Study Identifies Key Design Features for Ethical Children AI Characters

Educationally effective and ethically designed artificial intelligence characters for children aged three to eight require clear learning objectives, developmentally aligned content, semi-structured conversational pathways, balanced individual and shared engagement, and explicit transparency regarding their artificial nature. Qualitative findings published by researcher Sonia Tiwari in the journal AI, Brain and Child demonstrate that the ethical integrity of AI characters depends on concrete design factors spanning child needs, character persona, and interaction mechanics rather than broad abstract principles alone. Children are increasingly exposed to voice assistants, smart toys, and conversational agents, engaging in parasocial relationships that present novel psychological and developmental challenges. Drawing upon Chen and Lin’s POWER framework—which assesses purposeful, optimal, wise, ethical, and responsible technological implementation—the investigation systematically examined how digital, tangible, and hybrid conversational personalities function in early learning contexts, seeking actionable criteria to help learning designers create safe educational tools.

To examine these dynamics across real-world environments, the qualitative investigation gathered and analyzed public documentation from twenty diverse commercial artificial intelligence characters marketed for children aged three to eight or displaying applicable early-childhood use cases. The purposive sample deliberately spanned multiple product formats, including animated screen chatbots, intelligent tutors, social robotics, smart speakers, digital brainstorming tools, and tangible hybrid devices such as Duolingo, Khanmigo, Tolans, Buddy, Ello, and Korro. Rather than relying on controlled laboratory experiments or direct child interviews, the researcher gathered multimodal evidence from official product websites alongside sixty publicly available user-generated social media videos collected from YouTube, TikTok, and Instagram, allocating three distinct unscripted video observations to each character. Products explicitly designed around inappropriate relational dynamics, such as synthetic companions, boyfriends, or girlfriends, were excluded from the sample. This methodology allowed direct observation of authentic home and classroom routines, capturing family reactions, multimodal visual cues, and spontaneous dialogic exchanges while preserving participant privacy through facial blurring and omitting identifiable records.

The collected visual, textual, and video records were systematically classified using the AI Character Assessment framework, abbreviated as the AIC-A framework, which evaluates character implementations across three interconnected dimensions: child factors, character factors, and interaction factors. Employing a deductive manual coding protocol organized within a digital workspace database, the researcher mapped each character’s features against predefined evaluative criteria, linking video timestamps and website documentation to specific analytical categories. For instance, therapeutic movement systems like Korro, designed to assist children managing developmental conditions through interactive play, and animated language tools like Buddy, structured to deliver English instruction, provided foundational benchmark tables for the evaluation. This systematic classification was further interpreted through the lens of Chen and Lin’s POWER principles, synthesizing observed behavioral patterns with existing literature regarding joint media engagement, developmental psychology, and interactive digital interfaces to generate the AI Character Design planning tool, known as AIC-D.

Within the dimension of child factors, the investigation identified that characters designed around narrow, well-defined educational objectives provided substantially greater instructional clarity and emotional predictability for young learners than open-ended agents. Dedicated reading companions like Ello exemplified this purposeful approach by centering the entire interactive interface around early literacy acquisition, utilizing intuitive physical cues—such as an elephant character featuring exaggerated ears that visibly signal active listening while the child reads aloud—to establish explicit behavioral expectations. Furthermore, data security and limited contextual awareness emerged as vital operational components of ethical design. While dedicated educational tutors like Buddy, Ello, Khanmigo, and Duolingo’s Lily restricted their system memory strictly to functional instructional indicators like completed modules and acquired academic skills, robotic companions such as Dino and Miko recorded broader personal preferences, names, and conversational topics. Ethical implementations protected child safety by encrypting stored dialogue records and providing accessible parental management dashboards with direct data deletion capabilities.

Aligning character capabilities with the specific cognitive and emotional developmental stages of children proved essential for maintaining engagement while safeguarding well-being. The analysis highlighted how platforms like Buddy operationalize developmental appropriateness by segmenting instructional objectives and visual presentations according to age tiers spanning preschool, kindergarten, and first grade. Within this architecture, the character visually adapts alongside the child’s developmental maturation, transitioning from a simple avatar displaying a single letter on its chest for preschoolers into an elaborately costumed superhero persona for first graders. Grounding these interface shifts in early learning standards, such as those established by the National Association for the Education of Young Children, ensures that digital interactions match emerging linguistic, emotional, and cognitive capacities. Characters that respect these developmental boundaries scaffold learning effectively without overwhelming young users with complex cognitive demands or emotionally manipulative interactions.

