Univ. of Kansas Spinout VOISS Cuts Edtech Churn via Student Co-Design

User-Centered Design Powers Better AI With University of Kansas — Ai4

Ai4

Academic researchers must commercialize evidence-based tools through direct classroom co-design, because unvalidated startup rollouts trigger irreversible teacher churn while grant-funded laboratory projects die without commercial scale.

AI-assisted summary of User-Centered Design Powers Better AI With University of Kansas — original source. Check the original for full context.

Source published: 2026-08-14

Detailed notes

## Background Published on YouTube by Ai4 on August 14, 2026, this presentation features Amber Rowland, an Associate Research Professor at the University of Kansas and Chief Science Officer at VOISS. According to the page description, the session explores how integrating iterative user experience research, user co-design, and evidence-based learning science yields transformative AI edtech tools, using the commercialized social skills simulation platform VOISS as a blueprint.

## TL;DR Amber Rowland outlines the development and design methodology behind VOISS, an AI-powered simulation platform created to help neurodivergent adolescents practice critical social communication skills. Addressing the tension between academic rigor and startup velocity, Rowland warns that shipping rushed, unvalidated minimum viable products leads to catastrophic customer churn among time-strapped educators. She presents the 'Confidence to Scale Helix,' a six-step co-design framework that pairs classroom observations, student paper prototyping, and live developer coding sessions to iteratively earn product confidence. While AI handles roughly 80% of multi-device technical stress testing, she emphasizes that the remaining 20% must remain anchored in lived human experience. Rowland concludes with a call to action for university researchers to actively commercialize grant-funded innovations so proven interventions can endure and reach students in need.

## Atmosphere & style The presentation carries an authentic, pragmatic tone that balances academic authority with the gritty realities of startup commercialization. Rowland speaks with candid warmth and relatable humor, using self-deprecating remarks about researchers being 'slow' and Kansas tornado metaphors to keep technical and pedagogical discussions lively. Grounded by the poignant developmental journey of Oliver, a middle schooler finding his voice, the talk feels deeply empathetic, mission-driven, and refreshingly honest about the limits of technology without human-centered oversight.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=zilmb9q5BKM&t=4s) Building edtech on untested top-down assumptions leads to classroom failure, making direct user-centered design essential for real-world adoption. - [02:24](https://www.youtube.com/watch?v=zilmb9q5BKM&t=144s) The startup instinct to ship fast and fix later causes costly customer churn in education, as burned educators rarely return to broken software. - [02:24](https://www.youtube.com/watch?v=zilmb9q5BKM&t=144s) Developing effective AI tools requires active translation between pedagogical learning sciences and machine learning engineers so educational integrity is not compromised for speed. - [04:40](https://www.youtube.com/watch?v=zilmb9q5BKM&t=280s) The 'Confidence to Scale Helix' model illustrates how iterative co-design loops expand outward in reach and investment while building upward in design confidence. - [04:40](https://www.youtube.com/watch?v=zilmb9q5BKM&t=280s) AI-driven virtual simulations provide a psychologically safe environment for neurodivergent students to practice social communication and experience realistic peer consequences with zero real-world harm. - [06:58](https://www.youtube.com/watch?v=zilmb9q5BKM&t=418s) The six-step co-design pipeline (Listen, Confirm, Co-Design, Test, Refine, Scale) systematically filters raw stakeholder feedback through pedagogical and feasibility criteria before large-scale deployment. - [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) Placing software engineers and machine learning developers directly into classrooms alongside students transforms how usability issues are identified and resolved. - [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) Edtech creators must avoid a 'tech savior mentality' and remain transparent about human-centric safeguards to overcome student and educator skepticism toward AI. - [14:05](https://www.youtube.com/watch?v=zilmb9q5BKM&t=845s) While AI agents can automate roughly 80% of technical and multi-device stress testing, human testing remains irreplaceable for assessing emotional resonance and lived user experience. - [14:05](https://www.youtube.com/watch?v=zilmb9q5BKM&t=845s) Supporting multiple hardware platforms (Chromebooks, iPads, VR) is essential for accessibility, as many neurodivergent learners experience sensory aversion to facial VR headsets. - [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) Partnering with university research faculty strengthens commercial edtech by integrating peer-reviewed learning science into product development. - [19:06](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1146s) Academic researchers must actively pursue commercialization to prevent impactful, grant-funded educational technologies from dying when initial grant cycles conclude.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=zilmb9q5BKM&t=4s) Introduction and the Story of Oliver

Amber Rowland, an Associate Research Professor at the University of Kansas and Chief Science Officer (CSO) of VOISS, introduces the mission behind her work: creating extended reality simulations powered by AI to help adolescents learn and practice critical social skills. Grounded in user-centered design (UCD), the session focuses on how authentic user involvement ensures educational technology actually works in real classrooms. To illustrate why this matters, Rowland shares the story of Oliver, a middle school student in Denver, Colorado. Oliver struggles daily with social skills, anxiety, and communication—challenges that are especially daunting during middle school, an already awkward developmental period. His family worries deeply about his long-term prospects, such as whether he will be able to make friends, sustain relationships, or hold down a job.

These personal struggles take place against the backdrop of a nationwide workforce shortage among special education teachers and therapists, alongside a rising number of children diagnosed with anxiety and social communication needs. Rowland notes that VOISS was born out of deep frustration with top-down, district-mandated technology that routinely fails in practice: "I would sit down and try to do something and it wouldn't work the way I think it does. Like, did you even have teachers try this before you put it out? Drive me nuts." She argues that building edtech or AI products on untested assumptions about what kids or educators need is a recipe for failure, whereas user-centered design provides the empirical confidence needed to build tools that genuinely work.

- **Primary User Profile**: Oliver, a middle schooler with autism and anxiety in Denver, Colorado. - **Systemic Challenges**: Severe shortages of specialized educators/therapists combined with surging rates of youth anxiety. - **Core Stance**: Without direct user-centered design, developers make faulty assumptions that derail classroom adoption.

---

[02:24](https://www.youtube.com/watch?v=zilmb9q5BKM&t=144s) The Tension Between Research Rigor and Startup Speed

Operating at the intersection of academia and entrepreneurship reveals a sharp cultural contrast between research validity and startup velocity. In an academic research environment, Rowland’s role is to evaluate every possible failure mode and success factor to ensure absolute validity, reliability, and instructional integrity. While she self-deprecatingly refers to her exhaustive approach as "slow," her startup CEO prefers to call it "thorough." In a startup environment—VOISS having spun out into a commercial venture about a year prior—the prevailing instinct is to launch a minimum viable product (MVP) as fast as possible into the wild and patch mistakes later.

```mermaid flowchart TD accTitle: Academic Rigor vs. Startup Velocity subgraph Academic["Academic Research Mindset"] A1["High Validity & Reliability"] --> A2["Exhaustive Evaluation"] A2 --> A3["Thorough / Methodical Pace"] end subgraph Startup["Startup Mindset"] S1["Speed to Market"] --> S2["Rapid MVP Deployment"] S2 --> S3["Fix Later in the Wild"] end subgraph Result["User-Centered Design Bridge"] R1["Prevent Expensive Customer Churn"] R2["Translate Between Pedagogy & ML Engineers"] end Academic --> Result Startup --> Result ```

Rowland points out that relying on post-launch fixes in the education sector is an extraordinarily expensive strategy due to customer churn. Educators have limited time, high stakes, and very little patience for broken software; once burned by a flawed product, they rarely return. To bridge this divide, Rowland acts as an active translator between classroom realities (teachers and students) and technical specialists (designers, software engineers, and machine learning developers), ensuring that foundational learning science is never compromised for speed.

- **The Churn Risk**: Educators are an unforgiving user base; releasing buggy MVPs leads to immediate abandonment and reputational damage. - **The Translator Role**: Bridging content knowledge, classroom pedagogy, and machine learning engineering. - **Commercialization Status**: Spun out from the University of Kansas accelerator program with active commercial deployment.

---

[04:40](https://www.youtube.com/watch?v=zilmb9q5BKM&t=280s) The Confidence to Scale Helix & Grounding in Learning Science

To represent how product validation expands over time, Rowland developed the "Confidence to Scale Helix." When she originally drafted the model, she called it the "spiral," but user feedback quickly flagged that spirals have negative, out-of-control connotations—especially in Kansas, where spirals evoke tornadoes. The model was rechristened a "helix" to convey that each iterative cycle expands outward (engaging more users, requiring more investment, and driving more revenue) while simultaneously rising upward in design confidence. This upward trajectory guarantees that by the time a product reaches broad distribution, word-of-mouth adoption is propelled organically by educators and administrators recommending it to their peers.

