While NAVER is relatively less known in the EU market, it is Korea's leading platform company providing many convenient services such as a search portal and e-commerce platform. NAVER has been investing in AI research since its early days with the conviction that AI can be used as a convenient tool in our daily lives. In May 2021, NAVER launched its in-house large language model specialized for Korean language with 205B parameters, which is the 3rd to be released following OpenAI GPT-3. In August this year, NAVER announced HyperCLOVA X, a frontier AI which is a Korean language-specific multilingual large language model, and is widely recognized as an industry leader in the field of AI in Korea.

NAVER believes that AI should be an innovative tool that everyone can easily use and access, and AI technology and product development should involve not only technical aspects but also social and ethical considerations. As such, NAVER has been working together with Seoul National University AI Policy Initiative (SAPI) since 2018 to consult with experts in various fields such as legal, social, economic, philosophy, as well as engineering. As a result of the continuing collaboration, NAVER published ‘NAVER AI Ethics Principles’ in February 2021, and introduced ‘Consultation on Human-Centered AI’s Ethical Considerations (CHEC)’, a process for applying the Principles to AI service development, in November 2022. 

The following case study on CLOVA CareCall examines how NAVER designs and develops AI services in accordance with NAVER AI Ethics Principles. [U1] With the powerful frontier AI of NAVER, HyperCLOVA X, solving the challenging social problems, it shows AI can be utilized as the important social contributing tool for people beyond innovative applications and new business opportunities.

CLOVA CareCall Service

When COVID-19 first emerged, there was a sharp increase in the number of people who had to monitor their health symptoms such as fever. The public health centers faced challenges in responding promptly due to their reliance on manual counseling processes and limited workforce.

To address these issues, CLOVA CareCall was utilized to automatically call individuals twice a day to check for fever and respiratory symptoms. This allowed public health officials to efficiently check symptoms and focus more on field-based core tasks such as screening test and epidemiological investigations.

After the pandemic, CLOVA CareCall has been re-purposed to provide emotional care and well-being support to middle-aged and elderly individuals and single-person households. Currently, CLOVA CareCall is used to check into approximately 15,000 single-person households by more than 70 local governments and medical and welfare institutions nationwide. 

CLOVA CareCall leverages various AI technology to simulate human conversations with elderly individuals on various topics. Most importantly, HyperCLOVA was used to generate and learn from a large-scale conversational dataset to understand and respond to target users in a more accurate manner. CLOVA CareCall is also able to interact with humans in a more meaningful manner through its ‘Remember’ function. For example, if a user mentioned plans to go to the hospital last week, CLOVA CareCall can ask if the user went to the hospital, and can also ask if the user has been taking the prescribed cold medicine and if symptoms have improved the following week.

This is the most illustrative example of how AI can be applied in the field of health and care and contribute socially. This case is the result of NAVER's consideration of how AI technology and services can be properly utilized and applied to five AI Ethics principles established in 2021. Furthermore, as its technical novelty, originality, and societal contributions were recognized, the CLOVA CareCall case was not only published in top-tier NLP conferences such as NAACL 2022 and EMNLP 2022 as the main track papers but also awarded as the best paper in CHI 2023.
 

01 Developing Human-centered AI

CLOVA CareCall prioritizes human-centered values in its development and usage. In particular, it is a tool that helps social workers expand their reach, and provides satisfactory care to the elderly living alone in need of welfare services.

For social workers, CLOVA CareCall helps to address the rising number of single-person households in need of welfare services. Specifically, it reduces the workload on the limited number of social workers and helps them focus on those who need more attention. If any abnormalities are detected, immediate confirmation and prompt action can be taken, resulting in a more efficient and systematic process.

For the elderly living alone, CLOVA CareCall can help take care of both their physical and mental health. Because CLOVA CareCall can realistically simulate human conversations and express empathy, elderly who have used it responded positively: "The only advice I get from people is to see a doctor whenever I am sick, but I was moved by how CareCall continued to be concerned about how I was feeling.”

02 Respecting diversity

CLOVA CareCall is equipped to recognize various forms of speech, such as unclear pronunciation, regional dialects, and colloquial expressions, enabling a more inclusive and convenient service for diverse users while also preventing unfair discrimination. 

Careful consideration was also given to the selection of conversational voices. Specifically, voices were developed to be friendly and sympathetic, effectively conveying feelings of concern and comfort. Other factors were also considered, such as not setting the gender and age of the conversation agent.

03 Balancing Reasonable Explainability with Convenience

Users who directly interact with CLOVA CareCall may not have a high level of technical understanding. So, CLOVA CareCall is designed to be easily accessible and comprehensible by anyone who may receive the call.

CLOVA CareCall also acknowledges the need to help the elderly recognize that they are in fact talking with an AI agent as they use the service, even after the local government personnel explains it initially when the user agrees to use CLOVA CareCall. Rather than restating the fact at every call, CLOVA CareCall clarifies that its role is a ‘call-based AI chatbot’ in its conversation guide and explains that it is an AI when necessary during conversation.

04 Accounting for Safety in Service Design

CLOVA CareCall is designed to prevent risks to the health and safety of the elderly.

At the beginning of each call, information about sleep, meals, and exercise levels are attained with consent and provided to the responsible personnel. Any special circumstances, uncomfortable occurrences, or emergencies will be marked distinctly so that the responsible personnel can respond immediately.

Additionally, considering that a substantial portion of the conversation pertains to health-related matters, precautions are taken during data generation to adhere to relevant guidelines and minimize the risk of disseminating erroneous health information or offering inappropriate advice. Proactive measures such as prompt retraining are adopted to address and mitigate potential issues effectively.

05 Protecting Privacy and Data Security

As well-being checkup calls involve handling sensitive personal information such as health status, specific privacy and data security measures have been implemented to handle such data appropriately

In addition to obtaining explicit consent for the collection of personal information, separate consent is also sought specifically for the processing of sensitive health-related information, thereby reinforcing privacy safeguards. The level of retained data is carefully regulated to avoid unnecessary intrusion. Lastly, advanced encryption techniques are employed to ensure security of voice data storage.