[ad_1]
Artificial intelligence (‘AI’), particularly Large Language Models (‘LLMs’) such as Open AI’s ChatGPT, Bing Chat and Google Bard, poses obvious risks when lawyers use them without understanding their limitations. One need not look further than recent news, where the lawyer It angered a federal judge in New York To provide a summary of the cases invented by ChatGPT.
But we shouldn’t ignore technology just because it can be misused. These kinds of tools have the potential to revolutionize the legal profession, and have spawned a slew of legal technology startups and scholarly contributions.
The potential of LLMs remains largely untapped in international criminal law (ICL). How these tools reshape the ICL will depend on its ability to navigate its unique features.
The special features of ICL
International criminal proceedings are largely guided by complex documents. The massive investigations needed to determine the occurrence of international crimes often lead to an outpouring of documentary material. LLMs are designed to ingest large amounts of data, but their effectiveness can be hampered by poor quality scans and documents in languages not commonly used in their training.
Moreover, the jurisprudence of the ICC is diverse. Each ICL institution represents a jurisdiction with its own unique repositories for storing its case law. Although there are such central groups Database of legal instruments of the International Criminal CourtNo publicly available AI tool has been specifically trained on these groups. In contrast, US attorneys can integrate AI into their domestic practices more easily through various methods such as Westlaw Edge, Lexis+, and Casetext.
ICL also faces significant data security risks. Due to the young age of LLMs, their security implications remain largely undefined, which poses potential threats. Any data breaches in a war crimes or crimes against humanity trial can have catastrophic consequences, such as the identification and targeting of victims, witnesses and others at risk.
Present and future LLMs at ICL
However, LLMs offer tremendous benefits to ICL lawyers so far, not just in the distant future. These models can change the way lawyers work if they are appropriately understood and applied (look here for a general written survey).
Hiring LLMs requires a paradigm shift from traditional search engine queries and Boolean search terms. Attorneys need to engage in dialogue with these models, and craft claims with skill that can greatly enhance the quality of responses. Upcoming article by Daniel Schwartz and Jonathan H. Choi Provides an excellent overview of how LLMs can be motivated in legal contexts.
It should also be emphasized here that I am an ICL trained lawyer and not an expert in legal technology. Right now, I’m more of an enthusiast than a skilled user of these tools. It is very likely that my description of the LLM and its potential could be improved upon by more tech savvy individuals. Those interested in the technical aspects of how LLMs can be developed should consult View this video for May 2023 By Andrei Karpathy from OpenAI.
Applications of LLMs in ICL
- Jurisprudence / Filing Abstracts
Perhaps the most compelling application of the MA in International Criminal Law is its ability to summarize jurisprudence or parts of it. The quality of the abstract depends on how the triggering questions are asked. Asking any master of mainstream law to “summarize the Ongwen Appeal Judgment on the Offense of Forced Marriage” leads to a general and unhelpful summary. But copying and pasting the relevant portions of the Appeals Chamber’s Forced Marriage Judgment and LLM Claim ‘summarizing the following passage from the Ongwen Appeal Judgment: ‘(insert paragraphs)’ leads to a much higher quality summary. Placing quotation marks around the text better allows for follow-up. Makes sense with an LLM, where additional prompts can request answers with direct citations from the passage provided.
An obstacle to further growth? Character limits. Currently, the character limit for a ChatGPT4 subscription user is 4000 characters. For Bing Chat, the character limit is 2000 characters only. at recent days Stanisic and Simatovich The IRMCT Appeals Chamber ruling is 280 pages and over 700,000 characters long… ICL rulings can be much longer than that. GPT plugins are being developed that can circumvent these limits, and there are some pdf readers available – ChatPdf, for example – that can take entire pdf files and then allow users to “chat” with them through ChatGPT.