```mermaid flowchart LR accTitle: Confidence to Scale Helix Cycle subgraph Helix["Confidence to Scale Helix Cycle"] direction TB L["1. Listen"] --> C["2. Confirm"] C --> CD["3. Co-Design"] CD --> T["4. Test"] T --> RF["5. Refine"] RF --> SC["6. Scale"] SC --> L end ```

The underlying architecture of VOISS is grounded in rigorous academic research and randomized controlled trials (RCTs). It embeds evidence-based practices including social narratives, direct instruction, and simulated role-playing. In the virtual environment, students encounter realistic peer interactions—such as joining a basketball game. If a student makes an anti-social move, like stealing the ball and running to the opposite end of the court, the virtual peers react with realistic frustration. The student is then given immediate opportunities to replay the scenario, try alternate strategies, and achieve positive outcomes in a psychologically safe environment where "no kids harmed in the making of these social skills."

- **Evidence-Based Pillars**: Direct instruction, social narratives, and simulated role-play. - **Simulation Feedback Loop**: Real-time social consequence modeling paired with infinite, zero-consequence retries. - **Iterative Growth**: Expanding user numbers and investment with each loop of the helix.

---

[06:58](https://www.youtube.com/watch?v=zilmb9q5BKM&t=418s) The Co-Design Process and the 7-Step Cycle

The founding team behind VOISS combines deep academic and commercial expertise, featuring Amber Rowland, Maggie Mosher, Bruce Frey, Sean Smith, and CEO Paul, who was recruited via the University of Kansas accelerator to drive marketing and commercialization. Backed by a $4.5 million research grant, the team initiated their design cycle with a structured, multi-day think tank. They brought together educators, researchers, domain experts, and engineers for a day and a half to map out core pain points in social communication, organizing participants into both cross-disciplinary and role-alike working groups.

```mermaid flowchart TD accTitle: VOISS Seven Step Research to Scale Pipeline subgraph Discovery["1. Discovery & Validation"] S1["Listen: $4.5M Think Tank Brainstorming"] --> S2["Confirm: Filter Transcripts & Validate with Users"] end subgraph Design["2. Collaborative Creation"] S3["Co-Design: Paper Prototyping & Student Scenarios"] --> S4["Test: Button-Based vs. AI-Driven Testing"] end subgraph Deployment["3. Refinement & Growth"] S5["Refine: Live Developer Iterations"] --> S6["Scale: Multi-School RCT Deployments"] end Discovery --> Design Design --> Deployment ```

The co-design pipeline systematically processes stakeholder feedback from raw transcription to deployment: 1. **Listen**: Capturing every conversation during stakeholder think tanks to identify genuine pain points rather than assumptions. 2. **Confirm**: Filtering transcripts for feasibility, pedagogical evidence, and ROI, then reflecting the summary back to educators to confirm: "This is what we think we heard." 3. **Co-Design**: Partnering directly with students and teachers using paper prototypes (often accelerated by AI generation) for character and narrative arcs. 4. **Test**: Testing early scenarios with middle schoolers, including field research in Hawaii comparing a button-driven interface against an AI-driven conversational interface. 5. **Refine**: Having programmers implement real-time code updates directly alongside participating students. 6. **Scale**: Rolling out the validated platform across 10 schools in a Year 5 Randomized Controlled Trial (RCT).

A video testimonial illustrates the impact on Oliver, who shares that before using VOISS he rarely spoke to anyone at school out of fear and nervousness, but regular simulation practice taught him practical ways to initiate conversations without feeling overwhelmed. At Oliver's specialized school, students use the platform two to three times per week for 20 minutes per session to systematically reinforce social communication habits.

---

[10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) Practical UCD Strategies in EdTech

Rowland outlines several concrete, tactical practices that make user research effective in school environments:

- **Student & Teacher Screen Recordings**: Training classroom users to trigger screen recordings whenever a glitch or odd behavior occurs, eliminating the ambiguity of second-hand verbal descriptions. - **Developers in the Classroom**: Placing software engineers and ML developers directly in the room to observe students interacting with the platform in real time, which Rowland calls a transformative "game-changer." - **Task-Based Surveys**: Replacing vague "how did you like it?" surveys with rigid, step-by-step task scripts followed by targeted diagnostic questions. - **Pedagogical Oversight**: Requiring learning science and content experts to review every code iteration so features do not drift away from evidence-based practices. - **Direct Research Citations**: Actively publishing research studies and listing academic citations directly on `voiss.org` to prove efficacy.

Addressing the cultural climate in schools, Rowland warns against the "tech savior mentality." Many K-12 and higher education institutions exhibit strong pushback against artificial intelligence, with students frequently remarking that they refuse to use something simply because it contains AI. Drawing on insights from AI leader Pam Snyder, Rowland emphasizes that developers must be transparent and "loud about how you're ensuring it's responsible and human-centric." Furthermore, she highlights co-design methodologies from Sesame Workshop, such as scenario storyboarding and the "Meet the Martian" exercise—where children must explain concepts to someone dressed as an alien who has no prior knowledge of Earth—to elicit precise, assumption-free design specifications.

---

[14:05](https://www.youtube.com/watch?v=zilmb9q5BKM&t=845s) Balancing AI Speed with the Human Lived Experience

While artificial intelligence dramatically accelerates the design and validation pipeline, Rowland stresses that it must augment rather than replace human involvement. VOISS leverages large language models (LLMs) and autonomous agents for roughly 80% of technical validation tasks—conducting rapid stress testing, checking multi-device accessibility, discovering software edge cases, and proposing automated bug mitigations. This automated testing runs across all supported form factors, including iPads, Chromebooks, and Oculus VR headsets (an essential flexibility since many neurodivergent students with autism experience sensory aversion to wearing VR hardware on their faces).

However, the remaining 20% of testing must stay strictly human. Automated agents cannot replicate the subjective lived experience, emotional resonance, and unvarnished feedback of a child or teacher engaging with learning software. Maintaining this human-centered foundation creates a robust commercial advantage: investors gain confidence through measurable risk reduction, while school districts, parents, and enterprise buyers are significantly more receptive to products backed by validated, research-based learning science.

[15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) Partnering with Higher Education and Academic Research Publications

Adopting user-centered design and evidence-based learning science benefits both research rigor and the corporate profit and loss statement. While industry leaders often point out that edtech developers do not spend enough time directly engaging teachers on AI integration—a sentiment echoed by keynote speaker Fei-Fei Li—faculty at the University of Kansas School of Education, including Amber Rowland, Lisa Dieker, and Maggie Mosher, are actively bridging this gap. Rowland urges edtech teams to collaborate closely with higher education researchers who specialize in pedagogical integration and iterative design to replicate these outcomes at scale.

```mermaid flowchart LR accTitle: Confidence to Scale Helix Overview subgraph ResearchPhase["Evidence & Research"] A["Learning Science"] --> B["Iterative User Co-Design"] end subgraph ValidationPhase["Evaluation & Impact"] B --> C["Peer-Reviewed Rigor"] C --> D["Commercialization & Scale"] end ```

To help practitioners dive deeper into the literature, Rowland highlights an upcoming special issue of the *Journal of Special Education Technology*, where she serves as lead guest editor. The issue brings together 11 distinct publications demonstrating user-centered AI integration across various subject areas, including math, science, reading, writing, and social skill development.

- **Special Issue Overview:** The *Journal of Special Education Technology* special issue features peer-reviewed research across multiple disciplines, with early access available online ahead of the official print release. - **"Fewer Clicks Matter":** Rowland’s featured paper documents the user-centered design case study behind their team's first chatbot, translating the presentation's core methodologies into a formal research framework. - **Biometric Research:** The collection includes work by Lisa Dieker exploring the application of biometrics with student populations in educational settings.