Character-level analysis revealed that all twenty analyzed products intentionally adopted stylized or abstract aesthetic designs rather than photorealistic human forms. Stylized personalities—such as Ello’s illustrated elephant or Duolingo’s cartoon teenager Lily—and abstract entities—such as Miko’s robotic geometry, Tolan’s alien features, or Khanmigo’s expressive digital eyes—consistently evoked positive emotional reactions without associating the character with a specific human race, ethnicity, or biological gender. Social media documentation confirmed that children and parents routinely perceived these visual presentations as cute, humorous, and approachable, minimizing feelings of intimidation and reducing conversational anxiety. These aesthetic strategies directly reflect established psychological research on kawaii design principles, where rounded silhouettes, large expressive eyes, and gentle kinetic behaviors encourage focused attention and emotional warmth. The study noted that while playful abstraction establishes a safe foundation for parasocial interaction, visual charm must remain aligned with age-appropriate content to prevent deceptive superficial friendliness from masking inappropriate outputs.

Effective character persona design required a carefully calibrated balance between supportive mentorship and friendly companionship. Interactive companions like the robotic reading assistant Luka demonstrated this balance by exhibiting warm, playful reactions—such as giggling when a child patted its head or rubbed its belly—while reliably executing structured instructional functions like reading text aloud as the child turned physical book pages. Crucially, ethical character design demanded transparent artificiality, wherein systems explicitly disclosed their programmed, non-human nature rather than encouraging illusions of biological sentience or reciprocal romance. Products like Chatterbox, an educational build-it-yourself smart speaker, and Toybot embedded clear verbal disclosures directly into their conversational audio scripts, defining themselves explicitly as programmed computational tools rather than living companions. By openly stating their functional limitations and mechanical origins, these characters prevented excessive emotional over-attachment, supported healthy user expectations, and fulfilled core ethical recommendations outlined in contemporary youth technology frameworks.

Analysis of the third dimension, interaction factors, revealed that predictable, semi-structured conversational architectures were significantly superior to unconstrained, open-ended dialogues in sustaining instructional focus and preventing cognitive exhaustion. Systems such as Korro’s occupational physical therapy guides, Ello’s literacy feedback mechanics, and Khanmigo’s curricular prompts utilized tightly focused conversational loops to guide attention toward discrete, bite-sized tasks. In observational videos of reading sessions, Ello avoided interrupting children mid-sentence when reading errors occurred, instead offering patient pronunciation corrections after the passage concluded, demonstrating that semi-structured feedback fosters learning without inducing frustration. Similarly, smart speakers like the Amazon Echo Dot Kids and Apple HomePod Mini, alongside integrated voice-book systems like Readyland, demonstrated that simple, lightweight information retrieval interactions—such as answering brief factual queries about weather or animals—adequately supported spontaneous inquiry without requiring extensive data collection, intrusive tracking, or risky open-ended conversational simulations.

The versatility and educational effectiveness of artificial intelligence characters increased substantially when interaction mechanics accommodated both independent child use and collaborative joint media engagement involving caregivers, educators, or peers. Physical and hybrid devices like Toybot permitted flexible switching between solitary board-game play and multi-player family participation using audio cues, while maker tools such as Craft with Oki Pie generated explicit instructional prompts instructing adult educators how to pre-assemble components requiring advanced fine motor skills for kindergarteners. Across the sampled categories, dedicated educational tutors demonstrated the highest developmental quality by combining clear curricular targets with time-limited sessions that prevented compulsive overuse. Hybrid companions like Luka, Readyland, Toybot, and Chatterbox effectively integrated screen-free physical play with collaborative group learning, avoiding the social isolation associated with passive media consumption. Conversely, smart speakers provided safe, structured query responses but lacked systematic guided practice, while adult-mediated brainstorming tools prioritized collaborative parental setup.

To translate these empirical observations into actionable industry practices, the study synthesized its findings into the AI Character Design planning framework, which provides learning designers with structured checklists to evaluate child development targets, persona transparency, and interactive boundary structures before product release. However, the author outlined critical methodological limitations inherent to the study design, noting that the analyzed sample drew predominantly from United States documentation and potentially overlooked cultural nuances across broader global populations. Furthermore, reliance on public user-generated media introduced risks of unacknowledged commercial sponsorships among family reviewers, while the study’s deliberate focus on dedicated child-focused tools excluded high-risk general-market chatbots that frequently lack robust age-gating mechanisms. Future empirical research, including randomized controlled trials evaluating screen-free versus screen-based interfaces, realistic versus abstract aesthetics, and presence versus absence of transparent artificiality disclosures, will be necessary to substantiate these preliminary qualitative design principles across broader learning environments.

Subject of Research: Ethical design factors for artificial intelligence characters in early childhood education

Article Title: Designing ethical AI characters for children’s early learning experiences

Article References: Tiwari, S. (2025). Designing ethical AI characters for children’s early learning experiences. AI, Brain and Child, 1(1), Article 11. https://doi.org/10.1007/s44436-025-00015-1

Image Credits: AI Generated

DOI: 10.1007/s44436-025-00015-1

Keywords: Artificial Intelligence, Early Childhood Education, Ethical AI Design, AI Characters, Joint Media Engagement, Child Development, EdTech, Human-Computer Interaction, Designing, ethical, characters, children