AI in Law: The Unlearning Challenge and Ethical Implications

Many of my colleagues have begun utilising AI models like ChatGPT to draft various legal documents. Emerging technologies are increasingly assisting legal professionals in creating drafts more efficiently and swiftly. However, while engaging with these generative AI models, the personal details of clients are often shared, which, although seemingly harmless at first glance, raises significant ethical implications for legal professionals.

Much of the information an advocate handles falls within the realm of confidential data. The situation is further complicated when this data, which is personally identifiable, is fed into AI systems. This could potentially lead to professional misconduct if the information is compromised, raising ethical concerns for the legal profession.

The advent of technology in this field is inevitable and, in many ways, beneficial. It enables advocates to reduce client costs in the long run, making legal services more accessible and affordable. However, one of the significant challenges with these platforms is their inherent learning mechanism. Each time a user employs the program to generate a document or outcome, they are simultaneously feeding information into the system for potential future use.

Unlike traditional database models, where you can delete a specific entry, and the information is forgotten, AI models learn and store information differently. The data is not stored in a traditional sense but is incorporated into the model's learned patterns and behaviours. This makes the concept of simply "forgetting" the information more complex and technically challenging.

Machine unlearning, the process of making AI systems forget specific data points, is one of the biggest challenges currently faced by tech giants. While these models can quickly learn new information, unlearning presents a significant hurdle. Feeding these models confidential information could have long-term disastrous effects if not properly managed.

Therefore, it is crucial at this juncture to use these platforms judiciously. This could involve:

1. Thoroughly read a platform's terms and conditions before using it to understand how it uses and processes the input data.

2. Using generic terms such as "Party 1" and "Party 2" instead of actual party names while generating documents to ensure the information does not remain personally identifiable.

It is also incumbent on Bar Councils to frame appropriate guidelines for using these platforms. At the same time, tech developers need to understand the legal position and limitations while processing data and strive to create models that can address these concerns effectively.

Let's strive for a safer, more ethically sound world!