We are running another conference next April (TAIS 2025), and have just put out our call for papers. If you work in technical safety, philosophy or governance and want to contribute to building safety research capacity in Japan, please send a paper our way!
Important Dates
Submissions open: October 11, 2024
Submission Deadline: January 1, 2025
Notification of Acceptance: February 1, 2025
Camera-ready Deadline: April 1, 2025
Workshop Date: Saturday, April 19, 2025
All deadlines are 11:59PM UTC-12:00 (“anywhere on Earth”).
Call for papers
We are inviting submissions of short papers (maximum 8 pages) outlining new research, with a deadline of January 1, 2025. We welcome papers on any of the following topics, or anything else where the authors convincingly argue that it advances the field of AI safety in Japan.
Mechanistic Interpretability: the detailed study of particular neural networks. Interpretability researchers often take a circuits-style approach following the foundational work of Chris Olah, or use causal linear probes to understand directions in latent space. Work that investigates how and why structure emerges in neural networks during training can also fall under this category.
Developmental Interpretability: understanding the process by which neural networks learn. By applying the tools of Watanabe’s singular learning theory to modern neural network architectures, developmental interpretability researchers aim to detect phase transitions during training, thereby detecting new capabilities as they emerge.
Agent Foundations: the name given to a number of parallel research efforts, each of which aims to designing AI that is provably safe. Problems in agent foundations include embedded agency and cognition, corrigibility and the shutdown problem, natural abstractions, and game theory. With sufficient rigour, agent foundations research hope to build AGI that is safe by construction.
Scalable Oversight: a prosaic alignment approach that aims to make advanced AI systems amenable to human oversight. Often this involves training less powerful AIs to oversee more powerful AIs in a hierarchical way with a human at the root, as in Paul Christiano’s Iterated Amplification. Other scalable oversight approaches aim to make even the largest AIs directly overseeable.
ALIFE: a broad theme grouping approaches that understand artificial life as it relates to natural life. Often, as in active inference or collective intelligence, this involves replicating natural systems in silico such that we can understand and improve on them. Other ALIFE approaches overlap with the agent foundations research agenda, as in the study of emergent agentic phenomena in cellular automata.
Artificial Consciousness and Posthuman Ethics: thinking seriously about whether machines can be conscious or worthy of moral patienthood, including determining whether current systems are conscious or the intentional construction of conscious artificial systems. Works that address the ethical treatment of machines by humans and humans by machines also falls into this category.
AI Governance: research into how frontier AI systems can be effectively deployed to benefit Japan and the rest of the world and what regulation is necessary or sufficient to ensure the safety of the public, from both regular and existential risks.
We encourage you to consider submitting to TAIS 2025 even if you think your work might not be relevant; irrelevant submissions are not a substantial burden on our editorial team, and our editors like reading oblique and hard-to-categorize research.
We also welcome position pieces bringing clarification to complex topics (especially topics not prominent in the Japanese research community), discussing the need for research in overlooked areas, or pointing out structural issues that present obstacles to those who would otherwise do good research.
Submissions are non-archival. We are happy to receive submissions that are also undergoing peer review elsewhere at the time of submission, but we will not accept submissions that have already been previously published or accepted for publication at peer-reviewed conferences or journals. Submission is permitted for papers presented or to be presented at other non-archival venues.
Most accepted papers will be offered a poster presentation. In rare cases we might ask the authors to lead a workshop (1-2 hours). If you would like to be considered for a workshop, please indicate as such on your submission form.
TAIS 2025 is not peer-reviewed. Our team of editors will review each submission and decide whether to accept a paper as a poster, invite the authors to lead a workshop, or reject. Evaluation of submissions will be based on the originality and novelty, the technical strength, and relevance to the workshop topics. Notifications of acceptance will be sent to applicants by email.
Tokyo AI Safety 2025: Call For Papers
Link post
Last April, AI Safety Tokyo and Noeon Research (in collaboration with Reaktor Japan, AI Alignment Network and AI Industry Foundation) hosted TAIS 2024, an AI safety conference in Tokyo, Japan. You can read more about that conference and how well it went here.
