NTT Research has made significant strides in artificial intelligence (AI) and related technologies, as demonstrated at their annual Upgrade event in 2025. The highlight was the introduction of the new Physics of Artificial Intelligence (PAI) Group, among other critical developments.
Launch of the Physics of Artificial Intelligence Group
NTT Research introduced the Physics of Artificial Intelligence Group (PAI) to explore AI’s physical aspects, a growing trend in the industry. This initiative seeks to dissect the “black box” of AI and enhance trust and safety in AI systems. As the field of AI continues to expand, the importance of understanding the underlying mechanisms that drive these artificial systems has become increasingly apparent. The PAI Group aims to bridge these knowledge gaps by delving into the foundational physics that make AI operations possible. The establishment of the PAI Group is a strategic move by NTT Research to position itself at the forefront of this new research domain. This group represents a significant evolution from the already established Physics of Intelligence (PHI) Lab, suggesting a dedicated focus on AI’s physical underpinnings. Through its efforts, the PAI Group hopes to act as a catalyst for innovative breakthroughs in AI technology, making AI systems more transparent, reliable, and secure. The ultimate goal is to lay the groundwork for more robust AI applications that can be trusted by users across various industries.
Leadership and Vision
Led by renowned scientist Hidenori Tanaka, the PAI Group continues the interdisciplinary work of the Physics of Intelligence (PHI) Lab. Tanaka’s expertise in physics, neuroscience, and machine learning positions him perfectly to guide this pioneering initiative. Under his leadership, the group aims to make AI systems more energy-efficient and reliable, fostering harmonious human-AI collaboration. Tanaka’s vision is centered on demystifying the AI “black box” to enhance system transparency and user trust.
The PAI Group’s interdisciplinary approach builds on the foundations laid by the PHI Lab. By integrating insights from diverse fields such as neuroscience and physics, the group seeks to develop AI systems that can operate more efficiently while adhering to ethical guidelines. Ensuring that AI technologies are reliable and energy-efficient is critical in a world increasingly reliant on advanced AI systems for various applications, from healthcare to transportation. The group’s collaborative efforts with leading academic institutions will further enrich its research, bringing in fresh perspectives and expertise to tackle complex AI challenges.
Research Approach and Significant Contributions
The PAI Group’s research approach mirrors historical advancements in physics, driving forward by enhancing AI design through rigorous research. The group draws parallels between their work and past scientific endeavors, where understanding the intricate details of natural phenomena led to significant technological breakthroughs. For example, the principles of thermodynamics emerged from the development of steam engines, ultimately enabling advancements in semiconductor technologies. The PAI Group aspires to follow a similar trajectory by fostering deep comprehension of AI mechanisms to spur innovation. Significant contributions from the PAI Group and its predecessor, the PHI Lab, include notable advancements in neural network pruning and bias-removal algorithms for large language models. The group’s pruning algorithm has garnered over 750 citations in the past few years, highlighting its impact on the AI research community. Additionally, their bias-removal algorithm has been recognized by the U.S. National Institute of Standards and Technology (NIST) for its effectiveness. These contributions underscore the group’s role in advancing AI technologies while addressing pressing ethical and operational concerns.
Three-Pronged Mission
The PAI Group’s mission encompasses deepening understanding to integrate AI ethics, creating controllable AI environments, and enhancing human trust through better operations and data control. The first goal is to move beyond enforced ethical learning, aiming instead to embed ethics organically within AI systems. This involves developing a nuanced understanding of AI mechanisms to ensure that ethical considerations are inherently integrated into the design process.
Creating systematically controllable AI environments forms the second pillar of the PAI Group’s mission. Similar to experimental physics, the group seeks to create controlled settings where AI learning and prediction behaviors can be observed incrementally. This approach allows for fine-tuning AI systems in a manner that is both systematic and scientific. By breaking down complex behaviors into manageable observations, researchers can develop more precise and reliable AI models. The final aspect of the mission is to bridge the trust gap between AI and human operators. To achieve this, the PAI Group focuses on enhancing operational transparency and data control. By building AI systems that are more transparent in their operations, users can gain a better understanding of how these systems function and make decisions. This increased transparency is expected to foster greater trust and acceptance of AI technologies, paving the way for their broader adoption in various sectors.
NTT’s AI Inference Chip for Real-Time 4K Video
NTT Corp announced an advanced AI inference chip for real-time ultra-high-definition video processing. This chip is ideal for edge and power-constrained environments, significantly improving video processing capabilities. The new AI inference chip is capable of handling 4K resolution at 30 frames per second, making it an exceptional tool for applications requiring high-definition video analysis in real-time. This advancement is particularly notable in situations where traditional AI inferencing systems may struggle due to the need for video compression and high-power consumption.
