Thai artificial intelligence (AI) engineers from public and private organisations have teamed up to develop a Thai-language generative AI-powered chatbot, expected to benefit several sectors from education to business.
The National Electronics and Computer Technology Center (Nectec) recently revealed its partnership with three organisations to launch the OpenThaiGPT project.
The three organisations are the Artificial Intelligence Entrepreneur Association of Thailand (AIEAT), the Artificial Intelligence Association of Thailand (AIAT) and the NSTDA Supercomputer Center (ThaiSC).
Generative AI is a type of AI that can interact with users in natural language and create data, ranging from articles to multimodal content such as images, videos and audio.
The OpenThaiGPT project tracks the popularity of generative AI, launched when the ChatGPT chatbot took the world by storm last November. This AI-powered tool can answer questions in beautifully written sentences, assist programmers in writing code and even compose music.
Created by OpenAI, an AI lab in San Francisco, ChatGPT made headlines as it surpassed 1 million users within a week after its launch.
Now other global tech giants are jumping on the bandwagon to produce their own versions.
"It's important for Thailand to become a technology producer rather than just a user in the field of generative AI," said Thepchai Supnithi, director of Nectec's AI Research Group, at the centre's recent seminar titled "Challenges, Possibility, and Opportunities of Thai ChatGPT".
Mr Thepchai, who is also an AIAT representative, said Thailand must have the capability to develop, modify and harness the full potential of Thai language using AI.
According to Nectec, the OpenThaiGPT project is a chat-based assistant project where understanding the Thai language is crucial to enable dynamic information retrieval and a user-friendly interface.
He said OpenThaiGPT is being developed on a open source basis and will be distributed free, with the option for customisation and extension.
TIMING IS EVERYTHING
"This is the ideal time for Thailand to launch this project as the country has an abundance of local language data from Pantip.com as well as the largest supercomputer facility in Asean: ThaiSC's LANTA," said Mr Thepchai.
The latest release, OpenThaiGPT 0.1.0-alpha, is capable of various tasks such as questions and answers, machine translation, step-by-step explanation, paraphrasing and coding suggestions.
In future releases, the number of OpenThaiGPT parameters will increase from 3.74 billion to improve its language understanding and natural interaction, he said.
The project uses publicly available data, with Pantip.com one of the current data contributors.
The project is always mindful of data privacy and copyrights, while encouraging more data contributors to join the project, said Mr Thepchai.
Kobkrit Viriyayudhakorn, president of the AIEAT, said OpenThaiGPT can be connected with external systems or easily expanded, customised and further developed by everyone.
The project can also be integrated with other applications, serving as a building block for the development of the next generation of AI in Thailand, he said.
Possible use cases of OpenThaiGPT include chatbots, news summaries, automatic translation and document classification.
Prachya Bookwan, Nectec researcher for natural language processing, said OpenThaiGPT 0.1.0-alpha is capable of answering general questions, translating languages and coding.
The 1.0.0 version slated for release mid-year has up to 10 billion parameters, he said.
In addition, the OpenThaiGPT team aims to develop a multimodal option that can specify physical references beyond language.
Mr Prachya said the technology has the potential to work on smartphones in the future.
Wanchat Padungrat, the founder of Pantip.com, said he is willing to support the project so that it has a positive impact on society.
"If this project doesn't yield good results or outcomes, it can set us back," Mr Wanchat said.
TRUSTWORTHY AI
Viwan Jarerattanachat, head of scientific support and domain research at ThaiSC, said this project should improve AI capabilities in Thailand.
However, she acknowledged it is still a case of trial and error, requiring safety reviews to ensure it will not harm people or be misused.
The project needs to have transparency, accountability and responsibility, said Ms Viwan.
Apivadee Piyatumrong, Nectec's senior researcher for the AI Research Group, said there are guidelines and frameworks for creating trustworthy AI in many countries.
In Thailand, the Digital Economy and Society Ministry developed the "Digital Thailand-AI Ethics Guideline" in 2018 to include ethical considerations in the principles for the development of AI, said Ms Apivadee.
Regarding the development of OpenThaiGPT, she said there are three main areas to consider to ensure its trustworthiness.
First, safety measures need to be in place to prevent any harm to life or property that could arise from use of the technology, said Ms Apivadee. For example, if OpenThaiGPT is used in the medical field, it needs to be reviewed for safety and effectiveness.
Second, transparency is essential for OpenThaiGPT. It should be clear where the data and models used come from and what their quality and performance are.
Third, accountability and governance are important factors to ensure OpenThaiGPT is trustworthy, she said.
Chai Wutitiwatchai, executive director of Nectec, said Thailand has a National AI strategy master plan that was approved by the cabinet last year.
The country's AI Government Readiness Index score has improved from 60th in 2020 to 31st in 2022, he said.
According to Mr Kobkrit, the AIEAT study titled "Thailand's AI business status in 2022" reported total AI revenue grew 24% from 35.2 billion baht in 2020 to 40.8 billion in 2021.
The amount was mostly contributed by 105 AI companies, 50 system integrator companies and 50 hardware automation firms, he said.
Touchapon Kraisingkorn, co-founder of tech startup Amity, said he hopes OpenThaiGPT becomes the de facto open-source GPT model to gain wide acceptance in the Thai market.
He said GPT technology has immense capabilities to solve inefficiencies in both the public and private sectors.
"It is a once-in-a-decade technology that will allow Thailand to make leapfrog gains in productivity and technological competency," said Mr Touchapon.
ENTERPRISE USES
Panutat Tejasen, chief executive of the new AI startup ThaiGPT Co, said the company sees a business opportunity capitalising on generative AI technology and the popularity of ChatGPT.
The company is also open to using other generative AI models, he said.
Thai cryptocurrency exchange Bitkub is a main investor in ThaiGPT Co.
ThaiGPT aims to develop, integrate and customise AI solutions using Thai language to serve its enterprise customers, said Mr Panutat.
The company is already developing AI-based customer service solutions for a customer to provide answers to inquiries about the customer's products and services, as well as the ability to upsell and cross-sell.
He said the development of solutions should be finished in two months.
AUGMENTED INTELLIGENCE
Ruangroj Poonpol, group chairman of Kasikorn Business-Technology Group (KBTG), said in recent years many businesses have seen AI play a more important role in various work functions.
"The rise of generative AI like ChatGPT has caused a buzz in business circles that humans may soon be replaced by AI," he said.
A cooperative relationship between humans and AI will bring about the highest efficiency and sustainability for work processes, said Mr Ruangroj.
This cooperation results in the "augmented intelligence" concept, where AI is designed to help enhance humans' work and capabilities, not to replace them, he said.
"Through the creation of human-centric machine learning models, AI is trained to receive human feedback and turn that data into guidelines to help improve human decision-making and certain actions to be more accurate and precise," said Mr Ruangroj.
KBTG and MIT Media Lab developed AI through research called K-GPT (Knowledge-GPT), which leverages ChatGPT's capability to provide in-depth domain knowledge by using more natural language in its conversations.
The project enhanced its Thai language proficiency so that it can respond to questions and simultaneously offer advice from different angles.
KBTG Labs and MIT Media Lab also collaborated on proof of concept research called "Kookid", with two conversational agents Kana and Kacha that offer responses from different angles as needed for users' decision-making.
