A current critique in Taiwan: generative A.I.-produced legal documents

 September 26, 2023

Should the legal profession use generative AI to produce legal documents?  That’s the question that Taiwan is ruminating on this week, after the Judicial Yuan announced it would start using a bespoke A.I. model trained on court documents to generate draft decisions on cases of fraud, money laundering or unsafe driving.  That’s according to reporting by the Taipei Times.  A number of groups including the Taiwan Bar Association, Judicial Reform Foundation, the Taipei Bar association and the Taiwan Association for Human Rights questioned the move.  In particular, they brought three points to bear:

1. The system is questionable if it generates draft rulings from judgements and indictments and not from relevant files in the cases as well.
2. The public has the right to know about the contractor who built the system, in particular how it avoids problems, inadequate or biased information, and how the contractor and the Judicial Yuan would settle liability issues in the case of a dispute. 
3. Attorneys, prosecutors and other stakeholders should be allowed to try to access the system and this should allow the public to examine the biases and risks.  

The report goes on to note that AI reproduces structural biases regardless of cultural context between Taiwan and the U.S.  In particular, they note that the group’s research revealed “technology experts have warned that the extensive use of A.I. in the U.S. justice system, particularly in helping judges in sentencing based on risks of defendants becoming repeated offenders, might reinforce the prejudice and racial inequality that already exist in the justice system.”

One does not simply open up lines of communication.  That paraphrase is only here to say that there exist huge divisions in systems between the U.S. and Taiwan.  Common sense says to look at the underlying means and methods of reproduction here.  Taiwan is the spot in the global economy where most of the underlying structure of the world tech economy is laid.  The Taiwan tech economy, which makes the single most important component of any computing device, the silicon chip, stands to gain in a relatively large amount by any technology advance, even A.I.  The structural biases that we will see emphasized by A.I. in Taiwan and other countries will be different than the structural biases that would be emphasized in America: those would be the structural inequalities of this country and possibly its common-law cousins in Europe and the South Pacific, we should be clear about that, so the issue here is one of technology to anyone who acts or organizes in a common law system.  That being said there is an A.I. arms race in the case of Taiwan.  Chinese propaganda is getting their hands on A.I. as we speak.  A stark warning was published by Taiwan observer Courtney Donovan Smith in Taiwan News in March 2023 counseling about a type of A.I. that can “write passable articles and even entire books with limited prompts very quickly, opening the floodgates to fill fake news sites and social media accounts with an unimaginable volume of content,” while also Taipei Times commented that “a source with insight into the matter said that China’s use of algorithms in propaganda would likely increase, making it more difficult for users to distinguish fake stories from real ones in the future.”  So there is, arguably, an undercurrent of Taiwan Strait geopolitics at stake in A.I. adoption by the Taiwanese judiciary.  If the trend of A.I. LLMs fragmenting into smaller bespoke models can be traced by you here, 

1. the training data,
2. the contractual relations between the software engineer and the operator,
3. the transparency and accountability of the system,

are all technical considerations for critiquing the use of A.I. by government agencies. 

There is structural bias in every one of these three points, not only training data, which is obvious, but also in the characteristics of the generative system and what it was built to do, and any modifications that were made to avoid certain outcomes.  All of it can be laid open to critique.  Will Taiwan confront a faceless inquisition? — Will the U.S. face a massive increase in police powers? — And all because of A.I.?