Nippon Telegraph and Telephone (NTT) will develop a generative artificial intelligence model and offer its services to corporate clients sometime during the current fiscal year, through next March, Nikkei has learned.
NTT will adopt a unique approach by allowing AI to acquire expertise in specific sectors such as finance, medicine, and others, tailored to meet the specific needs of clients in each field. This distinguishes NTT from American tech companies that spend vast sums of money operating generative AI systems for general purposes.
The Japanese conglomerate will promote the service as relatively inexpensive and easy to use.
The service could become a model for Japanese-made AI that will compete with American tech giants’ systems.
NTT will develop a “large-scale linguistic model” as a foundation for its generative AI. The technology is capable of learning from a huge amount of written data, and then converting its knowledge into a range of applications such as writing, summarizing dialogues, and editing sentences.
Through NTT Data, NTT Docomo, and other group companies, the conglomerate will provide specialized services in the financial, medical, legal, and other sectors.
In the financial sector, the model will acquire knowledge of market research and analytical methods utilized in investment decision-making. In the medical field, one application involves assisting doctors by suggesting optimal combinations of medications based on patients’ medical history and current conditions.
In the future, NTT plans to make the system easier to use by adding image recognition functions and adopting voice recognition and camera technologies.
American tech companies such as OpenAI, which operates ChatGPT, Microsoft, and Google are providing AI to anyone on the internet, regardless if they are individuals or corporations. In contrast, NTT will focus on corporate-friendly services.
NTT plans to set a parameter range of 7 billion to 30 billion for its envisioned large-scale linguistic model. GPT3, developed by OpenAI in 2020 and used as the base of ChatGPT, has 175 billion parameters.
The number of parameters serves as an indicator of the performance capabilities of AI systems. An AI system with a large number of parameters can accommodate a broader range of knowledge, but these systems also require significant electricity consumption, which can be costly.
ChatGPT is said to guzzle USD 700,000 worth of electricity every day.
NTT intends to significantly reduce energy consumption by directing its AI towards specific business operations. It will develop a basic model capable of handling English and Japanese. Subsequently, the AI will be trained to acquire expertise in various sectors, including finance, medicine, and others.
In addition to telecommunications services, NTT has been promoting its IT services in recent years — including those based on AI — for corporate users. Groupwide, it plans to invest a total of JPY 8 trillion (USD 57.45 billion) in growth areas in the coming five years. More than JPY 3 trillion will be allocated to a program aimed at supporting the digital transformation initiatives of companies and local governments.
This article first appeared on Nikkei Asia. It has been republished here as part of 36Kr’s ongoing partnership with Nikkei.