NTT R&D Forum 2024 - IOWN Integral was held over five days from November 25 to 29, 2024.
The field of Innovative Optical and Wireless Network (IOWN), a next-generation information and communications infrastructure centered on optical technology, is steadily expanding from networks to artificial intelligence (AI), and continues to evolve toward the creation of a sustainable future society. The theme of NTT R&D Forum 2024 - IOWN Integral is “Integral,” which can mean both a math integral or “indispensable.” We incorporated these two aspects to symbolize the evolution of IOWN. The “integral” part means that IOWN will be applied to various fields and the pieces will accumulate, and the “indispensable” part means that IOWN will become “indispensable” to the Earth and humanity.
The technology exhibition was divided into three areas: “Research,” “Development,” and “Business,” and featured a total of 122 booths. The “Research” area showcased 49 research results across a wide range of fields, including networks, User Interface/User Experience (UI/UX), sustainability, security, bio/medical, and quantum. In the “Development” area, 52 examples of the latest research and practical applications related to generative AI and space were on display, with IOWN as the central focus. Furthermore, in the “Business” area, 21 initiatives by NTT Group companies were presented, showing the increasing real world implementation of IOWN.
This forum was organized as an invitation-only event by NTT Group companies, and was attended by a total of 19,261 people, exceeding last year’s attendance. The forum closed to great acclaim, with high expectations for the IOWN future.
Here we will report on the NTT R&D Forum 2024 - IOWN Integral event.
Three people took to the stage for the Keynote Speeches: NTT President and CEO Akira Shimada, NTT Research and Development Director Shingo Kinoshita, and NTT Senior Executive Vice President Katsuhiko Kawazoe. In addition, the technical seminar featured a panel discussion on the theme of “Next Generation AI,” bringing together experts from NTT, the University of Tokyo, Sakana AI, and the technology magazine WIRED Japan to discuss the direction in which AI should evolve. They confirmed the importance of collaboration, such as “moving from a single large-scale AI to the combination of multiple specialized AIs.” In addition, under the theme of “Photonics-electronics convergence technology and the future of supercomputers,” a dialogue was held between the representative of Preferred Networks (a company that has received acclaim for its exploration and achievements in deep learning software technology and computing infrastructure technology) and an NTT researcher who aims to solve societal problems using photonics-electronics convergence devices, starting with “reduced power consumption in the age of AI” and touching upon the very essence of computer architecture. Both the Keynote Speeches and the technical seminar were a great success, with attendees exceeding the capacity of the venue.
Day1Monday, November 25 1:30-2:10PM
Akira ShimadaPresident and CEO
In his Keynote Speech, NTT President and CEO
Akira Shimada touched on the rapid spread of
generative AI and its impact, while discussing
the current state and challenges of AI use in
companies, as well as new solutions
proposed by NTT.
While more than 90 percent of companies in
the United States have adopted AI, in Japan,
the figure is still only around 60%. Furthermore,
even in companies that have adopted it,
it is primarily being used for general-purpose
tasks, and fundamental changes such as expanding
it to more specialized tasks or creating new
services are yet to come.
Given this backdrop, President Shimada proposed
the Industry AI Cloud concept as a platform for
utilizing AI for highly specialized tasks in each
industry. Specifically, he introduced efforts to
create a society with zero traffic accidents in
collaboration with Toyota, and the creation of
a “virtual wholesale market” for agricultural
product trading to optimize the entire food and
agriculture value chain. In addition,
in August 2024, we established a new company,
NTT AI-CIX, which aims to achieve industry-wide
transformation using AI technology.
We have begun collaborating with Trial,
a nationwide supermarket chain, to optimize supply
chains in the retail and distribution industries,
and are working to optimize areas such as shelf
layout optimization and automated ordering.
On the other hand, the increased use of AI raises
concerns about the issue of electricity. With an
increase in power consumption on the horizon,
NTT is working to advance its low-power,
lightweight AI model
“tsuzumi,” and to
promote IOWN, which will create a low-power
computing infrastructure.
Of particular note, the company has announced that
it aims to create a Data Centric Infrastructure
(DCI) board that incorporates photonics-electronics
convergence devices in IOWN2.0, which will be
launched in 2025. At the NTT Pavilion at the Expo
2025 Osaka, Kansai, Japan, we have announced that
we will be implementing a server that aims to
reduce power consumption to 1/8th of the current
level, and will be providing an opportunity for
visitors to try it out. In addition,
Fujitsu unveiled an ambitious roadmap that calls
for a gradual movement toward commercialization
in 2026, optical communication between chips in
2028, and optical communication within
semiconductor chips from 2032 onwards, ultimately
aiming to reduce power consumption to 1/100th of
the current level.
The Keynote Speech emphasized NTT’s vision of
combining the Industry AI Cloud with IOWN to solve
real world problems in a sustainable manner.
It presented a concrete and achievable vision of
the future, in which AI is promoted in the highly
specialized areas of each industry, while providing
solutions to the problem of increasing power
consumption.
Shingo Kinoshita, Executive Officer and Director
of Research and Development Planning at NTT,
began his Keynote Speech by presenting an overview
of the forum and the IOWN roadmap. The latter in
particular demonstrated the gradual adoption of
optical technology from IOWN 1.0 in the networking
domain to the computing domain. The next version,
IOWN 2.0, will implement optical wiring between
server boards, IOWN 3.0 will implement it between
packages, and IOWN 4.0 will implement it inside
chips, tracing a steady path of evolution.
