined and independently vetted protective order helps save time and money by removing the need for parties to negotiate a separate order. It also eliminates a potential dispute. Finally, having a protective order in place from the beginning expedites resolution by removing a potential roadblock to early information exchange.

Rule 25(a) provides for confidentiality in the arbitration: "All Parties to the Arbitration and their counsel shall strictly maintain in confidence all details of the Arbitration and the Award, including the Hearing, except as necessary to participate in the Arbitration proceeding and the Hearing, in connection with a judicial challenge to or enforcement of a Decision, or unless otherwise required by law or judicial decision."

Confidentiality related to AI systems is often a key consideration for stakeholders, and it is one of the key advantages of arbitration compared to litigation. However, not all rules require that the above aspects of an arbitration be kept confidential.

Specialized Expert and Discovery Procedures

Rule 16.1(b) provides a specialized process for technical experts to review materials: "The production and inspection of any AI systems or related materials, including, but not limited to, hardware, software, models and training data, shall be limited to the Disclosing Party making such systems and materials available to one or more expert(s) in a secured environment established by the Disclosing Party. The expert(s) shall not transmit or remove any produced materials or information from such environment."

The Rule further provides for the optional appointment of an arbitrator-selected independent neutral: "If jointly requested by the Parties, the Arbitrator shall designate expert(s) to inspect AI systems or related materials. In which case, the Arbitrator shall first attempt to designate such expert(s) from a list of third-party experts maintained by JAMS, subject to availability and appropriate qualifications. All costs related to the use of Arbitrator appointed expert(s) shall be borne equally by the Parties, although the Arbitrator may shift fees at the Arbitrator's sole discretion, including in the Final Award. Expert testimony from an Arbitrator appointed expert shall be limited to a written report requested by the Arbitrator addressing questions posed by the Arbitrator, and testimony at the Hearing of such expert(s)."

These rules are important for several reasons. First, they reduce the risk of loss of confidential information and trade secrets inherent in traditional arbitration discovery by ensuring review in a secured environment. The value of AI-related trade secrets may otherwise dwarf the amount in dispute in an underlying matter, making discovery without these protections a risky prospect.

The rules also provide for the arbitrator to directly designate technical experts vetted by JAMS. This provides critical independence as opposed to allowing party-appointed experts, avoiding concerns associated with "hired guns," and it saves time and money by avoiding a battle of the experts.

Critically, expert opinion is also streamlined and focused on key issues articulated by the arbitrator. Otherwise, perverse incentives involved in both discovery and expert analysis have the potential to result in scorched-earth tactics and costly fishing expeditions on issues not directly relevant to resolution. Where underlying systems are particularly complex and may involve massive amounts of data, there is a greater risk of undermining the core benefits of arbitration.

As technology and disputes evolve, so too will ADR need to evolve. The JAMS AI Rules make it so that at least ADR does not lag behind technology.

AI And The Rule Of Law: Capacity Building For Judicial Systems

As the use of AI technologies advances, judicial systems are being engaged in legal questions concerning the implications of AI for human rights, surveillance and liability, among others. In addition, judicial systems are also using AI systems for judicial decision-making processes that have raised concerns for fairness, accountability and transparency in decision making by automated or AI-enabled systems.

The potential of AI is already being explored by many judicial systems that include the judiciary, prosecution services and other domain specific judicial bodies, around the world, in the criminal justice field, providing investigative assistance and automating/facilitating decision-making processes.

Nevertheless, the use of AI poses a wide range of challenges to be addressed: from pattern recognition, to ethics,biased decisions taken by AI-based algorithms, transparency and accountability. Self-learning algorithms, for instance, may be trained by certain data sets (previous decisions, facial images or video databases, etc.) that may contain biased data that can be used by applications for criminal or public safety purposes, leading to biased decisions.

Considering rapid developments in this field, the challenges and opportunities related to harnessing AI in the field of justice and how AI-based systems can help judicial actors in their roles within the administration of justice and to handle cases involving AI that impacts human rights must form part of discussions among stakeholders from the judicial ecosystem.

---