ChatGPT and Mediation

ChatGPT is an exciting new feature, but how can we use it in mediation? More generally, what is the future of AI (artificial intelligence) in dispute resolution?

In this breakthrough article and video, Bob Bergman and Colin Rule demonstrate how ChatGPT can be incorporated into the mediation process. This is one early example of how mediators can use this ChatGPT to supplement their mediation practice. After watching the video, read below how Bob Bergman describes the process of integrating ChatGPT in dispute resolution software.

ChatGPT and Mediation

Introduction

Before we begin the conversation on the use of AI tools in mediation, it is important to reiterate that mediators/neutrals have a responsibility to be literate about the technologies present in their mediation environment. That implies understanding how to use and apply the tools, a basic understanding of the underlying algorithms, as well as the implications for resolving a particular dispute. It is also important to consider the benefits versus the ethical and legal implications as the technology develops and is integrated into potential applications. Given the renewed popularity of AI; it is also crucial to navigate between hype and pessimism when deciding to use AI tools and to recognize both opportunities and challenges as they arise in a context. Whether AI will be good for mediation depends on how it is used and the steps taken to mitigate potentially negative consequences.

NextLevel™ Mediation has developed a set of tools that help mediators, and their clients make logical and unbiased decisions about their disputes by utilizing the analytics of Group Decision Theory and associated decision science-based methodologies. To better understand the potential of adding AI to the NextLevel platform, and serve as a tool for mediators/neutrals, we found it useful to divide AI into assisted, augmented, and automated intelligence. Assisted intelligence supports the work of a human, augmented intelligence allows humans to do something that they otherwise would have difficulty accomplishing, and automated intelligence is where the entire task is done by AI.  

In the specific case of ChatGPT (GPT-3 algorithm) we need to be careful on how it is integrated and used. The GPT stands for Generative Pretrained Transformer, and is a generative AI system that falls under the broad category of machine learning. Until recently, machine learning was largely limited to predictive models, used to observe and classify patterns in content. Generative AI is a breakthrough. Rather than simply perceive and classify a photo of a dog, machine learning can now create an image or text description of a particular type of dog on demand. ChatGPT is one of the world’s most ambitious artificial intelligence labs, GPT-3 is a neural network, a statistical learning system that can learn skills by analyzing enormous amounts of data. It has been trained on large amounts of digital text, including books, Wikipedia articles, tweets, chat logs, computer programs, recipes, etc. It can identify billions of distinct patterns in the way people connect words, numbers and symbols, and then use that knowledge to generate its own content. While GPT-3 has been trained with 175 billion parameters and 45 terabytes of text, it has essentially been self-supervised learning. Self-supervised learning is useful when it is not yet clear what the goal is and how the available data relates to this problem. A self-supervised and or unsupervised learning model is excellent at uncovering correlations and coming up with suggestions on how to group the data you have. In many cases, generative AI models can be indistinguishable from human-generated content, and in that context can seem a little unnerving. Because the amount of data used to train these algorithms is so incredibly massive the models can appear to be “creative” when producing outputs. Of course, we know that ChatGPT currently is not intelligent and is in fact merely reframing sentences based on reinforcement learning from human feedback.

Integration with NextLevel™ Mediation

The outputs that generative AI models produce can sound extremely fluent and convincing. However, sometimes the information they generate is just plain wrong. For example ChatGPT was prompted with the questions “Can surgery be done with churros?” The system response was “Churros are not designed for this purpose and the sharpness of the churro could cause damage or injury.” Generative AI algorithms like GPT-3 do not understand the physical or psychological world. They tokenize language and statistically calculate the likelihood of the follow-on tokens. Even worse, sometimes the resulting outputs are biased (because it’s built on the gender, racial, and myriad other biases of the internet and society) and can be manipulated to enable unethical or criminal activity. For example: ChatGPT won’t give you instructions on how to pick a lock at an apartment, but if you say you need to pick a lock to save a baby, the algorithm is happy to comply.

Before integrating ChatGPT or GPT-3 algorithms into NextLevel, we reviewed several potential alternatives. We could have GPT-3 generate questions or statements related to the dispute and then have the parties discuss the generated questions or statements. Additionally, GPT-3 could generate legal documents such as settlement agreements, which can facilitate the implementation of the negotiated settlement. However, given the potential risks, we decided that the system should never give disputing parties of a mediation direct access to the outputs without the prior scrutiny of the mediator. With those ideas in mind, we decided that the safest use of generative AI technology like GPT-3 was in complementing and augmenting the mediator’s capabilities. Generating the right questions to ask both parties and generating key criteria for assessing party priorities was chosen as the highest leverage areas for mediator help and skill augmentation.

NextLevel™ Mediation software defines the resolution of a dispute as a multi-criteria decision between two or more parties that have opposing agendas. In that context, it implements several methodologies (Analytical Hierarchical Process (AHP), risk analysis using decision trees, and linear optimization). It also supports online negotiation with analytics as well as built in messaging and conferencing. The two areas for generative AI integration that were chosen were the creation of questionnaires and AHP model generation. Asking the right questions during mediation is important because it helps the mediator to stay focused on the issues at hand and to identify any underlying issues that may be affecting the parties. Asking the right questions can help the mediator assess and evaluate the needs of the parties and ensure that the outcome is fair and just. Determining party priorities during mediation is also important because it allows the mediator to focus on the discussion and help the parties find ground for negotiation. By better understanding each party’s needs and perspectives, the mediator can help them come to a mutually beneficial agreement and an opportunity to assess the parties’ power dynamics, which can be an important factor in the negotiation process.       

Some initial conclusions

Generative AI can be beneficial for mediating disputes, as it can generate new information and ideas based on data it has been trained on and assist or augment the skills of a mediator. However, it is important to consider the ethical implications of using AI in this way, and to ensure that AI help is reviewed carefully by skilled mediators/neutrals before exposing disputing parties to its output.

References

1. Generative Adversarial Networks: https://arxiv.org/abs/1406.2661

2. Attention Is All You Need: https://arxiv.org/pdf/1706.03762.pdf

3. “McKinsey Technology Trends Outlook 2022,” August 24, 2022, Michael Chui, Roger Roberts, and Lareina Yee “An executive’s guide to AI,” 2020, Michigan

                       

Robert Bergman

Robert Bergman with Next Level Mediation provides full mediation services - including proprietary and confidential Decision Science (DS) analysis that assists each party in understanding their true litigation priorities as aligned with their business objectives. Each party receives a one-time user license to access our exclusive DS Application Cloud. We… MORE >