An artificial intelligence (AI) app that helps children with autism who have difficulty expressing themselves verbally communicate with their parents has been released. Researchers from the Korea Advanced Institute of Science and Technology (KAIST) and Naver AI collaborated.
KAIST announced on the 18th that a research team led by Professor Hong Hwa-jung from the Department of Industrial Design developed an AI-based communication tool called "AAcessTalk" to assist communication between minimally verbal autism (MVA) children and their parents in collaboration with Naver AI Lab and the Dodakim Child Development Center.
Minimally verbal autism refers to cases within the autism spectrum disorder (ASD) where individuals have limited speech. While linguistic proficiency may be significantly low compared to their developmental level, intelligence or social skills may be relatively high. AcessTalk is a tablet-based AI communication system designed to facilitate meaningful conversations between minimally verbal autism children and their parents.
Minimally verbal autism children primarily communicate through non-verbal cues, limited words, or fixed phrases. Parents need to interpret the subtle signals and initiate conversations, encouraging their children to participate in communication. In situations where the child offers little cooperation, many parents express feelings of isolation or frustration due to the one-sided interaction.
Parents use augmentative and alternative communication (AAC) systems, such as symbol cards, to enable their children to engage in communication. This method encourages autistic children to express themselves using visual symbols. However, AAC systems often focus solely on functional communication and may not reflect the user's personal context or true intent. Most AAC systems only provide children with "essential words" chosen by service providers or parents.
AAcessTalk utilizes AI technology to recommend personalized word cards in real-time, reflecting the child's interests and contextual situations. It provides parents with specific conversation guides based on the context. Once parents select a topic to initiate conversation, they can choose example sentences, while the child receives 12 word cards suitable for the situation. When the child selects a word card, the conversation returns to the parent. If a conversation pattern requiring attention emerges, the system alerts the parent.
The research team conducted field studies in 11 households over two weeks. During this period, a total of 232 conversations took place, with the child selecting 2,244 conversation cards. Unlike traditional AAC systems, which only provided fixed word cards, AcessTalk offered a variety of word cards according to the context, making actual conversations more realistic.
One caregiver of a participating autistic child said, "I was surprised when my child used unexpected words, which helped me understand their language abilities more deeply." Another parent noted, "It felt like I was having a real conversation with my child for the first time."
The findings of this study recently received the Best Paper Award at the international academic conference "ACM CHI 2025" held in Yokohama, Japan. ACM CHI is a prestigious conference in the field of human-computer interaction (HCI), and only 1% of the 5,000 submitted papers are awarded the Best Paper Award.
Professor Hong Hwa-jung noted, "This study confirms that AI can not only enhance communication efficiency but also serve as a tool that promotes genuine consolidation and understanding among family members," adding that they plan to expand practical technology applications and user experience-based research for socially vulnerable groups.
References
ACM CHI 2025, DOI : https://doi.org/10.1145/3706598.3713792