---

[17:26](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1046s) Industry Resources, Edtech Podcasts, and the Confidence Scorecard

Rowland points to several key external resources that reinforce user-driven development and learning sciences. She highlights the Tools Competition, an initiative where her team advanced to the third round. While they were not awarded funding in that specific cycle, the competition’s recent Trends Report offers invaluable insights into what makes edtech products and startups fundable, specifically emphasizing robust learning sciences and user-centered design.

```mermaid flowchart TD accTitle: Edtech Ecosystem Resources subgraph Competitions["Funding & Trends"] A["Tools Competition Trends Report"] end subgraph Media["Teacher & Design Podcasts"] B["AI Advocates Podcast (3-5 min tool walkthroughs)"] C["Co-Design with Kids (codesignwithkids.org)"] end subgraph Assessment["Implementation Tools"] D["Confidence to Scale Helix Scorecard (One-page team self-audit)"] end ```

To assist educators and product teams in staying informed and evaluating their own practices, several practical tools and media channels are available:

- **AI Advocates Podcast:** Hosted by Maggie Mosher and Lisa Dieker, this show delivers concise 3-to-5-minute episodes guiding educators through specific artificial intelligence tools and practical classroom integration strategies. - **Co-Design with Kids:** Run by the team at codesignwithkids.org, this initiative and its accompanying podcast provide practical frameworks for including youth directly in product co-design. - **Confidence to Scale Helix Scorecard:** A one-page self-assessment rubric developed by Rowland that allows teams to quickly score their level of user-centered design maturity across every development phase. Teams can request the scorecard by contacting Rowland directly via email.

---

[19:06](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1146s) The Imperative of Commercialization: Oliver's Story and Beyond

Rowland concludes by grounding the entire design philosophy in its real-world human impact, reflecting on the story of a student named Oliver. Video footage and feedback from both Oliver's parents and his teachers show the transformative personal growth and empowerment he achieved after using VOISS. For Rowland, this transformation illustrates why academic researchers must move beyond traditional publishing to embrace product commercialization.

Without taking the deliberate step to commercialize, impactful interventions developed in universities risk dying as soon as their initial grant funding concludes. Rowland issues a call to action for more higher education faculty to bring evidence-based technologies out of the laboratory and directly into the commercial marketplace, ensuring that students everywhere can access tools that foster genuine behavioral and social growth.

## Notable quotes > [00:04](https://www.youtube.com/watch?v=zilmb9q5BKM&t=4s) I would sit down and try to do something and it wouldn't work the way I think it does. Like, did you even have teachers try this before you put it out? Drive me nuts. — Amber Rowland

> [02:24](https://www.youtube.com/watch?v=zilmb9q5BKM&t=144s) slow — Amber Rowland

> [02:24](https://www.youtube.com/watch?v=zilmb9q5BKM&t=144s) thorough — Paul

> [04:40](https://www.youtube.com/watch?v=zilmb9q5BKM&t=280s) no kids harmed in the making of these social skills. — Amber Rowland

> [06:58](https://www.youtube.com/watch?v=zilmb9q5BKM&t=418s) This is what we think we heard. — Amber Rowland

> [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) game-changer — Amber Rowland

> [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) how did you like it? — Amber Rowland

> [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) tech savior mentality — Amber Rowland

> [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) loud about how you're ensuring it's responsible and human-centric. — Pam Snyder

> [10:46](https://www.youtube.com/watch?v=zilmb9q5BKM&t=646s) Meet the Martian

> [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) Fewer Clicks Matter — Amber Rowland

## Mentioned Amber Rowland, VOISS, University of Kansas, Oliver, Maggie Mosher, Bruce Frey, Sean Smith, Paul, Pam Snyder, Sesame Workshop, Fei-Fei Li, Lisa Dieker, Journal of Special Education Technology, Tools Competition, AI Advocates Podcast, Co-Design with Kids, iPad, Chromebook, Oculus VR

## Video description Amber Rowland - Associate Research Professor, University of Kansas

Effective AI-enabled edtech starts with users. Join us to explore how our team combines iterative user experience research, user co-design, and evidence-based learning science to create AI products that are intuitive, effective, and drive customer retention. We will share practical takeaways through the story of VOISS, our newly commercialized AI-powered simulation platform that helps adolescents build critical social skills. Grounded in academic research and proven success, you’ll see how our design process yields transformative tools and offers attendees a model for scalable, user-driven innovation. Whether you are developing AI tools, conducting research, or shaping learning experiences, this session will give you actionable strategies for building technology that is relevant, effective, and loved by users.

[Source](https://www.youtube.com/watch?v=zilmb9q5BKM)

中文简报

堪萨斯大学衍生企业VOISS通过学生协同设计降低教育科技流失率学术研究人员必须通过直接的课堂协同设计将循证工具商业化，因为未经实证检验的初创产品推广会引发不可逆的教师流失，而缺乏商业化规模的资助实验室项目则会走向消亡。 Amber Rowland 概述了 VOISS 背后的开发与设计方法论。VOISS 是一款人工智能驱动的模拟平台，旨在帮助神经多样性青少年练习关键的社交沟通技巧。针对学术严谨性与初创速度之间的冲突，Rowland 警告称，仓促推出未经实证检验的最小可行产品会导致时间本就紧迫的教育工作者出现灾难性的客户流失。她提出了“规模化信心螺旋”（Confidence to Scale Helix），这一包含六个步骤的协同设计框架将课堂观察、学生纸质原型制作以及开发人员现场编程相结合，以迭代方式逐步建立产品信心。尽管人工智能承担了约80%的跨设备技术压力测试，但她强调剩余的20%必须始终扎根于人类的切身生活体验。最后，Rowland 呼吁大学研究人员积极将基金资助的创新成果商业化，使这些行之有效的干预措施能够持续存在并惠及有需要的学生。教育科技若仅依凭自上而下的主观假设开发往往难以落地，而初创公司“快速发布再修补”的做法在教育领域极易引发昂贵的客户流失，因为受挫的教师很难再回头使用有缺陷的软件。 (2:24) 堪萨斯大学衍生项目 VOISS 提出“规模化信心螺旋”及六步共创流程（倾听、确认、共创、测试、优化、规模化），为神经多样性学生构建在零真实伤害下反复练习社交技能的心理安全环境。 (4:40) 让软件工程师与机器学习开发者直接走进课堂、与学生面对面观察实际交互，彻底改变了产品可用性问题的发现与实时迭代效率。 (10:46) 团队在测试中采用 80% AI 自动化技术排查与 20% 真实人类体验反馈相结合的策略，并强调高校前沿科研成果必须走向商业化，才能避免因项目资助结束而中断服务。 (14:05)

Reading summary