He also reported on the progress of All-Photonics
Network (APN) in the current IOWN 1.0. He says
the company has upgraded its APN service (was
launched in March 2023) to guarantee bandwidth
of up to 800 Gbit/s, the world's highest standard,
expand its interface, and significantly reduce
power consumption. Additionally, a proof-of-concept
experiment demonstrated a low latency of just
17 ms over a long-distance connection of
approximately 3,000 km between Japan and Taiwan.
He also mentioned ultra-fast data backup using
APN and the realization of highly efficient
remote production through APN. He also revealed
that in the future, they aim to achieve unparalleled
transmission capacity and power efficiency
through “on-demand optical path control,”
which will enable multiple wavelengths to coexist
within a single APN without colliding with
each other.
In addition, Director Kinoshita also introduced
the advances in NTT’s version of a Large Language
Model(LLM) "tsuzumi."
While maintaining its light weight and ability to
run on one GPU and one CPU, the system has
steadily improved in performance, including
multi-modal support and improved contextual
understanding. Of particular note is that it has
been adopted by Microsoft Azure and Salesforce
platforms, marking the first step towards global
expansion. He also touched on examples such as
the “AI agent” function for controlling a PC
on behalf of the user, and demonstrated part of
a vision for coexistence with AI where
“tsuzumi” is utilized,
including the implementation of an
"AI constellation" that combines multiple LLMs to
solve real world problems.
To conclude the Keynote Speech, he quoted the words
of Goro Yoshida, the first director of NTT
Laboratories: “Let us draw from the fountain of
knowledge, conduct research, and commercialize
it in order to provide concrete benefits
to the world.” He emphasized the importance of
a consistent approach across research, development,
and real world implementation, and laid out three
commitments: to establish its position as one of
the world’s best research institutions, to ensure
the commercialization of IOWN, and to achieve
meaningful real world implementation.
Day3Wednesday, November 27 10:50-11:30AM
Katsuhiko KawazoeSenior Executive Vice President
NTT Senior Executive Vice President Katsuhiko
Kawazoe gave a presentation in English about
NTT’s current status as a global company and its
key technologies. He explained how the NTT Group
has evolved into a global corporation with over 900
companies worldwide, with 45% of its employees
working overseas, and that it has 17 research
laboratories and 2,300 researchers engaged in
cutting-edge research and development.
He noted that while enhancing ICT technology is
essential in light of the crises modern society
faces, particularly in light of climate change,
natural disasters, and a divided world, there is
also a new challenge: a sharp increase in
electricity consumption due to an explosive
increase in Internet traffic. He also predicts
that data center power consumption will increase
13-fold by 2030. Vice President Kawazoe emphasized
that IOWN is NTT’s concept for overcoming this
limitation, and optical technology is the key.
In 2019, we succeeded in developing the world’s
first optical transistor, and this invention became
the origin of the IOWN concept.
Vice President Kawazoe highlighted several use
cases as examples of new capabilities that IOWN
will bring. One of these is the use of AI to
combat new cybersecurity threats. In particular,
against attacks that combine botnets and advanced
AI, he proposed a new form of defense in which
multiple AIs work together in the IOWN APN to
respond immediately.
He also touched on his efforts to create a
society with zero traffic accidents in
collaboration with Toyota. He made it clear that
they aim to build a safe mobility society by
combining three core technologies: real-time data
collection, constant connectivity achieved
through distributed computing infrastructure,
and learning from diverse data using a mobility
AI platform.
In addition, he also unveiled their vision for
expanding IOWN’s capabilities from the ground to
space. He said that by collaborating with SKY
Perfect JSAT, they aim to build an independent
network system in space and create an integrated
space computing network that can instantly
process and analyze data from observation
satellites in space.
To conclude the Keynote Speech, he spoke about
the capabilities that IOWN opens up for humanity,
illustrated by a project to support the work of a
DJ with ALS (amyotrophic lateral sclerosis).
“Human potential is infinite,” he said, sharing
with the audience his thoughts on creating a
sustainable society.
Susumu TakeuchiComputer and Data Science Laboratories
Senior Research Engineer, Supervisor
Yoichiro MiyakeProject Professor, Institute of Industrial Science, The University of Tokyo
Ren ItoCo-Founder and COO, Sakana AI
Michiaki MatsushimaHead of Educational Content, WIRED JAPAN
Moderated by WIRED Japan Editor-in-Chief
Michiaki Matsushima, the panel discussion
included the University of Tokyo Professor
Yoichiro Miyake, Sakana AI COO Ren Ito, and
Susumu Takeuchi, Senior Research Engineer at NTT
Computer & Data Science Laboratories,
in which they engaged in a lively discussion
about the limitations of current LLMs and their
future prospects.
At the beginning of the session, Mr. Matsushima
framed the discussion with the question “how is
it possible to live in a complex world while
maintaining complexity,” using the term
“technodiversity” proposed by Hong Kong philosopher
Yuk Hui. In response, NTT Senior Research Engineer
Takeuchi, based on the premise that a
“silver bullet” technology probably does not
exist, touched on the possibility that NTT could
find new ways to utilize AI by using
“reasonable LLMs with expert knowledge” that
are superior in terms of computational costs
and power consumption, and by creating
“AI constellations” that allow different
companies’ LLMs to discuss with each other.
Next, Mr. Ito expanded on the discussion, proposing
the construction of “evolutionary model merging,”
which he described as "embodiment of the concept
of AI constellations," that connects Small
Language Models (SMLs) to achieve performance
comparable to LLMs. The method, called
“Frankenstein merging,” combines existing models,
by selecting the top 10 from 10,000 possible
combinations and then repeating the process over
multiple generations. Amazingly, they achieved
performance equivalent to GPT-3.5 in just 24
hours and with $24. He presented technology
that automates human workflows based on these
techniques, such as examples of its use in writing
academic papers. They succeeded in automating the
entire process, from crafting an idea to proof,
paper writing, and peer review. This was the
first AI-based achievement to be featured in
Nature magazine, attracting a great deal of
attention.
