Big brains and deep pockets going big into AI
Tech companies are rushing to get ahead in the race to develop artificial intelligence (AI) systems with enhanced efficiency.
Despite critics suggesting the world has entered an AI winter, with McDonald's selling its Dynamic Yield AI subsidiary after disappointing results, they are investing with abandon.
Korean companies are aiming to develop a so-called "hyperscale" AI, an industry term referring to a system with more parameters and better capacity to process huge amounts of data.
Compared to Google's AlphaGo, the AI Go player that faced off against Korean Go master Lee Se-dol in 2016, a hyperscale AI model is hundreds to thousands times more efficient.
Domestic IT companies are ramping up investment to take the lead in the prospective market.
LG Corp.'s hyperscale AI engine Exaone was used to develop images utilized in designs featured during New York Fashion Week. Fashion designer Park Yoon-hee used the images to create the "Flowers on Venus" collection.
Naver is set to start a trial run of its CLOVA CareCall service in March. The CLOVA CareCall is a phone call service powered by the company's HyperCLOVA system. It will check in on 100 elderly people living alone in Daegu through phone calls once or twice a week. The service aims to provide a safety net as well as emotional support for the elderly, according to Naver.
The global AI market will reach $554.3 billion by 2024, a 69 percent growth compared to 2021, according to an IDC report released in February 2021. Grand View Research predicted that the market size will break $1 trillion by 2028.
Google introduced the Switch Transformer 1.6-trillion-parameter AI model in January 2021, while Microsoft and Nvidia jointly developed Megatron with 530 billion parameters in October. DeepMind Technologies, Alphabet's AI subsidiary headquartered in London and the developer of AlphaGo, introduced a 280-billion-parameter AI model Gopher last December. The Beijing Academy of Artificial Intelligence announced a 1.75-trillion-parameter Wu Dao 2.0 in June.
It is a huge leap from Google's Meena, an AI-based conversational chatbot with 2.6 billion parameters, or a 175-billion-parameter GPT-3 of Tesla's OpenAI. Both models were released in 2020.
In Korea, tech companies such as LG Corp., Naver and Kakao are leading the market.
LG Corp.'s Exaone has 300 billion parameters, Naver's HyperCLOVA 204 billion and Kakao's KoGPT 30 billion. Exaone and HyperCLOVA machine-learned some 560 to 600 billion word tokens, which equals fifty years worth of news articles on Naver.
"It is true that Korea is behind the United States and China in the global market," said an industry insider, "but we are quickly catching up with the front runners in every aspect including parameter and technology application."
Marketing is underway.
Naver adopted the hyperscale AI system in its search engine service last year. Kakaobrain, an AI subsidiary of Kakao, released minDALL-E in November, which is an AI-based drawing program that automatically creates an image based on what the user says.
LG AI Research introduced the Expert AI Alliance in collaboration with twelve companies — EBS, Woori Bank, LG U+, LG CNS, LG Chem, LG Electronics, Google, Shutterstock, VA Corporation, Elsevier, Korea University Medicine, and Hanyang University Medical Center — on Feb. 22.
"We established a foundation for the global hyperscale AI ecosystem, where Exaone and its partners can grow together and innovate the business models," said Bae Kyung-hoon, chief of LG AI Research.
Telecommunication companies are jumping on the bandwagon as well.
SK Telecom established a hyperscale AI development task force named Apollo last May, while KT organized the AI One Team with a goal to market an AI-based service this year. Woori Bank is collaborating with LG Corp. to develop an AI banker.
"Hyperscale AI technology has bigger extensibility than the existing AI models," said Lee Hwa-yeong, LG AI Research vice president, "and will be adopted in a range of industries and services such as customer counseling, banking and medical counseling."
BY KIM TAE-YUN [email@example.com]