A promotional image of A. (SK Telecom)
Named A., pronounced as “ay-dot,” the Korean-language virtual assistant takes on the appearance of a toddler in a raincoat. Powered by a natural language processing model, users will be able to initiate a conversation with the virtual assistant about topics like the weather or ask it to play music or videos. It also offers foreign language support in English, Chinese and Japanese.
“(A.) was developed based on GPT-3 language model, so this enables users to engage in a free-flowing conversation with the assistant,” Lee Hyun-a, vice president and chief product officer of SK Telecom, said in a media briefing Monday.
Lee added SK Telecom is accepting feedback from open beta users so it can help upgrade the virtual assistant, although the language processing engine consists of “tens of millions of variables.”
The launch of A. marks an expansion of SK Telecom’s artificial intelligence platform, Lee also said, as the A. system will be integrated into what serves as the backbone of its language-processing services such as smart speaker NUGU.
Users of the earliest version of A. will be allowed to choose from five different characters, eight voice tones and two modes of Korean language, formal and informal.
SK Telecom CPO Lee Hyun-a speaks at a press briefing Monday. (SK Telecom)
“We expect users to create emotional touch from characters because characters are offering a sense of intimacy based on verbal interactions,” Yea told reporters.
The app can currently search and recommend songs or media content through apps offered by SK Telecom’s affiliates, such as music streaming app Flo, internet-based streaming platform Wavve, outdoor navigation app T Map and web portal Nate.
SK Telecom said the pool of available content that A. is able to process will expand with more collaborations with third-party entities through this year.
Moreover, the multimodal understanding of A. will be available next year, so that it can understand and process input in the form of words and images.
As for the issue of personal data protection and information bias, SK Telecom Chief Technology Officer Lee Sang-ho said the company is “filtering out” sensitive and biased information for A. to process.
The open beta of the voice assistant is available for download for free on Google Play and One Store app store starting Monday, and will be available on Apple’s app store in June.
This comes against the background of a global competition to gain an edge in the virtual assistant market using hyperscale AI models.
Last week, Google announced its latest language processing model laMDA 2 for a more advanced dialogue application, as well as its features that removed the need for users to initiate talks with “Hey Google,” and understand verbal fillers.
Also in Korea, internet giants Naver and Kakao have expressed an interest in the market of hyperscale AI engines that understand and process Korean languages. LG Corp., a holding company of LG conglomerate, has been engaged in projects for multimodal language-processing projects with its engine called Exaone.