From the perspective of expanding AI models,
Professor Miyake presented three important
categories based on his experience developing
game AI. There is the “meta AI” that controls
the game as a whole, the “character AI” that acts
as the brain of the character, and the
“spatial AI” that assists with real-space
recognition. What is particularly noteworthy
is that these techniques go beyond simple uses
within games; they also contain implications
for urban design and the creation of smart
cities. For example, a cyclical approach in
which information from real space is collected
and analyzed using a digital twin, and the
results are then fed back into real space,
should strongly boost future urban development.
In the second half of the session, there was
an in-depth discussion about the direction AI is
evolving in. Mr. Ito pointed to the improvement
of “reasoning” abilities (logical thinking and
inference) as an important turning point in the
current development of AI, and noted that this
will be the key to enabling AIs to communicate
with each other and perform simulations. He also
predicted that the first stage in using AI would
be to improve task efficiency, followed by a
stage where it would be used as a sounding board
to bounce ideas off.
Professor Miyake made an interesting proposal
regarding the possibilities of using AI to
simulate meetings. He suggested the potential
for AI to exhaustively simulate options and
branching points that are often overlooked in
traditional meetings, supporting better
decision-making. He presented the possibility
of a new approach, in which AI would hold a
meeting before the actual meeting, and then
humans would hold a discussion based on
the results.
Regarding the future relationship between AI
and humans, Senior Research Engineer Takeuchi
pointed out its affinity with the Japanese
concept of “eight million gods.” He envisioned
a future in which AI with a variety of roles
will be omnipresent and coexist with humans.
In particular, he emphasized that with
generative AI technology, system integration
that has previously only been discussed
theoretically is now becoming a reality with
more concrete and advanced services.
What emerged from the discussion among the
three panelists was the form that
next-generation AI should take. This future
will not be dominated by a single large-scale
AI. Instead, AIs with a wide variety of
expertise will work together to support human
creativity and jointly seek solutions that
embrace diverse values and perspectives.
Day5Panel Discussion
Friday, November 29 1:00-2:00PM
Shinji MatsuoDevice Technology Laboratories / Basic Research Laboratories
Fellow
Toru NishikawaRepresentative Director & Chief Executive Officer, Preferred Networks
In a seminar held by Shinji Matsuo,
Fellow at NTT Device Technology Laboratories and
NTT Basic Research Laboratories, and Toru
Nishikawa, Representative Director & CEO of
Preferred Networks, a discussion was held on
the direction of technological innovation to
address the issue of increasing power
consumption.
First, the two men gave presentations based on
their companies’ initiatives. Fellow Matsuo
pointed out that data centers currently
account for 2% of the world’s power consumption
and 12% of power consumption in Japan’s Tokyo
metropolitan area. He also mentioned that the
NTT Group itself consumes approximately 0.7
percent of Japan’s total electricity, emphasizing
the need to reduce power consumption.
He demonstrated the potential of optical
technology as a solution. In particular,
he explained that advances in optical
communications technology, such as on-chip
photonics-electronics convergence technology,
will enable “miniaturization, low power
consumption, and low cost,” making it possible
to significantly reduce power consumption.
He presented benefits of introducing inter-chip
optical interconnection, including faster data
communication speeds while reducing power
consumption, ensuring power allocation for
computation, and making effective use of
computing resources independent of the physical
location of the hardware, along with a specific
technical path to get there.
Meanwhile, Mr. Nishikawa, who aims to
“commercialize cutting-edge technology in the
shortest possible time,” spoke about the
challenges facing computing infrastructure in
the age of generative AI. Citing the example
of the “MN-Core” processor they are developing,
he explained how a significant improvement
in power efficiency was achieved by focusing
on the computing unit and on-chip memory and
minimizing control circuits. Supercomputers
equipped with “MN-Core” have been ranked number
one in the world a total of three times in
the power consumption ranking “Green500,”
and the institute is continuing to further
develop its architecture with confidence.
In addition, he also revealed their development
plans for next-generation processors that will
employ different optimal architectures for
inference and learning. He stressed the
importance of creating a “new computer” by
integrating various technologies such as
interconnects and chiplets, not just highly
efficient AI chips.
In the dialog following the presentation,
there was a more specific discussion about
computing architecture in the AI era. What
attracted particular attention was the potential
and challenges of “distributed computing.”
While Fellow Matsuo pointed out that optical
technology can be used to connect memory up to
2 km away without any loss, he also hinted at
physical constraints, stating that some delay
is unavoidable. Given that current supercomputers
are limited to connections within 10 meters,
we asked a fundamental question: “Can a
supercomputer that takes up an entire datacenter
really be used for AI?”
Mr. Nishikawa responded by proposing a new
approach that utilizes AI in the design
itself, stating that as supercomputer systems
become more complex, they are becoming
increasingly difficult for humans to design
accurately. However, he also said that it is
difficult to design complex circuits from the
start, and emphasized the need to develop
circuits gradually, starting with small ones.
Opinions were also exchanged about the future
of packaging technology. Mr. Nishikawa explained
that next-generation processor development
requires a high-density, high-efficiency design
that separates inference and learning,
and he emphasized the particular challenge
of dealing with heat dissipation.