Amber Rowland, Associate Research Professor at the University of Kansas and Chief Science Officer of VOISS, contends that educational artificial intelligence must be grounded in user-centered design and evidence-based learning science to avoid costly educator churn and succeed in real classrooms. Reconciling Academic Rigor with Startup Velocity (2:24) Commercial technology startups typically prioritize speed to market by deploying minimum viable products (MVPs) and patching errors post-launch. Rowland warns that this strategy is exceptionally damaging in education, where time-constrained teachers have little tolerance for nonfunctional software and rarely adopt tools again after initial failures. To prevent high customer churn, edtech initiatives require a translational role between pedagogical researchers and machine learning engineers, ensuring instructional integrity before products enter classrooms. The Confidence to Scale Helix and Co-Design (4:40) To structure iterative product validation, Rowland introduces the Confidence to Scale Helix—a cyclical framework spanning six stages: Listen, Confirm, Co-Design, Test, Refine, and Scale. Rather than designing around top-down assumptions, developers iterate alongside users through structured think tanks, paper prototyping, and classroom testing. The platform itself grounds extended reality simulations in evidence-based practices such as direct instruction, social narratives, and simulated role-playing. In simulated scenarios—such as navigating peer dynamics during a basketball game—students encounter realistic social reactions without real-world negative fallout. If an interaction goes poorly, users receive immediate, zero-consequence opportunities to retry alternative social strategies. Initial development under a $4.5 million grant progressed through student-informed interface evaluations, including field tests in Hawaii contrasting button-based inputs with conversational AI, and led to a multi-school Year 5 randomized controlled trial. Tactical Classroom Research and Neurodivergent Usability (10:46) Direct user-centered research requires practical methods suited to school environments. Rowland advocates placing software engineers directly into classrooms to witness user friction, collecting student-triggered screen recordings of technical bugs, and replacing open-ended feedback with diagnostic, task-based survey scripts. Co-design exercises adapted from Sesame Workshop, such as the 'Meet the Martian' prompt, encourage students to explain social dynamics without implicit assumptions. Accessibility considerations must also accommodate neurodivergent needs. For students with autism who experience sensory aversion to head-mounted displays, VOISS supports multiple form factors—including iPads and Chromebooks alongside Oculus VR headsets. Rowland also notes that developers must counter classroom skepticism regarding 'tech savior' narratives by remaining transparent about responsible, human-centered AI governance. Balancing AI Automation with Human Experience (14:05) Artificial intelligence accelerates platform development by automating approximately 80% of technical validation tasks. Large language models and autonomous agents conduct multi-device stress tests, identify edge-case bugs, and evaluate baseline accessibility across diverse hardware. However, Rowland stresses that the remaining 20% of evaluation must remain strictly human, as automated systems cannot replicate the emotional nuance, lived experience, and unfiltered feedback of students and educators. Grounding edtech development in published research and randomized controlled trials reduces risk for commercial investors while driving organic adoption by school districts. Rowland cites the real-world impact on students like Oliver—a Denver middle schooler with autism and anxiety who developed conversation-starting skills through twice- or thrice-weekly 20-minute simulation sessions—as proof that user-centered methodologies translate to measurable behavioral growth. Rowland urges academic researchers to actively pursue commercialization rather than allowing evidence-based interventions to expire when grant funding ends. While autonomous AI can streamline back-end testing, sustainable edtech adoption ultimately requires human-centered validation, multi-stakeholder co-design, and transparent pedagogical alignment.

精炼阅读摘要

堪萨斯大学副研究教授兼 VOISS 首席科学官 Amber Rowland 指出，教育科技与人工智能产品必须依托以用户为中心的设计（UCD）和扎实的学习科学，才能避免因脱离课堂现实而导致的用户流失与失败。学术严谨与初创速度的张力 (2:24) 教育科技领域在学术研究标准与商业初创运作之间存在显著的文化差异。Rowland 介绍，学术研究要求穷尽评估所有可能的失败模式与成功因素，以确保工具的教学有效性与可靠性；而初创公司普遍习惯迅速推出最小可行产品（MVP）并依赖后期线上修补。Rowland 强调，在教育行业中，依靠上线后修复是极其昂贵的发展策略，因为教师时间有限且工作压力大，一旦遭遇存在缺陷的软件往往会彻底弃用并造成高昂的客户流失。规模化扩展置信螺旋与共同设计 (4:40) 为了在研发推进中系统性积累产品置信度，团队提出了“规模化扩展置信螺旋”（Confidence to Scale Helix），包含倾听、确认、共同设计、测试、完善与规模化扩展六个循环步骤。通过一项 450 万美元的研究资助，团队召集教育工作者、研究人员与工程师举行智囊会，梳理社交沟通痛点，并将讨论记录反馈给一线教师确认。在设计阶段，团队与学生及教师合作开展纸质原型设计，并在夏威夷的现场研究中对比了按键界面与 AI 对话界面的测试效果。课堂实地验证与反对技术救世主心态 (10:46) Rowland 提出了一系列务实的用户研究策略，包括培训师生在软件发生异常时录屏、让软件工程师与机器学习开发者直接进入课堂观察学生互动、用逐步任务脚本替代模糊的满意度调查，并由学习科学专家审查每次代码迭代。针对部分师生对人工智能的抵触情绪，Rowland 警告开发团队避免“技术救世主心态”，主张透明展示技术的责任机制与以人为本的属性，并借鉴芝麻街工作坊的“会见火星人”等共创练习，排除未经检验的设计假设。平衡自动化测试与不可替代的人类体验 (14:05) 在技术测试与验证流程中，VOISS 利用大语言模型与自主智能体承担约 80% 的常规任务，例如跨 iPad、Chromebook 和 Oculus 头显进行压力测试、无障碍检查与边缘案例发现；这种多设备适配也兼顾了部分自闭症学生对穿戴 VR 设备的感官排斥。然而，Rowland 明确指出剩余 20% 的评估必须由人类完成，因为自动化代理无法复刻儿童与教师真实的情感共鸣与主观体验，这种扎根于真实人类反馈的设计构成了降低投资风险与获取采购信任的核心优势。 Rowland 强调，以丹佛中学生 Oliver 通过模拟训练克服社交焦虑为例，教育科研人员应当积极推动学术成果商业化落地；若仅停留在传统论文发表阶段，具有实际干预价值的技术工具往往会在科研资助期满后中断，无法长期惠及需要行为与社交支持的学生群体。

Detailed notes — zh

## Background 该内容由 Ai4 频道于 2026 年 8 月 14 日发布在 YouTube 平台上，由堪萨斯大学副研究教授兼 VOISS 首席科学官安珀·罗兰（Amber Rowland）主讲。视频简介指出，演讲聚焦于如何将迭代式用户体验研究、协同设计与循证学习科学相结合，通过 AI 社交技能模拟平台 VOISS 的案例，展示打造直观、有效且高留存率 EdTech 产品的实操策略与规模化创新模型。 ## TL;DR 演讲围绕 AI 驱动的青少年社交模拟平台 VOISS，深入阐释了以用户为中心的设计（UCD）如何赋能教育科技产品的研发与商业化。罗兰教授介绍了平衡科研严谨性与创业敏捷度的“可扩展信心螺旋”模型，拆解了从智库研讨、儿童协同设计到随机对照试验（RCT）的七步闭环。她强调 AI 应作为加速测试与开发循环的工具而非替代真实人类体验，并主张通过产学研深度协同将高校研究成果推向市场，以实现对特殊需求青少年的长期真实赋能。 ## Atmosphere & style 整场分享节奏明快、兼具严谨的学术深度与浓厚的人文关怀，主讲人以兼具一线教师、学者与创业者的多重视角娓娓道来，风格务实而亲切。演讲穿插了自闭症学生的真实访谈视频与研发过程中的生动轶事，构建出从直面教育痛点到见证孩子蜕变的情感弧线。整体视听感受专业且充满感染力，让硬核的 AI 与工程方法论始终扎根于关怀具体生命的温度之中。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=zilmb9q5BKM&t=4s) VOISS 是一款基于扩展现实（XR）与 AI 技术的模拟平台，旨在通过安全且支持无限次试错的无风险社交演练，帮助面临社交障碍与焦虑的青少年建立关键社交技能。 - [01:34](https://www.youtube.com/watch?v=zilmb9q5BKM&t=94s) 在教育科技领域，教师与学生的容错率极低；传统初创团队‘先上线 MVP 再修 Bug’的快节奏试错成本极高，容易导致不可逆的用户流失，必须依靠以用户为中心的设计（UCD）来确保可用性。 - [03:53](https://www.youtube.com/watch?v=zilmb9q5BKM&t=233s) 团队提出了‘可扩展信心螺旋’（Confidence to Scale Helix）模型，通过倾听、提炼确认、协同设计、测试、优化与规模化验证的循环迭代，在学术严谨性与敏捷开发之间平衡风险并逐级放大产品信心。 - [05:57](https://www.youtube.com/watch?v=zilmb9q5BKM&t=357s) VOISS 的底层设计深度融合了社交叙事、直接教学法与角色扮演等经过随机对照试验（RCT）验证的循证实践，而非缺乏理论依据的快餐式软件。 - [07:19](https://www.youtube.com/watch?v=zilmb9q5BKM&t=439s) 研发落地中采用七步闭环，包括跨界闭门智库研讨挖掘真痛点、纸质原型与师生协同设计，以及程序员随队在场根据实时反馈快速修正等实战机制。 - [09:34](https://www.youtube.com/watch?v=zilmb9q5BKM&t=574s) 为打破师生与工程团队的沟通壁垒，团队推行全员屏幕录制反馈机制，并组织软件开发与算法工程师亲临教室观察真实使用场景。 - [11:09](https://www.youtube.com/watch?v=zilmb9q5BKM&t=669s) 摒弃宽泛的主观问卷，改用基于任务的定向调研，同时警惕‘科技救世主心态’，主动公开科研理论依据以消除用户对新兴 AI 技术的戒备与抵触情绪。 - [13:13](https://www.youtube.com/watch?v=zilmb9q5BKM&t=793s) 在儿童协同设计中引入故事板共创和‘见见火星人’等趣味技法，引导青少年用具象语言表达日常社交规则与产品期待。 - [13:40](https://www.youtube.com/watch?v=zilmb9q5BKM&t=820s) AI 在研发中的定位是‘加速开发而非取代人类体验’：80% 的常规兼容性与压力测试交由 LLM 和智能体完成，而 20% 核心生命体验与情感交互测试必须保留真实人类参与。 - [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) 产学研协同能将高校深厚的学习科学底蕴转化为企业的商业竞争力与用户留存表现，为 AI 教育产品提供扎实的理论护城河与 P&L 支撑。 - [19:06](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1146s) 高校科研成果如果仅停留在学术层面，往往会随着研究经费耗尽而终止；唯有通过商业化转型将产品推向市场，才能让技术真正惠及更广泛的现实用户群体。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=zilmb9q5BKM&t=4s) 认识 VOISS 与核心用户 Oliver：解决青少年的社交焦虑与教育资源短缺安珀·罗兰（Amber Rowland）是堪萨斯大学（University of Kansas）的副研究教授，长期致力于研发帮助学生学习与教师教学的教育科技产品。同时，她也是扩展现实（XR）模拟平台 VOISS 的首席科学官（CSO）。VOISS 是一款利用 AI 与模拟技术帮助青少年学习并练习社交技能的工具。罗兰开门见山地指出，在短短 20 分钟的分享中，她必须以极快节奏阐述以用户为中心的设计（User-Centered Design, UCD），而展开这一切的起点正是真实的用户故事。她以丹佛市的一名初中生奥利弗（Oliver）为例。奥利弗每天都在与焦虑、社交障碍和沟通困境作斗争；在面对课业学习的同时还要处理人际交往，让本就尴尬敏感的初中阶段变得更加艰难。他和家人对未来充满担忧——他将来如何建立并维系人际关系？又如何找到并保住一份工作？当前教育系统面临严重的教师与专业治疗师短缺，而像奥利弗这样被诊断出焦虑和社交需求的孩子却在激增。VOISS 的诞生正是为了给这类孩子提供可及的解决方案。 - **核心用户痛点**：青少年在初中阶段面临学业与人际交往双重压力，社交焦虑严重影响其长期独立生活与就业前景。 - **行业背景**：全美特殊教育与心理支持师资力量严重匮乏，难以满足日益增长的青少年干预需求。 --- ### [01:34](https://www.youtube.com/watch?v=zilmb9q5BKM&t=94s) 为何 EdTech 必须以用户为中心：学术严谨性与初创公司“快慢博弈” 罗兰谈及自己深耕 UCD 的初衷源于过去的教师经历。作为一线教育工作者，她受够了学区采购的技术产品——那些软件在实际操作中往往无法按预期工作，让人忍不住质疑：“你们在发布前到底有没有让老师试用过？”她直言，无论开发的是 AI 还是普通 EdTech 产品，如果不采用以用户为中心的设计，失败是注定的，因为开发者不可能凭空臆测学生或教师的真实诉求。作为一名研究型教授，罗兰习惯了追求高信度、高效度与严谨性的科研节奏。她自嘲这种方式为“慢”，但新公司的 CEO 则委婉地称之为“彻底（thorough）”。随着团队在过去一年转型为初创企业，她切身感受到了学术思维与创业文化的碰撞：CEO 渴望速度，追求尽快推出最小可行产品（MVP）并投放市场“先用起来再修 Bug”。然而团队很快发现，在教育领域这种做法成本极其高昂。教师群体的容错率极低，一旦初版体验不佳并产生抵触情绪，用户流失（Churn）将无法挽回。因此，在产品推向市场前通过 UCD 确保其真正可用，反而是更便宜、更稳妥的路径。 --- ### [03:53](https://www.youtube.com/watch?v=zilmb9q5BKM&t=233s) “可扩展信心螺旋”模型（Confidence to Scale Helix）在团队分工中，罗兰兼具内容与教学法（Pedagogy）的专业知识，同时充当一线师生与技术工程师、机器学习专家的“翻译官”。为了在严谨与敏捷之间找到平衡，团队提出了“可扩展信心螺旋”（Confidence to Scale Helix）模型。罗兰解释说，这套方法在斯坦福设计思维模型等传统框架的基础上进行了扩展，每一次在循环中迭代推进，团队对产品的信心就会增加一分，所覆盖的用户群体、资金投入和潜在收益也随之层层放大。罗兰分享了一则有趣的命名插曲：起初她将该模型称为“螺旋”（Spiral），但在征求用户反馈时，大家纷纷指出这个词带有负面消极的暗示（如跌入下行漩涡）；加之她身处常年受龙卷风侵袭的堪萨斯州，团队最终决定改用代表向上攀升的“双螺旋”（Helix）。在教育市场中，教师之间、校长之间的口碑推荐是决定商业成败的核心，只有通过层层验证建立起足够的采纳信心，产品才具备真正规模化推广的底气。 ```mermaid flowchart TD accTitle: Confidence to Scale Helix Process subgraph S1["前期洞察与共创"] A["1. 倾听 (Listen) 跨界智库痛点研讨"] --> B["2. 提炼与确认 (Confirm) 证据筛选与需求回溯"] B --> C["3. 协同设计 (Co-design) 纸质原型与师生共创"] end subgraph S2["测试迭代与落地"] C --> D["4. 测试 (Test) AI vs 按钮对比评测"] D --> E["5. 优化 (Refine) 程序员现场调优"] E --> F["6. 规模化验证 (Scale) 多校 RCT 随机对照试验"] end F -. "增强信心并进入下一轮迭代".-> A ``` --- ### [05:57](https://www.youtube.com/watch?v=zilmb9q5BKM&t=357s) VOISS 的科研底座与无痛试错机制 VOISS 的核心竞争力在于其深厚的实证学术支撑，绝非市面上缺乏理论依据的快餐式软件。平台深度融合了社交叙事（Social Narratives）、直接教学法（Direct Instruction）以及角色扮演（Role-playing）等经过多项随机对照试验（RCT）严格验证的循证实践（Evidence-Based Practices）。这些科学理论为模拟场景提供了坚实的底层逻辑。在具体产品机制上，VOISS 为孩子构建了一个完全安全、支持无限次试错的无风险社交演练空间。罗兰举例说明：在虚拟篮球场场景中，如果一个学生冲进球场抢走篮球并跑到另一端，游戏中的同伴就会表现出生气反应。此时系统会给予该学生重新尝试的机会，引导其调整行为以达成理想的社交结果。罗兰风趣地总结道：“在这些社交技能的习得过程中，没有任何真实的孩子受到伤害。” - **团队核心成员**：包括罗兰本人（CSO）、参会的麦吉·莫泽（Maggie Moser）、布鲁斯·弗莱（Bruce Frey）、肖恩（Shawn），以及通过堪萨斯大学孵化器项目引入的商业化 CEO 保罗（Paul）。 - **理论依据**：完全基于循证实践与认知学习科学，并在官网（voiss.org）公开所有文献与研究出版物。 --- ### [07:19](https://www.youtube.com/watch?v=zilmb9q5BKM&t=439s) 践行七步设计法：从 450 万美元智库研讨到夏威夷 AI 对比测试依托获得的 450 万美元项目资助，团队全面推行了七步设计闭环。第一步是**倾听（Listen）**：团队组织了为期一天半的闭门智库研讨，邀请教育工作者、学术研究员、领域专家和工程技术人员共同参与。会议分别设置了“跨界同构桌”与“角色同类桌”（如一线教师专桌），全面挖掘社交技能培养中的真实痛点，并对全部研讨过程进行录音与转录。接下来的步骤环环相扣： 1. **筛选与确认（Filter & Confirm）**：团队对转录文本进行深度过滤，评估各项需求是否符合循证实践、是否具备投资回报率（ROI）与工程可行性，随后将提炼出的核心问题返还给参与者进行二次确认； 2. **协同设计（Co-design）**：保留适合青少年心理的“纸质原型法”（Paper Prototyping），并利用生成式 AI 辅助加速草图绘制，由学生参与角色与情境设计； 3. **测试（Test）与优化（Refine）**：团队在夏威夷等地开展测试，对比学生使用“传统按键选择交互”与“AI 动态驱动交互”的体验差异，程序员随队在场根据实时反馈快速修正； 4. **规模化验证（Scale）**：在第五年的随机对照试验中，平台成功推广至约 10 所学校进行常态化应用并采集长期反馈。 --- ### [09:34](https://www.youtube.com/watch?v=zilmb9q5BKM&t=574s) 倾听学生真实声音与屏幕录制等落地实操策略演讲现场播放了一段简短的真实学生访谈视频。镜头中的奥利弗患有自闭症谱系障碍（Autism），他坦言自己刚转入这所学校时完全不敢和任何人说话，而 VOISS 帮助他找到了与他人交流的方法，并有效缓解了内心的紧张与胆怯。在奥利弗就读的这所专门招收有社交发展需求学生的学校里，孩子们每周使用 VOISS 练习 2 至 3 次，每次约 20 分钟，以此持续建立社交自信。为了打破一线用户与后端工程团队之间的沟通鸿沟，罗兰总结了数项极为关键的操作策略： - **全员培训屏幕录制（Screen Recordings）**：直接培训师生在软件发生异常或逻辑别扭时随手录屏反馈，避免由中间人转述造成信息失真与理解偏差； - **让开发者走出办公室**：安排软件开发与算法工程师亲临教室，现场观察孩子们玩模拟器时的微表情与真实操作，这种直观冲击彻底改变了技术人员的研发视角。 --- ### [11:09](https://www.youtube.com/watch?v=zilmb9q5BKM&t=669s) 现场观察、任务驱动调研与打破“科技救世主”心态在用户调研方法上，团队摒弃了“你觉得这个软件怎么样”等宽泛低效的提问方式，转而采用**基于任务的定向问卷（Task-based Surveys）**。研究人员会给师生下达清晰明确的具体操作指令，并针对特定环节设计精准问题（如“针对这一具体操作，你认为哪些地方需要改进？”）。同时，每一轮产品迭代都必须由教学法与学科内容专家严格审查，防止开发团队在追求炫酷功能时偏离学习科学的核心轨道。罗兰特别警示了当前行业内蔓延的“科技救世主心态”（Tech Savior Mentality）。她指出，当下大学校园与初中校园内对 AI 普遍存在抵触和警惕情绪，许多学生一听到某款工具由 AI 驱动便会断然拒绝使用。为此，团队必须耐心地向用户解释 AI 如何切实改善其体验，并遵循专家帕姆·斯奈德（Pam Snyder）的倡议——“大声宣讲你是如何确保 AI 的负责任应用与以人为本”。 - **调研方式升级**：用步骤明确的任务式测评取代泛化打分，获取具象化的改进点。 - **透明度原则**：主动在官网公开理论依据与论文出处，以学术公信力消除用户对新兴技术的恐惧与抵触。 --- ### [13:13](https://www.youtube.com/watch?v=zilmb9q5BKM&t=793s) 儿童协同设计技法与 AI 加速（非替代）的人本商业价值在儿童协同设计技巧方面，罗兰推荐了芝麻街工作室（Sesame Workshop）的方法库。例如团队常用的**故事板共创（Storyboarding）**，让孩子们分组画出他们心目中的社交情景剧本；或者采用**“见见火星人”（Meet the Martian）**游戏，由工作人员扮演从未到访过地球的外星人，迫使孩子们用最极致、具象且细致的语言来解释日常社交规则。罗兰重点强调了 AI 在开发流程中的定位：**AI 应当加速研发循环，而非取代人类体验**。在 VOISS 的实践中，80% 的常规测试已交由大语言模型（LLM）与智能体系统完成，用于快速排查程序漏洞、执行压力测试、评估无障碍支持，并覆盖 iPad、Chromebook 以及 Oculus VR 头显等多终端兼容性测试（考虑到部分自闭症儿童排斥面部佩戴设备，团队特意保留了平板与电脑端）。然而，剩下的 20% 必须始终保留真实人类的参与。罗兰强调，人与系统交互时所产生的真实“生命体验”（Lived Experience）与情感共鸣是 AI 永远无法模拟的。正是这种扎根于人本的 UCD 流程与严谨的科研背书，赋予了投资人、学校、家长以及企业极大的信任感，构成了教育科技产品从实验室走向商业成功的最强护城河。 ### [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) 产学研协同：高校科研如何赋能教育科技商业化高校科研与企业产品开发的结合不仅能确保学术上的严谨性，更能直接体现在企业的利润与亏损表（P&L）中。高校学者在教育学与学习科学领域深耕多年，正是教育科技产品最理想的合作对象。针对李飞飞（Fei-Fei Li）在早先演讲中提到的“目前与一线教师探讨如何落地融合 AI 的人还远远不够”，演讲者回应道：“来堪萨斯大学（KU）吧，因为这正是我们日复一日在做的事情。” 现场参会的 Lisa Dieker、Maggie Mosher 均来自堪萨斯大学教育学院。然而，单靠一所高校的力量远远不够，整个行业需要更多高校学者参与进来，将前沿研究与产业实践结合，共同运转“规模化信心螺旋”（Confidence to Scale Helix）。 ```mermaid flowchart TD accTitle: 产学研结合与以用户为中心的设计闭环 subgraph Academic["学术科研支撑"] A["学习科学与严谨研究 (Learning Science)"] --> B["联合设计与实证检验 (User Co-Design)"] end subgraph Industry["商业化落地"] C["AI 教育科技产品 (VOISS 平台)"] --> D["损益表现与用户留存 (P&L & Retention)"] end Academic <--> Industry ``` - 演讲者担任客座主编的《特殊教育与科技期刊》（*Journal of Special Education and Technology*）推出专刊，收录了 11 篇围绕以用户为中心的 AI 教育应用论文，涵盖数学、科学、写作、阅读及社交技能等多学科领域。 - 专刊中题为《Fewer Clicks Matter》的文章详细复盘了团队打造首个 AI 聊天机器人的全过程，为企业提供了清晰的用户中心设计（UCD）落地案例。 - 同事 Lisa Dieker 亦在专刊中发表了关于在儿童群体中应用生物识别技术（Biometrics）的研究论文。该专刊已开通线上预先访问（Online First Access），正式刊期定于 9 月出版。 --- ### [17:26](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1046s) 行业资源与学习工具推荐在产品孵化和设计实践中，有诸多优秀的外部资源值得教育科技创业者与开发者参考。