[Topic-2] AI出力に対する規制 / Regulation of AI output

本研究では，公共空間をAI接続カメラで監視するシステムを例に，プライバシー保護の具体策を検討しています．このシステムの利活用目的として，行方不明の子供の追跡，通学路の子供の見守り，異常事態の検知，犯罪やテロ行動の分析を挙げています．提案する「Verifiable record of AI output」は，AIをOutbound-Firewallで包み込み，社会的議論で事前承認された利用目的に適合する場合に限り，AI出力を許諾します．さらに，こうしたAI出力の判定過程・出力内容を検証可能で改ざん不可能な形で記録・公開し，不正利用を防ぐ仕組みです．この仕組みを適用することで，AIシステムの透明性と市民の信頼性を確保します．また，他のAIシステムへの展開可能性も示唆しています．本研究の成果は，AI & Society誌に掲載されました[3]．また，本論文に基づく研究計画が立てられ，日本側（群馬大学）とシンガポール側（南洋理工大学）との国際共同研究プロジェクトの予算申請*がされている．（採択された場合，2025年4月研究開始．）

In this study, we explore specific measures for protecting privacy using a system that monitors public spaces with AI-connected cameras as an example. The intended uses of this system include tracking missing children, monitoring children on school routes, detecting abnormal situations, and analyzing criminal or terrorist activities. The proposed "Verifiable Record of AI Output" encapsulates AI within an outbound firewall, allowing AI outputs only for purposes that have been pre-approved through societal discussions. Furthermore, this mechanism records and discloses the decision-making process and output details of the AI in a verifiable and tamper-proof manner, thereby preventing misuse. By implementing this system, we aim to ensure the transparency of AI systems and foster public trust. Additionally, the study suggests the potential for expanding this mechanism to other AI systems. The results of this research were published in AI & Society [3]. Based on this paper, a research plan has been developed, and a budget application for an international collaborative research project between the Japanese team (Gunma University) and the Singaporean team (Nanyang Technological University) has been submitted. (If approved, the research will commence in April 2025.)

[3] Yusaku Fujii, “Verifiable record of AI output for privacy protection: public space watched by AI-connected cameras as a target example”, AI & Society (2024).
https://doi.org/10.1007/s00146-024-02122-8

本論文では，あらゆる地点が1つ以上のAI接続街路灯カメラでモニタされる広範な公共空間 「AI見守り公共空間」（人類未経験）の利点と欠点を検証しました．具体例として，誘拐された子供や徘徊する高齢者の位置特定，路上犯罪者の追跡と逮捕など，社会安全性の飛躍的向上が挙げられます．また、街路灯交換時にカメラ内蔵型街路灯を導入することで、低コストでの実現可能性を指摘しました．

This paper examines the advantages and disadvantages of an extensive public space, termed the "AI-Monitored Public Area" (an unprecedented concept for humanity), where every location is monitored by one or more AI-connected streetlight cameras. Specific examples of its benefits include significant improvements in social safety, such as locating abducted children or wandering elderly individuals and tracking and apprehending street criminals. Additionally, the study highlights the feasibility of implementing such a system at low cost by introducing streetlights with built-in cameras during routine streetlight replacements.

一方で，個人のプライバシー侵害や監視社会化のリスクも大きな課題とされています．これに対する対策として，本研究では「Verifiable record of AI output」という新しいプライバシー保護の概念を提案しました．この概念は，AI出力の透明性と信頼性を確保する仕組みであり，民主主義国家における「AI見守り公共空間」の受容性を高めることを目指しています．「Verifiable record of AI output」は，Big Dataなどの膨大なAI inputの匿名化や，ブラックボックス化しているAI内部プロセスの制御などよりも，AI outputだけに対してルール適合性確認のみの単純な出口規制をかける方が，現実的ではないか，というアイデアに基づきます．

On the other hand, significant challenges include the risk of personal privacy infringement and the potential for a surveillance society. To address these concerns, this study proposes a new concept for privacy protection called the "Verifiable Record of AI Output." This concept is a mechanism to ensure the transparency and reliability of AI outputs, aiming to enhance the acceptability of the "AI-Monitored Public Area" in democratic societies. The "Verifiable Record of AI Output" is based on the idea that a simpler approach focused solely on exit regulations—verifying compliance of AI outputs with predefined rules—may be more practical than attempting to anonymize vast AI inputs, such as Big Data, or control the opaque internal processes of AI systems. This approach seeks to strike a balance between leveraging AI's benefits and addressing privacy concerns.

さらに，本研究は「AI出力の検証可能記録」の社会安全性への効果とプライバシー侵害の懸念を検証するプロセスを通じ，新たな技術や発明，特許の創出を期待しています．この試みが社会に受け入れられた場合，本研究は世界に大きな貢献をもたらす可能性があります．

Furthermore, this study anticipates the creation of new technologies, inventions, and patents through the process of evaluating the impact of the "Verifiable Record of AI Output" on social safety and addressing concerns about privacy infringement. If this approach gains societal acceptance, the research holds the potential to make a significant contribution to the world.

2025年は，現在，申請中予算（＝シンガポールとの共同研究）が採択された場合は，申請内容に従い，精力的に研究を進めたいと思います．もしも，採択されなかった場合は，次の予算申請のチャンスを待ちながら，以下を行います．
[a] コンセプト自体の深化・具体化．
[b] コンセプト自体に関する論文発表
[c] [b]に関するプレス発表，世界主要メディアのOp-Edでの発表．

In 2025, if the currently applied-for budget (for the joint research with Singapore) is approved, we plan to actively advance the research in accordance with the application details. If the application is not approved, while waiting for the next opportunity to apply for funding, we will focus on the following tasks:
[a] Refining and concretizing the concept itself.
[b] Publishing academic papers related to the concept.
[c] Issuing press rel