At the end of the discussion, both men offered
messages to young researchers. Fellow Matsuo
stated that “it is important to take into
account the entire planet and society as
a whole,” and expressed his enthusiasm for
“creating a world where energy savings can
be achieved through AI”. Mr. Nishikawa concluded
by reaffirming the importance of comprehensive
research and development such as that by NTT,
stating, “Understanding and integrating both
hardware and software make it possible to achieve
revolutionary innovations that could not be
achieved by either one alone.”
Technical exhibits at NTT R&D FORUM 2024 were divided into three areas: “RESEARCH”, “DEVELOPMENT”, and “BUSINESS.” They were arranged thematically on the basement, first, and second floors of the venue site, allowing visitors to view the entire exhibition in a systematic manner. Exhibits were colored blue in the RESEARCH area, green in the DEVELOPMENT area, and purple in the BUSINESS area, allowing visitors to understand the relationship between the areas at a glance.
「In the RESEARCH area, 49 exhibits showed NTT’s latest basic research achievements in the fields of Network, UI/UX, Sustainability, Security, Bio/Medical, and Quantum. Exhibits of technologies progressing toward practical application drew the attention of visitors in particular. These included technologies for a novel quantum computer, non-invasive wearable glucose sensor, and active noise cancellation.
Exhibit No. γ08-02
Progress is being made in the research of quantum
computers, which leverage quantum properties to
solve computing problems that are difficult to
solve with conventional technologies. While there
are various methods, superconducting quantum
computing and neutral atom quantum computing,
considered the most mainstream methods, require
an environment cooled to near absolute
zero (-273°C) for stable operations. As a result,
they face the issue of limited chip scaling because
the chips must be cooled.
NTT is engaged in R&D to realize
“continuous-variable optical quantum computers.”
Using the company’s cultivated high-performance
optical device technologies, this quantum computer
is capable of large-scale operations at room
temperature. Technologies used in optical fiber
communication can be applied to many of the elements
that make up the continuous-variable optical quantum
computer. When new technologies are needed, extending
optical device technologies cultivated to date by NTT
holds great promise. Integrating optical communication
technologies and optical quantum technologies makes
it possible to achieve large-scale calculations at a
bandwidth 1000 times that of conventional computers,
even at room temperature. This achievement heralds
the arrival of a new innovator in the field of
quantum computing.
For this exhibit, Professor Akira Furusawa of the
Graduate School of Engineering at the University
of Tokyo, whose lab is collaborating in the
development of this quantum computer, participated.
A demonstration connecting RIKEN, where the developed
computer is installed, and NTT was carried out for
the first time (Fig. 1).
The continuous-variable optical quantum computer
is expected to be applied to optimization problems,
which seek solutions that maximize or minimize the
objective function from a vast number of choices
while satisfying constraints, and to neural networks.
The computer is considered to be particularly
advantageous compared with conventional computers
for problems that deal with continuous data
and systems.
The exhibit also touched on the development of a
system that places a cloud between users and the
quantum computer. It has the role of sending
specified parameters to the quantum computer and
sending the results of the job executed by the
quantum computer to the user, allowing the creation
of a more convenient quantum computing
environment.
Applying optical devices developed by NTT, we
anticipate realizing a rack-sized optical quantum
computer around 2030 and a chip-based optical
quantum computer around 2050.
Exhibit No. γ09-01
NTT is advancing R&D in the biomedical field
using its sensor and data collection technologies.
The goal is the realization of precision medicine,
which allows total medical care, including health
promotion, preventive medicine, treatments, and
prevention of worsening of illnesses, tailored to
each individual.
An R&D achievement is a glucose sensor that
carries out measurements using microwaves. NTT is
aiming to commercialize this technology by
FY2028.
Conventionally, a person’s glucose level,
which reflects the amount of sugar in the
bloodstream, a factor related to diseases such as
diabetes, is measured out by sticking a disposable
sensor into one’s body with a needle. However,
this method is highly burdensome for users, as it
brings pain and discomfort. The barrier to its use is
an issue.
To address this problem and allow visualization of
a person’s glucose level without using needles, NTT
is developing a non-invasive sensor in wearable form
that makes contact with the skin and carries out
measurement using microwaves (Fig. 2). NTT has many
equipment for evaluating the response of different
microwave frequencies in communication devices, and
it has identified frequencies that are highly
responsive to glucose.
At present, NTT is conducting trials on glucose
measurement while the user is at rest. Going forward,
our goal is to allow convenient visualization of
changes in the glucose level on a daily and constant
basis by researching and developing technologies to
suppress error factors. Our glucose sensor technology
promises to contribute to supporting appropriate
dietary guidance and exercise regimen to prevent an
increase in the glucose level.
Exhibit No. γ05-01
NTT is pursuing a variety of R&D initiatives
to realize the “ultimate private sound space” where
users hear only the sounds they want to hear and
allow others to hear only the sounds they wish for
them to hear. We are developing and selling
earphones and headphones that allow users to hear
sounds without the phones blocking their ears.
These products provide optimal sound control
tailored to each individual with functions such as
noise cancellation and benefits such as increasing
value at performance venues. NTT seeks to enhance
customers’ experience based on sound.
The exhibit at R&D Forum 2024 provided several
experiential programs. When guests enter a sound
dome, noise is eliminated by several
noise-cancelling installed speakers. We envision
the use of this system in entertainment venues and
general business facilities (Fig. 3). The exhibit
depicts a vision of the future where users can
simply sit at a particular table and easily
converse without environmental noise.
The demonstration of the “Sound SyReal” technology
combined sounds in cyberspace and sounds in the
real world. Sounds flowing into the participants’
earphones and sounds flowing in the exhibition hall
were adjusted in real time using low-latency
transmission. Participants experienced the acoustic
effect of hearing sounds from behind them, even
though no speakers were placed there. Another
demonstration presented “acoustic cross reality.”
