The Laws Of Automation, NTT Research Details 'Physics Of AI' Group

Ensuring AI safety requires establishing robust guardrails and developing explainable AI services free from bias and hallucinations. Intelligence must be not only smart but also trustworthy and accountable.

NTT Research's newly formed Physics of Artificial Intelligence Group, a spin-off from its Physics & Informatics Lab, aims to establish fundamental laws governing AI physics, economics, motion, and mathematics. Led by Dr. Hidenori Tanaka, a specialist in physics, neuroscience, and machine learning, the group employs an interdisciplinary approach to understanding AI.

Unveiling AI's Inner Workings

NTT Research President and CEO Kazu Gomi emphasizes the group's role in fostering societal understanding of AI and its impact on our relationship with technology. The group seeks to address concerns and biases surrounding AI, paving the way for harmonious human-AI coexistence.

The group recognizes the importance of understanding the "black box" nature of AI and machine learning, particularly in developing energy-efficient systems. Similar to the evolution of cloud computing, where security concerns followed initial adoption, AI's trustworthiness and safety are now receiving increased attention.

Bridging Biological and Artificial Intelligence

The Physics of Artificial Intelligence Group will collaborate with academic researchers to explore similarities between biological and artificial intelligences, aiming for a harmonious fusion of human and AI collaboration. This involves understanding AI training, knowledge acquisition, and decision-making processes. NTT draws a parallel to physics, where understanding fundamental principles (like thermodynamics) led to technological advancements. The group's research will similarly shape the future of AI. Collaboration with Harvard University's Center for Brain Science will further this endeavor.

Neural Network Pruning and AI Inference

Prior research involved analyzing neural network pruning algorithms—a process of removing less important parameters in deep learning networks to improve efficiency. This work focuses on the minutiae of AI's internal mechanics, including the impact of pruning weights versus biases on model performance.

NTT's recent platform updates include a new AI inference chip, a large-scale integration (LSI) designed for real-time processing of ultra-high-definition video. This low-power technology is ideal for edge computing and power-constrained devices like IoT sensors and remote patient monitoring equipment.

Power Efficiency in Edge Computing

The LSI overcomes power limitations in edge and constrained environments by implementing an NTT-created AI inference engine. Using the YOLOv3 object detection algorithm, this engine reduces computational complexity while maintaining accuracy. Collaboration with NTT DATA, Inc. leverages attribute-based encryption (ABE) technologies for enhanced data security and access control. ABE determines data decryption based on attributes and policies, ensuring fine-grained access control.

Other PHI Lab groups are exploring energy-efficient AI computing through optical computing and lithium niobate thin-film technology. Inspired by the energy efficiency of biological brains, the new group will investigate similarities between biological brains and artificial neural networks.

A Harmonious Future for AI and Humanity

The Physics of Artificial Intelligence Group's three-pronged mission includes: deepening our understanding of AI mechanisms and integrating ethics; creating systematically controllable AI environments for observation; and rebuilding trust between AI and human operators through improved operations and data control. Dr. Tanaka highlights trustworthiness as key to harmonious human-AI coexistence, and the group's research aims to develop more natural intelligent algorithms and hardware.

Considering NTT's extensive infrastructure in Tokyo, with its network of underground cables, the company possesses a unique understanding of complex technological systems, providing a strong foundation for its AI research.

The above is the detailed content of The Laws Of Automation, NTT Research Details 'Physics Of AI' Group. For more information, please follow other related articles on the PHP Chinese website!