'We need artificial intelligence developed by Vietnamese people'

Vietnamese people can create domestically-generated AI products to help ensure information safety and accuracy and best serve the government and society, according to Dr. Dao Duc Minh.

The message was shared by Dr. Minh, General Director of VinBigdata at the AI Summit forum on the morning of September 22 within the framework of the Vietnam Artificial Intelligence Festival (AI4VN) 2023, in Ho Chi Minh City. He said that when chatGPT was launched at the end of 2022, it created an underground wave of competition between countries and large corporations in researching and launching generative AI models and large language models. Specifically, in the US, in addition to OpenAI's chatGPT tool, there are other generative AI models such as Google's Bard, Amazon's Titan. In China, there is Baidu's Earnie Bot, Sense Time's SenseChat, Tencent's Hunyan. In Korea, there is Naver's HyperClova X...

Defining Generative AI, Dr. Minh said that this is a form of artificial intelligence that can create new content and ideas in many different forms such as text, images, videos, music. From there, AI can create new songs, drugs, movies, games... with the use of large language models (LLMs). According to a recent estimate by McKinsey, Generative AI can contribute 4.4 trillion USD to the world economy. The study was evaluated based on 60,000 applications of Generative AI for different industries.

Dr. Dao Duc Minh, General Director of VinBigdata, shared at AI4VN on the morning of September 22. Photo: Thanh Tung — Dr. Dao Duc Minh, General Director of VinBigdata, shared at AI4VN on the morning of September 22. Photo: *Thanh Tung*

Great potential, but according to Dr. Minh, if AI applications are created based on large language model platforms available in the world, Vietnam is at risk of facing many risks. He explained that with large foreign language models, Vietnamese data only accounts for a very small proportion, mainly English, so it is difficult to ensure accuracy, data security, and suitability for business needs... For example, large foreign language models have hundreds of billions of parameters. Running a real model requires a very large computing infrastructure, but the quality is not commensurate because AI models can give wrong answers, especially in historical and cultural knowledge, and issues specific to each country.

Currently, VinBigdata experts have successfully built large language models in Vietnamese with the goal of ensuring data security, improving accuracy, reducing costs, and being suitable for the current situation and domestic problems. The group of experts built a model with hundreds of times smaller parameters than chatGPT, but it was built on the amount of data of Vietnamese people. The model can answer questions with local characteristics with high accuracy. Specifically, when asked about the law with two different violations, the chatGPT model had the same answer and general content. Meanwhile, the AI model using Vietnamese data can answer accurately based on laws, decrees, fines, information sources... very specifically. "Our goal is to develop AI applications with Vietnamese data that can be applied to millions of people and businesses," said Dr. Minh.

From a national perspective, he said that mastering generative AI domestically is important because it can help the government master content, avoid misinformation, ensure national data security, and bring Vietnamese technology to the world. "Vietnam has every opportunity to narrow the gap with the world in the field of generative AI," he said.

Sharing the same view, Mr. Ha Jung Woo, Head of Cloud Innovation Department of Naver Company (Korea), shared that mastering data and technology is an important factor that many countries consider in the future, overcoming the risk of dependence on other countries. Many developing countries also want to create their own AI ecosystem. Naver representative said that currently many large companies develop AI generators mainly using English. Therefore, when the model runs in another language, the information will be incomplete, have errors... creating an unnecessary "bias".

Mr. Ha Jung Woo, representative of Naver, shared about artificial intelligence at the event. Photo: Thanh Tung — Mr. Ha Jung Woo, representative of Naver, shared about artificial intelligence at the event. Photo: *Thanh Tung*

Mr. Pablo Fuentes Nettel, Senior Consultant at Oxford Insights, provided data that there are currently about 60 countries with AI development policies, including Vietnam. According to the speaker, Vietnam has an advantage in AI development thanks to the support of the Government. Vietnam has a bright future if it focuses on investing in AI. "Vietnam has a dynamic economy, a young population, and many talents. I think your country has many advantages and great potential to open up a bright future for AI," he affirmed.

Vietnam Artificial Intelligence Day (AI4VN 2023) is directed by the Ministry of Science and Technology, organized by VnExpress newspaper in coordination with the Information and Communication Technology - Faculty - Institute - School Club (FISU). The two-day event attracted more than 2,000 participants. Over the past 5 years, AI4VN has attracted more than 10,000 participants, more than 100 speakers to discuss the development of a sustainable AI ecosystem in Vietnam.