Using the aforementioned open-ear headphones,
particular sound effects can be played for only
audience members watching a play as well as render
real sounds on the stage clearer.
NTT has proposed technologies for “precise control
of sound,” allowing users to not only spend time
with peace of mind and security in a world where
people and noises are mixed, but to also experience
increased value in all sorts of venues.
Exhibit No. γ03-01
NTT is also taking on the challenge of R&D in
the “brain tech” field, which integrates psychology,
neuroscience, and technology to elucidate the
physical mechanisms of brain functions, leading to
the development of various applications and
services. Presenting NTT’s “motor-skill-transfer
technology,” the exhibit demonstrated a movement
support device that uses brainwaves, which differ
in each person.
For the demonstration, the user sits in an electric
wheelchair and wears brainwave-measuring headgear.
Next, a tutorial similar to that of videogame asks
the user to visually imagine moving to the right and
to the left. Mentally visualizing the movements
causes changes in brainwaves. The “motor imagery
estimation AI” analyzes the brainwave patterns and
converts them to control commands. This technology
allows the user to operate connected devices or an
avatar on a screen (Fig. 4). With this technology,
users can control equipment such as an electric
wheelchair by means of their brainwaves as they
visualize movements without moving their
bodies.
NTT’s research is strong in the tuning of the
“motor imagery estimation AI.” The technology
consists of a proprietary deep learning model that
incorporates parameters that account for the timing
of intentional switching in motor imagery and a
weakly supervised learning system that allows the
model to be updated sequentially as it adapts to
brainwave changes during the use of the measurement
headgear. In this way, reduction in estimation
accuracy is prevented.
NTT’s research seeks to increase the level of
freedom of movement for people with physical
limitations, such as those with disabilities, as
well as for people who cannot move temporarily.
Furthermore, enabling smooth control of avatars in
cyberspace will make it possible to eliminate the
need for control devices, allowing more people than
before to spend life in the metaverse.
The motor cortex, the part of the brain that is
responsible for motor commands, has been mapped in
detail for locations that correspond to movement
commands. Improving the accuracy of the “motor
imagery estimation AI” can thus be accomplished by
measuring brainwaves in greater detail going
forward. Another possible approach for increasing
accuracy is to combine brainwave data with
different data such as electromyography (EMG),
which measures slight changes in the electrical
field when muscles contract.
Exhibit No. γ01-01
Until now, it has been difficult to easily measure
the motor skills of hands and feet because
traditional assessment of hand and foot dominance
uses specialized and expensive equipment. To address
this issue, NTT has developed technology that uses
smartphones to easily evaluate motor skill.
Measurement of motor skill is done by simply holding
a smartphone running a dedicated app with one hand
or attached to a foot with a strap, and then tracing
the circle drawn on a sheet of paper with the
smartphone (Fig. 5). Deviations in circular movements
are quantified by NTT’s proprietary algorithm to
calculate the level of variability. The algorithm
evaluates the user’s repeated movements in a short
time by measuring the trajectory of the accelerator
sensor. By comparing the level of variability with
other users in the same age group, the user’s motor
skill can be easily quantified.
Through large-scale acquisition of data on motor
skill, which has been difficult until now, it is
possible to use the data to see changes in motor
skill according to age and to conduct a large-scale
survey on the effects of correcting children’s
dominant hand. Representative use cases of the
technology are assumed to be training and
rehabilitation, with the recovery of motor skill
in the hands and feet visualizable through regular
measurements of the level of variability.
Experiments already conducted jointly with sports
equipment manufacturers resulted in data showing
that the level of variability falls until the
mid-teens and increases as one ages. We expect that
this technology will be applied in the future to a
variety of fields including sports, medicine, and
the nursing care industry.
Exhibit No. γ06-01
Well-being (not only good physical and mental
health but also good social and economic
conditions) has become a global concern together
with SDGs. It is being emphasized not only in
individuals’ lives but also in group settings such
as companies and schools. To realize a sustainable
society, it is necessary to continually measure
the actual level of well-being. However, until now
it could not be done at low cost and with high
accuracy.
At NTT, we are also promoting the enhancement of
employees’ experience. To realize this, we are
advocating and researching the concept of
“social well-being,” which emphasizes the balance
between group harmony and individual autonomy.
In considering social well-being, in addition to
individual well-being, the involvement of
individuals in the group to which they belong is
critical. It is also essential to evaluate the
health of the group.
First, in joint research with the Kanazawa Institute
of Technology, integrating the findings of social
psychology and educational psychology, NTT has
formulated “well-being competencies” composed of
practical qualities and abilities needed to live a
life of well-being. “Well-being competencies” allow
individuals to subjectively understand steps to
increase their well-being. As an aid, NTT has also
developed technology that combines “well-being
competencies” with objective information. For
example, NTT’s “text risk assessment technology”
evaluates how intimidating text that is posted
comes across to other people. By using this
technology, it is possible to not only objectively
evaluate the effects of an email on its readers,
but also highlight intimidating statements
expressed in meetings from meeting minutes. NTT has
also developed a similar assessment technology for
collective well-being (Fig. 6).
By developing technology for measuring and
supporting well-being and establishing practical
know-how that connect these two areas, NTT seeks to
contribute to the realization of a sustainable
society by supporting individual and collective
well-being in both their subjective and objective
aspects.
Exhibit No. γ04-03
Optical communication has become the standard at
all kinds of communication sites. However,
the computer circuits that drive optical
communication are still electrical.