The implementation of this AI inference chip is set to revolutionize various sectors, including infrastructure inspection and public safety, by enabling drones and other devices to detect objects or individuals from significant heights and distances. For instance, drones equipped with this chip can identify objects or people from up to 150 meters above the ground, far exceeding the capabilities of existing real-time AI video inference technologies. This enhancement not only improves operational efficiency but also reduces labor costs by enabling operations beyond visual line-of-sight, a critical requirement in many industrial applications.
Applications and Advantages of the AI Inference Chip
This chip’s implementation, such as in drones, enhances object detection from considerable heights and distances. Its low-power AI inferencing is poised to revolutionize sectors like public safety and live sports broadcasting. By enabling real-time processing of ultra-high-definition video, the chip can be utilized in various edge applications where traditional AI systems may fall short. The technology is particularly well-suited for scenarios where power consumption is a significant concern, ensuring that AI applications can operate efficiently without compromise.
The advantages of this AI inference chip extend beyond simple video processing. In public safety, for instance, the chip can be used in surveillance systems to monitor large crowds or critical infrastructure with unparalleled accuracy. In live sports broadcasting, the ability to analyze player movements and game dynamics in real-time could enhance viewer experiences and provide valuable insights for coaches and analysts. The chip’s innovative design addresses many limitations of current AI inferencing systems, positioning it as a game-changer in the industry.
Integration with the IOWN Initiative
Future prospects include integrating the AI inference chip within NTT’s Innovative Optical and Wireless Network (IOWN) Initiative. This integration aims to address networking infrastructure challenges through high-speed, low-latency solutions. The IOWN Initiative’s data-centric infrastructure (DCI) is geared toward overcoming modern networking hurdles such as scalability, performance limitations, and high energy consumption. By leveraging the IOWN All-Photonics Network, the AI inference chip can achieve its full potential in terms of speed and efficiency. The integration of this chip into the IOWN Initiative is expected to unlock new possibilities for AI applications across various sectors. The combination of high-speed optical networking and advanced AI inferencing capabilities promises to deliver unprecedented performance, enabling more sophisticated data processing and analysis. This development represents a significant step forward in realizing a more connected and efficient digital infrastructure, setting the stage for future innovations in AI and networking technologies.
Collaboration and Security Enhancements
NTT is also collaborating with NTT DATA to enhance the AI chip with proprietary Attribute-Based Encryption (ABE) technologies, ensuring secure data sharing integrated with existing applications and data stores. ABE technology provides fine-grained access control and policy setting at the data layer, offering robust security measures tailored to specific needs. By incorporating ABE into the AI inference chip, NTT aims to address the growing concerns surrounding data security and privacy in modern AI applications. This collaboration underscores NTT’s commitment to developing secure AI solutions that can be trusted by users across various industries. The integration of ABE technology ensures that data is securely managed and shared, preventing unauthorized access and enhancing overall system integrity. With the increasing reliance on AI-driven systems and the growing importance of data security, this joint effort with NTT DATA represents a proactive approach to addressing these critical challenges.
Broader Vision and NTT’s Commitment
NTT’s broader vision of a data-driven society is detailed in The Identity of IOWN, a book by Akira Shimada and Katsuhiko Kawazoe. The book outlines the potential transformations that the IOWN Initiative can bring to society using innovative optical and wireless technologies. These breakthroughs underscore NTT Research’s commitment to innovation, scientific advancement, and addressing critical challenges in AI technologies. By fostering a deeper understanding of AI systems and their underlying mechanisms, NTT aims to create a more sustainable and secure technological landscape. The vision articulated in The Identity of IOWN highlights the transformative power of advanced AI and networking technologies. The book explores how these innovations can contribute to a more connected, efficient, and data-driven society. This broader vision aligns with NTT’s ongoing efforts to push the boundaries of AI research and development, ensuring that technological advancements serve the greater good and address pressing societal needs.
Conclusion
NTT Research showcased impressive advancements in artificial intelligence (AI) and related technologies during their annual Upgrade event in 2025. The event highlighted major breakthroughs and introduced new initiatives, particularly the formation of the Physics of Artificial Intelligence (PAI) Group. This new group aims to explore the intersection of AI and physics, pursuing innovative approaches that could revolutionize the field. The PAI Group’s creation signifies NTT Research’s commitment to pioneering in AI, embracing both theoretical and practical aspects. This addition to their portfolio reflects a broader trend of interdisciplinary efforts in AI research, where understanding the fundamental principles of physics can lead to more sophisticated and efficient AI systems.
Other notable developments at the event included advancements in machine learning algorithms, improvements in AI-driven data analysis, and enhancements in AI’s applications across various industries. NTT Research is clearly at the forefront of pushing the boundaries of AI, setting new standards for technological innovation and integration.
By combining expertise in physics and artificial intelligence, the PAI Group’s work is poised to make significant contributions to the future of AI, potentially leading to groundbreaking discoveries and applications that could redefine how AI technologies are developed and utilized globally.