- legal research
LLMs can answer legal research questions that go beyond summarizing a specific, well-known file. Such claims tend to provide more accurate information for simpler and widely discussed topics (“What are the elements of a common criminal act?”) than for more ambiguous information (“When is it necessary to disclose witness expenses in the ICTY?”). Dialogue is often key to obtaining a meaningful answer, as it may be necessary to ask the LLM to “develop your answer part on (separate topic)” to get more information about what really interests you.
An obstacle to further growth? hallucinations Currently LLM holders tend to confidently post inaccurate citations and even fake statuses. Not checking the results can lead to serious consequences. It is critical that we think of LLMs as a supplement to diligent practice, not as a substitute for it. One other, smaller area of growth in legal research is that ChatGPT has only been trained on internet data as of the end of 2021, so it currently doesn’t have access to the latest case law (note that Bing Chat doesn’t have that same problem).
- Drafting and editing
When properly called upon for patterns of facts and applicable law, LLMs can actually generate draft paragraphs that can serve as a starting point for drafting an ICL. They can also help improve already drafted text to improve clarity and readability. Intentional specificity with regard to style may help in this regard – asking an LLM to write part of a legal brief in the style of “William Schabas” or “ICC Judge” might produce better results than asking you to phrase something more generally.
An obstacle to further growth? The extent of the reviews. The mainstream LLMs that are now available cannot create clips with references that can be drafted without further modification. Suggested changes in word choice can cause unintentional errors when certain precision is required, and LLM students may not always be able to provide a clear citation of why they wrote the sentences they did (this in itself may be an indication that the LLM Business administration hallucinates in his answer).
Within the constraints of confidentiality and specific institutional procedures, it may be more beneficial for ICL attorneys to use the existing drafting capabilities of the LLM as inspiration before drafting a preliminary document, or as a tool to significantly improve a finished document. No matter how the LLMs are incorporated into the written work, the generated segments cannot simply be copied/pasted into a broader work without further scrutiny.
- Disclosure research and evidence collection
LLMs trained on large evidence pools can provide a wealth of information about the pool. LLMs can be asked for information from pending disclosure requests or specific facts relevant to ICL trials. They can also make it easier to obtain evidence during a trial, such as summarizing witness testimony or isolating key details.
An obstacle to further growth? Secrecy. The risk of errors in current technology makes it impossible to place sensitive information in publicly available LLMs. This means that solutions must be developed for confidential ICL datasets that inspire sufficient confidence to allow them to be used in practice. There are many tech startups developing such tools—Casetext’s new joint advisory tool is a case in point (and it can do many things besides eDiscovery)—but ICL organizations may need custom solutions to gain widespread adoption. . New Evidence Submission Initiative of the Office of the Prosecutor of the International Criminal Court – while relying on artificial intelligence and machine learning – a welcome development in this regard.
- Legal analytics
The ability of LLMs to answer very specific questions in the face of big data could yield some very interesting ‘exploratory reports’ on how ICL litigants react to particular litigation strategies. A particular judge’s rulings on disclosure violations, how a particular defense attorney cross-examines inside witnesses in past trials, what ranges of rulings the Trial Chamber might be willing to accept for a particular crime – all of these types of data omissions are technically possible with the right datasets and claim .
An obstacle to further growth? small samples. There simply aren’t many ICL trials, which can greatly limit the ability of canonical analyzes to make meaningful predictions. There is also a frequent turnover rate — ICC judges are elected for non-renewable nine-year terms, for example, which makes it very likely that judges hearing their first trial will favor the institution. Legal analyzes of international and legal law may become more useful for recurring elements of trials that can lead to larger datasets more quickly—the record of the presiding judge when questioning objections, for example—than for rare events for which the dataset will be inherently small.
Conclusion
Whatever the limits of its current use, the potential of LLM for ICL is unmistakable. They are also rapidly evolving, making predicting when or how they will be widely adopted in this field difficult. But it would be beneficial for ICL to explore how to maximize these tools and ensure their responsible use.
Photo: United Nations International Residual Mechanism for Criminal Tribunals, Arusha, Tanzania (Roman Boyd, 2019)
[ad_2]
Source link