If everything from circuits to networks could be
implemented optically, significantly low latency
and reduction of power consumption could be
realized.
As an initiative to achieve this reality, NTT is
engaged in the development of photonic cryptographic
circuits that carry out cryptographic operations
using light. This exhibit presented an algorithm
that uses a photonic encrypting circuit that converts
electronic bit values to photonic addresses to
realize low latency. Also presented was a method of
converting a cryptographic circuit into a photonic
circuit using NTT’s optical integration technology.
NTT has made noteworthy achievements in this area,
such as the carrying out the world’s first one round
of symmetric key cryptography on a photonic
circuit.
Furthermore, NTT is developing an all-photonic
chip (Fig. 7). At present, we are at the stage of
improving the technology, such as conducting
research on suppressing heat generated by reducing
the number of switching elements. We aim to
establish low-latency and low-power communication
systems by deploying this technology in servers and
other equipment. The introduction of “optical
semiconductors,” which NTT is engaged in, is also
expected to give a big boost to the Japanese
semiconductor industry.
Exhibit No. γ10-13
When responding to a large-scale network failure,
it is essential to analyze the management
information of multiple networks. However, because
the management of each network is separate, until
now it has been difficult to grasp the impact of a
failure that spans multiple networks and quickly
determine responses.
NTT’s network failure impact assessment and
countermeasure navigation technology overcomes
these difficulties, enabling faster network
recovery. First, when a network equipment failure
occurs, information is collected using NTT’s NOIM
common data model to identify the failure and impact.
This model enables unified management and
visualization of various networks, such as RAN,
transmission, and core networks. In addition,
the use of NTT’s network failure location
estimation technology (knowledge-based autonomous
failure-event analysis technology for
network (Konan)) allows automatic analysis of
communication service failures (Fig. 8). After the
failure location and impact are identified,
the autonomous scenario creation LLM agent is
connected to the failure information. The LLM agent
references equipment operation manuals and other
documentation and carries out repairs and recovery
in real time. Instead of carrying out pre-defined
responses, NTT’s technology collects and utilizes
information appropriate to the failure situation
to carry out the appropriate response. What’s more,
NTT’s LLM agent provides navigation that is more
accurate than what is generally available.
Until now, experienced personnel surmised failure
locations from past alerts and action logs, and
responses were carried out by looking for the
necessary operation manuals and looking up
information. NTT’s new network failure impact
assessment and countermeasure navigation technology
makes it possible to automate the entire process
from identifying the network failure to recovering
the network using the LLM agent.
We are aiming to refine the technology to accurately
and autonomously visualize and navigate network
failures for multilayered network infrastructure,
which support social services. Our goal is to enable
issuing an initial report on the impact of a network
failure on services within 30 minutes and achieve
network recovery within 60 minutes.
In the DEVELOPMENT exhibition area, 52 exhibits, including exhibits on NTT’s latest IOWN-related research initiatives and practical applications, generative AI, and space technologies were presented. Examples include the collaboration between human beings and AI based on NTT’s evolving LLM “tsuzumi,” “Media Production Digital Transformation (DX),” which realizes a video production facility through IOWN-based network, sharing, and cloud technologies, and innovative technological developments like the NTT Group’s space business brand “NTT CONSTELLATION 89 PROJECT (NTT C89).”
Exhibit No. δ01-01
As can be understood from the exhibit’s key message
“Empowering Business and Life with a New Level of
Comfort through ‘tsuzumi,’” the biggest
difference from last year’s “tsuzumi” exhibit is the
addition of a technology to leverage “tsuzumi” in practical
applications. This technology is the “Visual
Comprehension Adapter” (Fig. 9). Documents such as
Web pages and PDFs contain not only text but also
images and diagrams. Their layout holds meaning.
The exhibit presented visual comprehension
technology that allows the meaning of a document
as a whole to be understood.
As an example of intra-company DX made possible
by “tsuzumi,” the
exhibit demonstrated “tsuzumi’s” automation of
the creation of order slips for desired products.
“tsuzumi” successfully
created the order slips by referencing internal
work manuals while operating the actual system
screen.
Many data and IT systems created for human use
become difficult to operate due to their age and
the technologies used in their creation, becoming
burdens in daily life and work. Especially in places
such as large corporations and government agencies
where large-scale system upgrades require time and
capital expenditure, legacy technologies are kept
in use through extensions. Against the conventional
approach to upgrade enterprise systems, “tsuzumi” shows the way to
new business transformation thanks to its ability
to understand and perform operations on current
systems. What’s more, “tsuzumi” is continuing to
evolve as it facilitates collaboration between
humans and AI through information acquisition
and assimilation by means of natural human dialogue
and response and document comprehension.
Exhibit No. δ03-07
In recent years, the proliferation of Internet
media has brought about intensified competition
among video content companies. “Media production
DX” using IOWN APN promises to address the need
to develop high-quality contents and improve
operational efficiency.
The exhibited technology that uses the IOWN APN to
implement a wide-area network spanning multiple
locations to connect a large number of shooting
sites and production bases. IOWN APN is ideally
suited because it features high capacity, low
latency, and no jitters. For example, to broadcast
live sports, conventionally staff and a production
truck loaded with production equipment were
dispatched to the event. Using IOWN APN, however,
makes it possible to directly transmit source video
from the event site to a private cloud and edit
the video in real time in a studio at a separate
location.
At this exhibit, a media content production
environment that may become the new standard was
presented. Its capabilities include actually
broadcasting a TBS TV information program from the
Musashino R&D Center with video switching and
mixing (Fig. 10) and activating an international
IOWN APN spanning 3,000 kilometers by connecting
with Chunghwa Telecom in Taiwan.