全球知名教育科技赛事 Tools Competition 刚刚发布了关于以用户为中心设计的趋势报告（Trends Report），深度拆解了获奖初创企业因何具备投资价值——其核心就在于将用户中心设计与学习科学紧密结合。此外，高校团队与一线实践者也在通过播客等轻量化形式普及 AI 工具与以儿童为中心的设计理念。 - **Tools Competition 趋势报告**：堪萨斯大学团队曾进入该赛事的第三轮评选；报告全面分析了具备资助潜力的产品如何借助学习科学与用户调研赢得青睐。 - **《AI Advocates》播客**：由 Maggie Mosher 与 Lisa Dieker 主持，每期节目仅需 3 至 5 分钟，专为中小学教师打造，每次精讲一款 AI 工具的操作与教学融合方案。 - **Co-Design with Kids**：倡导让儿童深度参与研发的组织（codesignwithkids.org），其播客分享了基于循证研究与儿童协同设计的落地实战方法。 - **规模化信心螺旋评分卡（Scorecard）**：演讲者制作了一份单页评估工具，团队可联系邮箱 `amber.rowland@ku.edu` 获取，用于快速自测产品在各研发阶段的“以用户为中心”成熟度。 --- ### [19:06](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1146s) 商业化的初衷：从科研成果到真实世界的长期赋能演讲在学生 Oliver 的真实转变故事中走向高潮。视频中的 Oliver 谈及自身进步时神采奕奕，教师与家长也一致见证了他使用系统后发生的显著蜕变。这种赋能正是研究团队坚定推进商业化转型的根本动力。如果只是停留在学术研究层面，项目往往会随着科研经费（Grant）的耗尽而归于沉寂，从而失去帮助更多孩子的机会。 > "His parents talk about how much Oliver has changed as a result of the work that we did and the empowerment that we offer through our product. And so, that's why I do what I do. And that's why I commercialize the product because I want more kids to have access to this. And if I didn't, it would have died with the grant." 高校内部沉淀了大量前沿、优质的研究成果，唯有将这些学术创新推向市场、完成商业化转化，才能让技术真正走出实验室，避免极具价值的成果被埋没，持续赋能更广泛的学习者群体。 ## Notable quotes > [01:34](https://www.youtube.com/watch?v=zilmb9q5BKM&t=94s) 你们在发布前到底有没有让老师试用过？ — Amber Rowland > [05:57](https://www.youtube.com/watch?v=zilmb9q5BKM&t=357s) 在这些社交技能的习得过程中，没有任何真实的孩子受到伤害。 — Amber Rowland > [11:09](https://www.youtube.com/watch?v=zilmb9q5BKM&t=669s) 大声宣讲你是如何确保 AI 的负责任应用与以人为本 — Pam Snyder > [13:13](https://www.youtube.com/watch?v=zilmb9q5BKM&t=793s) AI 应当加速研发循环，而非取代人类体验 — Amber Rowland > [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) 目前与一线教师探讨如何落地融合 AI 的人还远远不够 — Fei-Fei Li > [15:28](https://www.youtube.com/watch?v=zilmb9q5BKM&t=928s) 来堪萨斯大学（KU）吧，因为这正是我们日复一日在做的事情。 — Amber Rowland > [19:06](https://www.youtube.com/watch?v=zilmb9q5BKM&t=1146s) His parents talk about how much Oliver has changed as a result of the work that we did and the empowerment that we offer through our product. And so, that's why I do what I do. And that's why I commercialize the product because I want more kids to have access to this. And if I didn't, it would have died with the grant. — Amber Rowland ## Mentioned Amber Rowland, VOISS, 堪萨斯大学 (University of Kansas), Oliver, Confidence to Scale Helix, Maggie Mosher, Bruce Frey, Paul, Pam Snyder, Sesame Workshop, Lisa Dieker, Fei-Fei Li, Journal of Special Education and Technology, Tools Competition, AI Advocates, Co-Design with Kids, Oculus VR, Chromebook, iPad ## Video description Amber Rowland - Associate Research Professor, University of Kansas Effective AI-enabled edtech starts with users. Join us to explore how our team combines iterative user experience research, user co-design, and evidence-based learning science to create AI products that are intuitive, effective, and drive customer retention. We will share practical takeaways through the story of VOISS, our newly commercialized AI-powered simulation platform that helps adolescents build critical social skills. Grounded in academic research and proven success, you’ll see how our design process yields transformative tools and offers attendees a model for scalable, user-driven innovation. Whether you are developing AI tools, conducting research, or shaping learning experiences, this session will give you actionable strategies for building technology that is relevant, effective, and loved by users. [Source](https://www.youtube.com/watch?v=zilmb9q5BKM)

Source quotes

[1:34] Amber Rowland: 你们在发布前到底有没有让老师试用过？ [5:57] Amber Rowland: 在这些社交技能的习得过程中，没有任何真实的孩子受到伤害。 [11:09] Pam Snyder: 大声宣讲你是如何确保 AI 的负责任应用与以人为本 [13:13] Amber Rowland: AI 应当加速研发循环，而非取代人类体验 [15:28] Fei-Fei Li: 目前与一线教师探讨如何落地融合 AI 的人还远远不够 [15:28] Amber Rowland: 来堪萨斯大学（KU）吧，因为这正是我们日复一日在做的事情。 [19:06] Amber Rowland: His parents talk about how much Oliver has changed as a result of the work that we did and the empowerment that we offer through our product. And so, that's why I do what I do. And that's why I commercialize the product because I want more kids to have access to this. And if I didn't, it would have died with the grant.

Source quotes

[0:04] Amber Rowland: I would sit down and try to do something and it wouldn't work the way I think it does. Like, did you even have teachers try this before you put it out? Drive me nuts. [2:24] Amber Rowland: slow [2:24] Paul: thorough [4:40] Amber Rowland: no kids harmed in the making of these social skills. [6:58] Amber Rowland: This is what we think we heard. [10:46] Amber Rowland: game-changer [10:46] Amber Rowland: how did you like it? [10:46] Amber Rowland: tech savior mentality [10:46] Pam Snyder: loud about how you're ensuring it's responsible and human-centric. [10:46] Meet the Martian [15:28] Amber Rowland: Fewer Clicks Matter

Original transcript

[0:04] Hello, my name is Amber Rowland and I'm an associate research professor at the [0:07] University of Kansas where I do research and development on products to help students learn [0:14] and teachers teach. And so, today I'm going to talk to you about a product that I am also the [0:19] CSO of, chief science officer, and that's Voice. And it's an extended reality simulation that helps [0:25] adolescents learn and practice social skills. So, I'm going to talk about the user-centered design [0:30] work that we do. I only have 20 minutes, so I got to talk fast. So, here we go. [0:34] But, I'm going to use Voice as the example so that you're aware of what it is. These are the slides. [0:40] And I have to turn it on. Okay, here we go. I'm going to tell you about Oliver. This is [0:45] a user-centered design session, so I feel like it's important to tell you about my users. So, [0:49] Oliver is an example of one of our users. He's a middle school student in Denver, [0:54] Colorado. Every day he struggles with anxiety and social skills and communication, and that's [0:59] hard when you're trying to learn simultaneously. And middle school's already kind of a hard time. [1:04] Maybe you remember kind of feeling awkward at that time. So, so much so that he's got a lot of [1:09] concerns, he and his family have a lot of concerns about what's his future going to be like. How's he [1:13] going to make and keep relationships? How's he going to make and keep a job? So, he is seeking [1:19] solutions. And unfortunately, right now we have a workforce shortage in education with teachers and [1:24] the therapists who can help him. And we also have a lot more kids who are being identified [1:28] with anxieties and social skill needs. So, we created Voice to try to help kids like Oliver. [1:34] And then, in addition