We are running another conference next April (TAIS 2025), and have just put out our call for papers. If you work in technical safety, philosophy or governance and want to contribute to building safety research capacity in Japan, please send a paper our way!
Important Dates
Submissions open: October 11, 2024
Submission Deadline: January 1, 2025
Notification of Acceptance: February 1, 2025
Camera-ready Deadline: April 1, 2025
Workshop Date: Saturday, April 19, 2025
All deadlines are 11:59PM UTC-12:00 (“anywhere on Earth”).
Call for papers
We are inviting submissions of short papers (maximum 8 pages) outlining new research, with a deadline of January 1, 2025. We welcome papers on any of the following topics, or anything else where the authors convincingly argue that it advances the field of AI safety in Japan.
Mechanistic Interpretability: the detailed study of particular neural networks. Interpretability researchers often take a circuits-style approach following the foundational work of Chris Olah, or use causal linear probes to understand directions in latent space. Work that investigates how and why structure emerges in neural networks during training can also fall under this category.
Developmental Interpretability: understanding the process by which neural networks learn. By applying the tools of Watanabe’s singular learning theory to modern neural network architectures, developmental interpretability researchers aim to detect phase transitions during training, thereby detecting new capabilities as they emerge.
Agent Foundations: the name given to a number of parallel research efforts, each of which aims to designing AI that is provably safe. Problems in agent foundations include embedded agency and cognition, corrigibility and the shutdown problem, natural abstractions, and game theory. With sufficient rigour, agent foundations research hope to build AGI that is safe by construction.
Scalable Oversight: a prosaic alignment approach that aims to make advanced AI systems amenable to human oversight. Often this involves training less powerful AIs to oversee more powerful AIs in a hierarchical way with a human at the root, as in Paul Christiano’s Iterated Amplification. Other scalable oversight approaches aim to make even the largest AIs directly overseeable.
ALIFE: a broad theme grouping approaches that understand artificial life as it relates to natural life. Often, as in active inference or collective intelligence, this involves replicating natural systems in silico such that we can understand and improve on them. Other ALIFE approaches overlap with the agent foundations research agenda, as in the study of emergent agentic phenomena in cellular automata.
Artificial Consciousness and Posthuman Ethics: thinking seriously about whether machines can be conscious or worthy of moral patienthood, including determining whether current systems are conscious or the intentional construction of conscious artificial systems. Works that address the ethical treatment of machines by humans and humans by machines also falls into this category.
AI Governance: research into how frontier AI systems can be effectively deployed to benefit Japan and the rest of the world and what regulation is necessary or sufficient to ensure the safety of the public, from both regular and existential risks.
We encourage you to consider submitting to TAIS 2025 even if you think your work might not be relevant; irrelevant submissions are not a substantial burden on our editorial team, and our editors like reading oblique and hard-to-categorize research.
We also welcome position pieces bringing clarification to complex topics (especially topics not prominent in the Japanese research community), discussing the need for research in overlooked areas, or pointing out structural issues that present obstacles to those who would otherwise do good research.
Reviewing and Submission Policy
Please submit your paper using the form on our website.
Submissions are non-archival. We are happy to receive submissions that are also undergoing peer review elsewhere at the time of submission, but we will not accept submissions that have already been previously published or accepted for publication at peer-reviewed conferences or journals. Submission is permitted for papers presented or to be presented at other non-archival venues.
Most accepted papers will be offered a poster presentation. In rare cases we might ask the authors to lead a workshop (1-2 hours). If you would like to be considered for a workshop, please indicate as such on your submission form.
TAIS 2025 is not peer-reviewed. Our team of editors will review each submission and decide whether to accept a paper as a poster, invite the authors to lead a workshop, or reject. Evaluation of submissions will be based on the originality and novelty, the technical strength, and relevance to the workshop topics. Notifications of acceptance will be sent to applicants by email.