Exhibit No. δ05-09
The NTT Group has launched the “NTT C89” brand for
its space business and industry. The brand is
engaged in creating space services and addressing
global issues such as climate change. In recent
years, as the cost of launching satellites and
rockets fall, businesses and services making use
of outer space have increased.
One service is planetary survey by a “rover.”
This planetary exploration hardware is indispensable
for space exploration. Wide-area surveys by a rover
can be carried out either through automation or
remote control, making it an important partner in
advancing exploration of the lunar surface. However,
there is drastic environmental difference between
“daytime” on the moon, during which the surface
temperature rises to 110℃, and “nighttime,” during
which the surface temperature falls to -180℃. This
presents a barrier in the utilization of rovers,
as it is difficult to use batteries as a power
source and costly to transport power cables
from Earth.
NTT has thus developed a method to supply stable
power wirelessly to the rover, unaffected by the
surface temperature or sunlight on the moon. This
method transmits highly efficient, contactless
energy using materials that can be procured from
the lunar surface. Power is sent to a rover equipped
with a receiver from the transmitter. The
transmitter, which includes a proprietary field
resonance antenna that generates strong electric
field waves, uses “lunar sand” called regolith as
the transmission path (Fig. 11). Compared with
conventional methods such magnetic resonance or
microwaves, the surface area for transmission is
expanded by more than 100-fold and the transmission
efficiency by more than 10-fold. Put another way,
the technology constructs a “lunar surface power
transmission network” using lunar regolith. NTT
seeks to apply this technology not only to
exploring the moon, but also to provide power to
space elevators from 2050 onward.
Exhibit No. δ02-02
Existing servers consume great amounts of CPU power
for GPU control and communication. It is difficult
to aggregate GPUs in a remote location because costs
and delays in communication are incurred. NTT has
therefore developed a technology that makes
efficient use of accelerators and realizes
low-latency connections in Data Center
Interconnection (DCI), contributing to the reduction
in data center energy consumption. The exhibit
presented a server system using NTT’s DCI controller
software (Fig. 12). The controller software issues
instructions for photonics-electronics convergence
switches and allocates accelerators to the
composable server. This allows accelerators to
communicate directly with each other at high speed
at photonics-electronics convergence switches. At
the composable server, accelerators are flexibly
allocated in the resource pool in response to
changes in the workload. In experiments, applying
the DCI contributes to a 34–62% reduction in
energy consumption, even in AI analysis of
images.
In addition, the use of “low-latency remote
connections,” which realizes data copying between
memory at remote locations (remote direct memory
access (RDMA)) on the APN, allows the construction
of an environment where information can be processed
even when the “data aggregation” base and the
“data ingestion” base are several hundred kilometers
apart. As a result, besides a reduction in CPU
overhead, latency in transmission and power
consumption could be reduced by up to 60%. Going
forward, NTT envisions use cases such as security
monitoring and behavior analysis using multiple
cameras across various locations, which is required
for the realization of smart cities and digital
twins. In the future, this technology is expected
to further reduce costs and save energy by allowing
GPUs to be used efficiently through consolidation
across remote locations.
Exhibits No. δ01-10, δ01-11
The use of generative AI is being sought not just
by businesses for their work sites but also by
individuals and local governments. Companies are
in fierce competition to provide LLMs, which can
create new text by learning textual data and
understanding human languages at a high level,
as services. However, the output results depend
greatly on each LLM’s training data and algorithms.
There are cases also where the output is a general
solution and cases where the result comes only from
a single perspective. NTT is thus engaged in the
development of AI technologies based on the
philosophy of “AI Constellation.” Instead of
depending on a vast monolithic LLM, multiple small
LLMs having their own expertise and individuality
are linked to improve diversity in results. This AI
architecture is inspired by the image of stars
connecting to form constellations, hence its
name.
One of the technologies that makes this possible
is the “Facilitator AI,” which links LLMs to advance
discussion. A facilitator is someone who stimulates
debate from a neutral standpoint in meetings and
group work in order to support mutual understanding
and consensus-building among participants.
The Facilitator AI plays the exact same role.
For example, in Omuta City in Fukuoka Prefecture,
a public participation workshop used AI
Constellation, with multiple LLMs and the
Facilitator AI, to brainstorm ideas on ways to
prevent the need for nursing care in the
city (Fig. 13). LLMs were given roles such as that
of an internist, occupational therapist, social
epidemiologist, dentist, and welfare commissioner.
They each presented ideas considered necessary in
their field of expertise. The Facilitator AI
coordinated these contents and led discussions so
that participants could provide diverse
perspectives while minority viewpoints were
respected. At the workshop, participants experienced
how debate could be promoted based on conversations
raised by LLMs, similar to conversations among
human experts.
NTT is also developing technology that allows AI
agents to acquire knowledge from conversations
between LLMs and grow autonomously. We envision a
future where autonomously growing AIs and humans
work together to perform complex tasks with ever
increasing precision.
Exhibit No. δ01-13
Is it possible to create “smart robots” which,
instead of just operating when instructions from
humans are received, understand their own tasks
and situations and act flexibly on their own?
Spurred by this question, NTT researchers are
developing a control technology for service robots
that incorporate generative AI. Instead of robots
that operate based on predetermined rules and
algorithms, the aim is develop robots that act on
their own initiative by deriving results that come
from combining in the LLM the perspectives of
“role and business knowledge,” “one’s own ability,”
and “situational awareness.”
