Good software iterates. We know that almost every longstanding software platform or tool and every operating system or application goes through a series of logical incremental progressions and builds, generally referred to as an iterative process, or just ‘iteration’. Even when completely new hardware and software development methodologies, programming languages and automations exist, there is still an organic evolutionary process involved as software engineers cut their teeth on emerging technologies while new business (or life) deployment scenarios are explored.
NTT has taken this core truth to heart and explained how its Innovative Optical Wireless Network (IOWN) technology platform is now in a position where the company hopes software engineers, systems architects and other technologists will embark on a process of embryonic exploration with it.
Why is this process such an exploratory mission? Primarily because NTT’s IOWN concept is a new type of communications infrastructure capable of providing high-speed (low-latency) broadband communication and enormous computing resources by using low-power technologies including optical technologies. At the eye of these optical technologies is the use of photonics. With the longstanding truth of Moore’s Law governing silicon microprocessor development now becoming a memory, photonics-based processors promise to reimagine the way we create the ‘wiring’ of chips with light-based technology that can perform high-speed arithmetic calculations.
Why embryonic prototyping is good
In his role as president and CEO, NTT Research, Inc., Kazuhiro ‘Kazu’ Gomi has spoken of the need to tackle challenges such as unsustainable energy consumption. He has also been vocal on the need to combat data and personal privacy issues and to address the rising costs of medical care in an ageing world population.
“Amid so many challenges, invention is necessary — and non-linear thinking is at the beating heart of our enterprise,” said Kazu Gomi, who calls for embryonic experimentation at the heart of what the firm says are ‘sophisticated research modalities’ today. Focused on what the company refers to as ‘basic research’ i.e. technology innovations that will have a deep and long-term impact on the way we live our lives, the call to software developers from NTT is as broad and wide as it is open.
“Building with photonics is an embryonic process and we want to encourage software engineers at all levels to touch these platforms,” enthused Kazu Gomi. “We need to do things that are a bit disruptive and move away from the notion of linear growth or linear development in the normal sense. Building the technology for a Quantum Neural Network and harnessing the power of photonics is a massive exploration of resources, techniques and approaches – and it is definitely an industry-wide effort.”
When we talk about ‘basic’ R&D inside NTT Research, we are focused on what we could also call pure (as opposed to applied) research.
“To clarify further, in applied research, an engineering team might typically be working to deliver a defined technology product for a Q3 release (for example) with relatively defined functionality and deliverables that fit to what might be a preexisting roadmap,” explained Kazu Gomi, speaking to press and analysts this April 2024.
With the majority of the work carried out at NTT Research’s facility in Silicon Valley being of the pure/basic embryonic type, the engineers do positively dovetail with the teams in Tokyo, where there are different sets of resources available to tap into. The question therefore arises, how do software developers and other technology engineers feel about working on embryonic-level products, some of which might never come to market?
Finding the finding out factor
“For individuals, it’s all about the ‘finding out’ factor and diving into the process of figuring out how things work. These are people who thrive on discovery and they know that they’re part of a team unit that can take innovations forward and start to productize them at various stages of any given technology’s evolution,” said Chris Shaw, chief marketing officer at NTT Research. “In reality, some people are naturally great at being scientific researchers on esoteric emerging projects – and equally, some people are inherently suited to working on (for want of an example) supply chain manufacturing applications for ERP systems. We know that team philosophy at NTT Research is: put the right people in the right roles and allow them to do what they do best.”
Shaw says that this level of engineering professional is not so worried about failure. As he noted, the fulfilment and motivation factor for them is all about finding out the basic fundamental way that things (any thing) works. They know that someone else (in the productization strategy team, or elsewhere) will see that spark if they find it and perhaps turn it into something.
As an essentially still-nascent technology – and with this discussion centering around the work of IT engineers in the pure research space – are we at the point where we start talking about app developers, web developers, cloud developers, mobile developers and now photonics developers?
“That reality is coming up,” assured NTT Research’s Kazu Gomi. “In terms of validation for these technologies there are plenty of working examples, in terms of codification (in areas such as certification) we’re not there yet, but that’s okay, this is a journey. The systems that result from IOWN and the use of photonics will impact our technology use cases at every level over the next couple of decades, but at this stage it’s important to remember that we’re still building the hardware for many of these innovations i.e. the software services typically come afterwards, obviously. Photonics today is still having an impact on efficiencies that we can apply now i.e. there is so much communication between microprocessors and memory cells, so light-based engineering is already changing the parameters here, but much more will come.”
Eye AI, NTT’s tsuzumi LLM
As a case in point to illustrate what NTT says is some of the most progressive work happening in its research facilities in Silicon Valley and Tokyo, the company’s current work with its ‘tsuzumi’ Large Language Model (LLM) is fairly esoteric. Now integrating visual reading technology to operate like the human eye, NTT’s tsuzumi LLM can now more comprehensively understand and process graphical elements within documents, such as images, charts, diagrams or icons.
Proprietary to NTT and lightweight in terms of software code footprint and component ‘parameters’ (factors that govern how an AI model is capable of interacting with its data and making predictions from it), the company says that this visual reading technology functions bolsters existing ‘reading’ capabilities of tsuzumi and does so in a low-cost and energy-efficient manner. Currently undergoing enterprise trials to explore prototype use cases, the technology has been developed in collaboration with Professor Jun Suzuki of Tohoku University’s Center for Data-driven Science and Artificial Intelligence.
“LLMs have become capable of handling high-level natural language processing tasks with high accuracy and multimodal models, including those that integrate vision and language, are beginning to emerge,” said Kyosuke Nishida, senior distinguished researcher at NTT. “However, there remain significant challenges in comprehending documents or computer screens that contain both text and visual information, such as charts and tables. By integrating our visual reading technology with tsuzumi, NTT aims to give ‘eyes’ to AI-powered tools, unlocking new applications and functionality,” said Nishida.
Lightweight low-power LLMs
While traditional LLMs require large amounts of power for training, tsuzumi’s smaller size (an ultra-lightweight version has 600 million parameters and a lightweight version has 7 billion parameters) reduces the energy consumption and costs associated with training, inference, and tuning, making it a more sustainable and cost-effective option for businesses. Additionally, tsuzumi supports twenty human languages including both English and Japanese and allows for inferencing on a single GPU or CPU. It is compatible with both visual and audio modalities and can be specifically tuned for particular industries or enterprise organization use cases.
This year we know that NTT researchers envision four initial primary use cases for enterprise deployment of tsuzumi and the visual reading technology. According to NTT, “These include customer experience solutions including call center automation; employee experience solutions for tasks involving manual searching and reporting, including electronic medical recordkeeping in the healthcare industry; transforming the value chain for industries including life sciences and manufacturing; and software engineering for systems and IT departments, including development and coding assistance and automation.”
NTT is currently conducting commercial trials of tsuzumi and has consulted with over 500 companies from around the world on the potential introduction of the technology into their systems.
Our ephemeral embryonic IT future
Of all the emerging work being carried out here, it is not unusual to talk about experimental prototyping when it comes to the development and deployment of Artificial Intelligence – especially in the generative AI space. Enterprise technology platform companies used to be frightened of saying that they were at an ‘early experimental stage’ with any given technology – refreshingly perhaps, this is no longer the case.
Because so many technologies at the real cutting edge of tech are so embryonic, still-nascent and often somewhat ephemeral in changing cloud networks, we need to encourage and embrace experimentation – especially of course new and radically different fields such as photonics. There is light at the end of this tunnel, in every possible sense of the term.
