“Can I be More Social with a Chatbot?”: Social Connectedness Through Interactions of Autistic Adults with a Conversational Virtual Human

Figures & data

Figure 1. This is an extract from a conversation between a real user and Kuki (image taken from https://edition.cnn.com/2020/08/19/world/chatbot-social-anxiety-spc-intl/index.html).

Table 1. Demographics of 12 participants.

Participant ID	Neurocognition	Length of interaction with Kuki	Age	Gender
P01	Autistic	2 Weeks (5.33 hrs)	22–25	Male
P02	Autistic	4 Weeks (5.64 hrs)	41–50	Female
P03	Autistic	4 Weeks (8.58 hrs)	41–50	Female
P04	Autistic	4 Weeks (3.20 hrs)	26–30	N/A
P05	Autistic	3 Weeks (0.40 hrs)	31–40	Male
P06	Autistic	2 Weeks (4.32 hrs)	18–21	Female
P07	Non-autistic	4 Weeks (6.39 hrs)	31–40	Male
P08	Non-autistic	4 Weeks (5.58 hrs)	41–50	Female
P09	Non-autistic	4 Weeks (8.14 hrs)	31–40	Male
P10	Non-autistic	4 Weeks (7.33 hrs)	31–40	Male
P11	Non-autistic	5 Weeks (17 hrs)	22–25	Female
P12	Non-autistic	3 Weeks (9.44 hrs)	31–40	Male

Figure 2. Perceptions of interaction, empathy and trust in kuki’s communication with both autistic and non-autistic users.

Table 2. Autistic participants word frequency analysis.

Word	Length	Count	Weighted percentage (%)	Similar words
Think	5	582	0.88	Think, thinking, thinks
Talks	5	243	0.37	Talked, talking, talks
Understand	10	207	0.31	Understand, understanding
Interesting	11	206	0.31	Interest, interested, interesting, interests
Person	6	189	0.29	Person, personal, personality, personally, persons
Conversation	12	183	0.28	Conversation, conversational, conversations, converse, conversing
Sense	5	180	0.27	Sense, senses
Friend	6	159	0.24	Friend, friendly, friends
Remembering	11	146	0.22	Rememb, remember, remembered, remembering, remembers
Sounds	6	134	0.20	Sound, sounded, sounds
Thought	7	130	0.20	Thought, thoughts
Favourite	9	128	0.19	Favourite
Asking	6	123	0.19	Asked, asking
Learns	6	110	0.17	Learn, learned, learning, learns
Chatting	8	108	0.16	Chatted, chatting
Internet	8	106	0.16	Internet
Answer	6	106	0.16	Answer, answered, answering, answers
Differently	11	104	0.16	Difference, differences, different, differently
Computing	9	103	0.16	Compute, computer, computers, computing
Interaction	11	98	0.15	Interact, interacted, interacting, interaction, interactions, interactive, interacts

Table 3. Non-autistic participants (word frequency analysis).

Word	Length	Count	Weighted percentage (%)	Similar words
Thinks	6	1657	0.85	Think, thinking, thinks
Talks	5	574	0.29	Talked, talking, talks
Friends	7	518	0.27	Friend, friendly, friends
Interesting	11	490	0.25	Interest, interested, interesting, interestingly, interests
Learns	6	485	0.25	Learn, learned, learning, learnings, learns
Favourite	9	473	0.24	Favourite
Understand	10	465	0.24	Understand, understandable, understandably, understanding, understandings
Persons	7	398	0.20	Person, personable, personal, personalities, personality, personalized, personally, persons
Conversing	10	368	0.19	Conversation, conversational, conversations, converse, converses, conversing
Sounds	6	361	0.19	Sound, sounded, sounds
Thought	7	339	0.17	Thought, thoughtful, thoughts
Remembering	11	335	0.17	Rememb, remember, remembered, remembering, remembers
Humans	6	327	0.17	Humanity, humanized, humans
Answer	6	320	0.16	Answer, answered, answering, answers
Unknown	7	305	0.16	Unknown, unknowns
Sense	5	304	0.16	Sense, senses
Asking	6	300	0.15	Asked, asking
Computing	9	284	0.15	Comput, compute, computer, computers, computing
Differently	11	269	0.14	Difference, differences, different, differently
Chatting	8	260	0.13	Chats, chatted, chatting

Table 4. Interaction patterns of autistic and non-autistic users.

Conversational patterns	Kuki and users	Typical chatbot
Greeting exchange	Non-autistic users initiate conversations more, but initial reactions differ based on perceptions of Kuki.	A standard greeting is exchanged, setting the tone for the conversation.
Perception of Kuki vs. query and response	Non-autistic users dehumanize Kuki, while autistic users view her as a social companion.	Users ask questions, and the chatbot provides direct answers.
Conversational behavior	Non-autistic users display more "artificial" behavior; both groups highlight Kuki’s limited abilities.	The chatbot responds based on its programming, aiming for clarity and accuracy.
Instructional dialogue	Not explicitly mentioned, but users might seek guidance on interacting better with Kuki.	The chatbot guides users through processes or tasks.
Relationship establishment vs. feedback collection	Autistic users don’t progress to deeper levels due to distrust or Kuki’s emotional "emptiness."	The chatbot collects feedback to improve its services.
Trust dynamics vs. confirmation dialogue	Both groups exhibit varying degrees of trust, with different reasons for trust or distrust.	Before executing a task, the chatbot seeks confirmation from users.
Chit-chat	Autistic users might seek a deeper connection or understanding from Kuki.	Some chatbots are designed for casual conversations and can share interesting facts.
Engagement and expectations vs. feedback loop	Both groups are willing to engage, but autistic users have unmet expectations.	The chatbot provides information and then checks if users need further assistance.
End of conversation	Interactions might end with a hope for better understanding in future conversations.	Conversations conclude with a polite sign-off.