to that, we use user-centered design because [1:38] I got really tired as an educator of using the usually the district-provided technology. [1:44] I would sit down and try to do something and it wouldn't work the way I think it does. Like, [1:47] did you even have teachers try this before you put it out? Drive me nuts. So, we do a really, [1:52] really good job, I think, at user-centered design. I'm I'm to tell you all about it. And I'm going to [1:56] share a lot of practical strategies and techniques for how we do this specifically with kids, [2:00] but also with educators. Um so, anyway, if you don't use in user-centered design, I would argue [2:06] that you're going to fail at whether you're doing AI or any other edtech program, because I think if [2:11] you're trying to make assumptions about what kids or educators need, unless you happen to be one, [2:15] you're not going to make the right assumptions in all areas, and I think user-centered design [2:19] can give you that confidence you need to do it that well. So, just a little more background, [2:24] research professor here. So, I do my job is to do all the things, figure out all the ways something [2:29] can go well, all the ways something can go wrong, and then give examples or give recommendations for [2:36] the way we should proceed when we're developing products and um materials for teachers. [2:41] And so, it's it's a lot of validity, it's a lot of reliability, it's a lot of integrity, [2:46] and that's one of the ways that I keep my job and my my reputation as a researcher is I work [2:51] really hard, and it takes a lot of time. So, I'm thorough. What I call I call slow, but my CEO now, [2:57] he calls it thorough. Uh and so, now we are in a startup. We've been doing it for about a year now. [3:02] And so, I'm learning the startup lingo, and I'm understanding what my CEO wants. He wants [3:06] to move fast, right? It's all about speed. Um he wants to get a a minimum viable product out [3:13] there. He wants it in the wild. He wants people using it, and we'll fix it later. And what we're [3:16] realizing is that's kind of an expensive way to go. And so, I'm I may be preaching to the [3:20] choir here for some of you, but it's that balance that we're always trying to strike. Um and so, [3:26] uh in order to do so, we argue that you've got to use user-centered design. It's the cheaper way. [3:32] Kind of already made this argument. And then also, if you don't, I think you have a churn problem, [3:36] because what happens is it gets out there in the wild, people don't want to use it, and um [3:40] I don't know if you've met educators before, but they're not very forgiving. And they're not going [3:44] to come back, and they get real irritated. And so, we've found that we've really got to make [3:48] sure it's working before we go launch. And the way we do that is the people that we have on our team. [3:53] Um I would argue that at least at the University of Kansas, I bring content and pedagogy know- [3:58] knowledge, and I think that's so critical. I'm going to talk about um the science of learning [4:02] and why it's important to make that a foundational piece for anything ed tech, but I don't have all [4:08] that I need. I'm not a kid who's a middle school student in this current day and age, and I don't [4:13] necessarily have I I was a teacher, but that was like 20 years ago. And so, it's important that [4:18] I'm always bringing those users in. We also tend to go out and find our designers, our developers, [4:23] our engineers. Um and so, I end up serving as the translator between those, right? I've got [4:28] to translate between what a what a classroom kid teacher might need and a and a a designer or a [4:35] machine learning engineer might need. And so, um we have to have a process, and we use something [4:40] called what I'm calling the confidence to scale helix. Uh it's user-centered design with a few [4:46] extra steps. So, if you're looking at the Stanford model, for instance, I've got a couple other [4:50] steps than you might be used to. But, I did that purposefully, and I'm going to tell you about each [4:54] of those as we go. But, um Jim and I helped me a little bit. I was trying to explain that each time [4:59] you go around this circle, I feel like you can build a little bit more confidence. You can build [5:03] a you've got to use a few more people. It usually takes a little more investment. You may be making [5:08] a little bit more profit, but every time you go around that circle, you get a little bit bigger. [5:12] And at first, I called this the the um what do I call it? I called it the spiral. And of course, [5:17] I'm user-centered design, so I'm asking people, "What do you think about my confidence to scale [5:20] spiral?" And they're like, "That has negative connotations. Don't call it that." And then I'm [5:24] realizing I'm from Kansas, so we have tornadoes. I definitely don't want this to be that connotation, [5:29] right? So, we went with helix. So, this idea that you're always getting a little bit bigger, [5:33] but you're also getting a little bit taller. Uh so, the reason we do this, the confidence to scale [5:38] helix, each time you go around, you're getting a little more confident that what you're creating [5:41] is something people are going to adopt, and then they're going to tell their friends about. And [5:44] that's what's really important education, because teachers talk to teachers, principals talk to [5:48] principals. So, and whatever area you happen to be in, I think it's probably the same thing. When [5:53] users start recommending it to others, then we've got a really great product that's going to sell. [5:57] So, I mentioned voice. Uh it is grounded in very rigorous academic research. You can imagine with [6:04] what I've said so far, but we have um social narratives, which are evidence-based practice [6:08] for teaching social skills. We have direct instruction. We have role-playing. These are [6:12] all evidence-based practice. Lots of evidence. Lots of randomized control trials that show that [6:17] it works. So, we put it into our simulation to make sure that we can be confident that that [6:22] grounding is solid. We also built it with kids and with teachers. I'm going to tell you more about [6:27] how we did that. Uh but it is commercialized. It is active. And just to give you an example, [6:32] you'll see a a very brief video of what it looks like here in just a minute. But if you go into [6:36] our simulation, a kid wants to play basketball, there's kids playing basketball. If you go in, [6:40] you steal the ball and you run to the other side of the