The exhibit’s demonstration presumed a nursing home
as the setting. The LLM was given the role of
“assisting a client in the nursing home.” The
following scenario was shown: A man sitting down
takes a beverage from a refrigerator nearby but
leaves without closing the door. The service robot,
which saw what happened, calls out to the man in
synthesized speech: “The refrigerator door is open.”
When it judges that the man is not returning, it
moves toward the refrigerator by itself and closes
the door (Fig. 14).
The LLM understands each step taken in the users’
action history, such as “drinking a beverage” and
“opening a can.” From this history, it infers
“things to keep in mind” and “candidate actions to
carry out.” It then determines the action that
should be carried out from among the candidates,
which is then executed by the service robot.
There is wide room for what service robots, such
as the demonstration robot, can do to provide
hospitality, even when it comes to social issues
such as a declining working population and the need
to provide support for the lives of the elderly.
Exhibit No. δ04-03
Especially since the advent of COVID-19,
livestreaming services are facing challenges in
providing flexible resources to meet customer needs
and achieving real-time streaming without a sense
of delay using conventional optical networks.
Furthermore, livestreaming is also drawing attention
as a business tool. To achieve rich contents using
AI processing and other technologies, specialized
hardware and staff are employed.
For example, consider a case where a person’s
movements are captured by cameras from different
locations to produce an avatar in real time through
graphics processing and then streamed on a single
screen. In such a case, the avatar’s movements may
be jittery and appear unnatural due to latency in
transmission or processing, resulting in an
undesired experience for the user.
NTT has developed a technology for real-time video
processing within a network and subchannel circuit
exchange (SCX), which allows distributed data
centers to be realized. In this exhibit, the
streaming of high-quality video by simple equipment
was demonstrated. This was made possible by
connecting video processing functions implemented
in the network with video capture sites using
APN.
Even when video from multiple locations is used as
input and advanced video processing using AI and
other technologies is carried out, a stable
communication path could be freely constructed
thanks to SCX technology. Redundancy required for
video services could thus be realized, as could
support for bandwidth change (hitless resizing),
thus ensuring that streamed video is
synchronized (Fig. 15). This APN-based end-to-end
optical video streaming architecture will realize
low-latency live streaming that connects multiple
locations on demand.
In the BUSINESS area, NTT Group companies showed their initiatives in applying R&D results to actual business through 21 exhibits. Their efforts were divided into three categories: Communication & Computing, CX (Customer Experience) & DX, and Security & Privacy. Socially implemented technologies directly connected with commercialization were introduced, including “Rectification of Urban Road Traffic” using NTT’s “4D Digital Platform®” to reduce social burden and “Trust Technology,” which is closely connected to generative AI technology.
Exhibit No. β01-10
The use of “digital twins” as a method for
approaching real-world problems is growing. The
technology reproduces real-world objects and
conditions and monitors and simulates them in the
virtual world. NTT has incorporated “digital twin
computing (DTC)” into the IOWN concept, and is
developing and utilizing its “4D digital platform®”
as a supporting technology. The platform includes
real-time integration of sensing data with
high-precision location and time information in
the “Advanced Geospatial Information Database,”
which contains rich semantic information, and
enables high-speed analysis and prediction.
“Traffic rectification” is drawing attention as
methods that make use of the 4D digital platform®.
To apply them, it is essential to evaluate and
predict the effects of individuals’ behaviors on
traffic as a whole under Transportation Demand
Management (TDM). At present, NTT Group companies
and the Hanshin Expressway Company are jointly
studying the effectiveness of urban road traffic
rectification. For example, they are conducting
traffic simulations that connect a variety of
times-series spatial data and providing traffic
recommendation information to users on their
smartphones. Also being planned is a demonstration
of integrated data analysis and usage that includes
training for increasing the value of
recommendations based on the results of behaviors
carried out by users in response to
recommendations (Fig. 16).
Improving technologies for linking, integrating,
and utilizing diverse time-series and spatial data
promises to benefit not traffic rectification but
also fields such as mobility and marketing.
Exhibit No. β03-06
The use of generative AI in speech synthesis is
bringing advancements to technologies such as
text-to-speech, which reads written text aloud in
a natural tone, and technologies that samples the
voices of real people for use in speech synthesis.
On the other hand, the practice of training the
voices of characters for generative AI without the
permission of actors or copyright holders and
producing different contents with the trained data
is problematic. This has led the Japan Actors
Union, which includes voice actors and actresses,
to launch a public awareness campaign with a video
calling for the necessity of creating rules for the
use of human voices for AI. Besides promoting the
fair use of training data, a system that provides
fair revenue sharing is also being sought.
Combining NTT’s “Multilingual Technology” and
“Trust Technology” provides a solution to these
issues. First, NTT’s multilingual speech synthesis
technology allows a tailor-made speech synthesis
model to be constructed by learning the base speech
synthesis model together with the speaker’s voice
in a session ranging from as short as a few minutes
to at most about 10 minutes. In addition, the
technology can use this model to achieve
cross-lingual speech synthesis in Japanese,
English, Chinese, and Korean. NTT’s “Trust
Technology” applies blockchain technology’s
transparency and tamper-proofing functions, which
are also used in cryptocurrencies, to tags attached
to voice data, speech models, and speech synthesis
so that the authenticity of each data can be
verified. This technology also facilities revenue
sharing by providing histories of data usage.
At this exhibit, visitors experienced a
demonstration where the generative AI learned the
voice of a real-life idol or announcer and answered
questions with a voice similar to them. The
technology also provided responses in multiple
languages (Fig. 17).
NTT seeks to create opportunities to use voice
contents globally by connecting partners who wish
to provide their own voice as intellectual
property (IP) with parties who wish to use AI speech
in services and goods.