The Pitfalls of Mass-Produced Pubic Commentary - Why AI is Needed to Protect Taxes and Democracy
The following manuscript
(title here)
(Heading here)
Recently, the number of "mass-produced pub-comments" has been increasing rapidly. For the February 2025 Basic Energy Plan, it was reported that only 46 people submitted 3,940 opinions (an average of about 86 per person), many of which were created using generative AI.
(Heading here)
The call for public comments on the draft ministerial ordinance on the reuse of decontaminated soil (Ministry of the Environment) (period of call: January 17 to February 15, 2025) received an unprecedented number of approximately 200,000 comments. Behind this extraordinary number was an organized call and posting instructions on social networking sites. The "Association Against the Spread of Radioactivity," a citizens' group that opposes the nationwide reuse of contaminated soil from the nuclear power plant accident, and others took the lead in organizing a slogan "Please stop reusing radioactively contaminated soil! On Twitter, posts with hashtags such as "#Let's make public comments together" and "#radioactive materials #environmental pollution" (see above) shared the link to the Ministry of the Environment's call page, the deadline for submitting opinions, and a model sentence (sample sentence) for submitting opinions. The example of the model sentence was also shared. As a result, the Ministry of the Environment received 207,850 comments, of which approximately 96% were completely identical down to the punctuation.
Due to the large number of copy-and-paste submissions of the same text, the opinions that were identical in content were combined into one and the number of unique opinions was reported to be 8277.
Normally, less than 10 comments are received in a public comment period, but this time the number of comments reached 207,850. Of these, 199,573 (96%) were exactly the same as one of the remaining 8,277, showing the systematic mobilization that resulted.
There were several confirmed cases of a single person posting more than 1,000 times using the same name, clearly an organized and mechanical mass posting (DDoS-like sabotage). In the end, the majority of the comments were negative in nature, such as "opposition to the proposed amendment and the spread of contaminated soil.
Since public officials are obligated to review all comments regardless of the number of submitter, a huge amount of taxpayer money and staff time is being sucked up in responding to these "mass and similar comments. In order to protect the trust and financial resources of the system, it is essential to create a mechanism to quickly bundle duplicate and similar comments using AI.
1 Why it matters - the gap between "quality over quantity" institutional design and reality.
1.1 Pub. comm. is evaluated on substance.
The Administrative Procedure Act clearly states that opinions received are to be judged on the basis of content, not number. Therefore, even if you copy and paste the same sentence, it is counted as only one "opinion".
1.2 Generation AI to mass-produce "nearly identical" text.
Recently, generative AIs that can create an infinite number of similar sentences by simply putting the main points in a prompt have become widespread, and on social networking sites, "how to change the wording with sentence templates + ChatGPT" has been shared, and a single person has been observed posting dozens of such posts.
1.3 SNS mobilization spreads quickly.
A call to "submit a pubic comment" spread through hashtags and video feeds, and there have been a series of instances where tens of thousands of opinions were concentrated in a short period of time.
2 Social costs - taxes and public employee time lost
"shall give due consideration to the comments submitted." The obligation to "give due consideration to the comments submitted" forces staff to sort through an endless stream of similar sentences. (Section 42 of the Administrative Procedure Act.) The U.S. Government Accountability Office GAO warns that "processing duplicate comments is a significant administrative burden."
If an additional 100,000 bubkomes are received, an additional 208 man-days of work will be required, even if the summarizing process takes only one minute per case.
It has been pointed out that if mass-produced comments pile up, the highly specialized minority opinions may be buried, and the quality of policy may suffer.
3 International Best Practices.
In the United States, the Administrative Conference of the United States (ACUS) recommended the use of technology to combat "high volume, AI-generated and similar comments" and issued Recommendation 2021-1, which includes AI clustering and duplicate detection. Federal agencies have actually reported instances where NLP tools have automatically sorted millions of cases, reducing confirmation time by approximately 90%.
4 So we need AI countermeasures
Instead of human visual confirmation of synonyms and paraphrases, summarize similarities using AI-based similarity judging.
Instead of staff copying and pasting the data into Excel, make statistics and full-text searches available on the dashboard.
Save a few minutes per case x tens of thousands of cases = hundreds of man-days of cost and more time to pay attention to minority opinions.
The following effects can be expected if AI is utilized
1.Tax savings-Machines do the prep work, people focus on analysis.
2.Restore fairness-bundle similar opinions and make unique viewpoints stand out.
3.Maintain confidence in the system-prevent the misconception that "it's a numbers game after all".
5 Technicians are "behind the scenes".
It may not sound glamorous to hear the term "administrative DX," but it is a job that quietly protects the infrastructure of our democracy. The person who develops a system that can automatically organize large numbers and similar pubic comments is a key player in protecting the quality of taxation and decision-making for all of us.
The act of wasting taxpayers' money by mass-producing pubic comments is a serious issue that undermines social benefits, and we urge you to support the engineers and researchers who are taking on the challenge of addressing this issue with AI.
---
This page is auto-translated from /nishio/量産型パブコメの落とし穴──税金と民主主義を守るためにAIが必要な理由 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.