court, you're going to make all those [6:43] people mad. And so, you then have the opportunity to do it again. And you want to get the outcome [6:48] that you're interested in. And so, we give kids a safe place to learn and practice social skills [6:53] as many times they need to. No kids harmed in the making of these social skills. [6:58] This is our team. Um we have founding members. Maggie Moser's at this conference. You may have [7:02] seen her. Uh Bruce Fryer, Shawn, and then Paul is our CEO. Again, we don't have I don't have a [7:08] PhD in in marketing or commercialization, but we were able to find a CEO who could help us do that [7:14] through our accelerator program at the University of Kansas. It was awesome. So, here's our here's [7:19] our seven steps. The first one is listen. And this one is so critical to us because we don't develop [7:24] anything unless we know it's a pain point. Like we don't make assumptions. I got some ideas of what I [7:29] think could be developed next, but I always ask. And so, we bring um educators and researchers and [7:36] experts and engineers all together. We did a think tank when we first got this round of funding, [7:41] which is $4.5 million. Um we brought them together for a full day and a half to talk about [7:46] the problems that are in social skills. We knew that we knew we wanted to use extended learning. [7:51] And so we put like-minded people at the same tables and had them talk and brainstorm. And then [7:55] we had role-alike people, all the teachers in the room, get around, talk about things. And the idea [8:00] here is that we are trying to get as many voices as we can from the beginning. We then record all [8:06] of that. So each of us are in those tables and we're recording that information. And we take it [8:10] we take the transcripts and we filter it down. And we pull out everything that we think is going to [8:15] work. Is it evidence-based practice? Is it a good return on investment? What we call feasibility. [8:20] All those things. Next we confirm it with them. We send that back. We say, "Okay, we filtered [8:24] it out. Here's what's bubbling to the top. You agree? This is what we think we heard." And then [8:28] they give us some confirmation on that. And then the co-design. So I was talking to my um machine [8:33] learning guy about this talk and I said, you know, paper prototyping is something we do a lot cuz we [8:37] work with kids. And I think humans still need that paper. And he totally agreed with that. But we can [8:43] have AI make those paper prototypes, right? We can help us um make things go faster. So I still have [8:51] students helping with the character and scenario development. We still have educators helping [8:55] develop some of the scenarios and the materials. I'm going to get to AI here in the next slide. [9:00] Um test. We go into testing it. So the first scenario we had middle school kids test it, tell [9:05] us what they thought. And then we went to Hawaii and we had kids test the difference between our [9:10] voice that has buttons to push and voice that has AI that runs it. And we gave them the opportunity [9:14] to give us feedback. And we learned so much. And those programmers were with us. And then they went [9:19] in and they refined it. And they said, "Do we do we hit the mark?" And then finally that scale. [9:24] So we ended up having about 10 different schools ultimately using it in our year five RCT. [9:29] And they were able to give us lots of feedback after that as well. Um so I wanted to just let [9:34] you hear from the kids. I'm going to give you only like less than 30 seconds. We're [9:38] not going to listen to the whole video, but if you want to hear from a bunch of kids, you can. [9:44] That would be really care about them. And and I I think Here's Oliver. that it would be really um [9:50] helpful if people used it more and it could um [music] help people [9:55] um with their emotions in a variety of things. When I'm uh talking to people, [10:00] like when I first uh [music] came to this uh school this year, I didn't really talk to anybody. [10:07] Yeah, and uh Voice like um [music] helped me start talking to people more because it like um [10:13] like hel- helped me um learn more ways [music] to talk to people like and not be nervous. [10:22] Be Okay, there's lots of kiddos that are talking in that, but Oliver has autism and he struggles, [10:27] as I mentioned at the beginning, with social skills. And Voice is one of those that he really [10:31] really enjoyed. In fact, most of the kids at this particular school are there because they have a [10:35] social skill need, and so they love Voice. They play it two to three times a week, 20 minutes [10:40] a day, uh just to get that practice and build that confidence that they need. So, here's those [10:46] practical things I was telling you about. Um as I said, I mentioned that I serve as the translator, [10:52] and one of the things we quickly figured out was the screen recordings were critical. So, [10:55] we trained the students and the teachers to take screen recordings. Anything Anytime something went [10:59] wonky or weird, we asked them to record it and then send it to us because otherwise I'm trying [11:03] to serve as the middle person explaining what the kids or the teachers were telling me. So, screen [11:07] recordings are critical. The other one was having developers in the room actually watching the kids. [11:12] And I I developed a couple of products without doing that, and then the the moment we brought the [11:16] developers into the room, it was game-changer for everybody. So, I I highly recommend getting out of [11:20] the office and going and watching the kids play it live. Um then we do task-based surveys. So, [11:26] instead of just sending a product to the kids or the teachers to say, "What do you think?" [11:29] you know, really broad, we give them exact steps that we want them to follow and then we ask them [11:33] very specific questions. You know, what did you think about this? How would you make this better? [11:37] Um, what are some things that we could have improved in this exact area? And that way [11:41] we're a little bit more specific about what we're getting. We also always have experts in pedagogy [11:45] and content reviewing each iteration because we want to make sure that whatever the developers [11:50] are creating hasn't strayed too far from the learning science or the evidence-based practice [11:55] that we know. Cuz again, I can make a great claim right now that Voices research-based. There aren't [11:59] a ton of edtech products out there that can make that claim. And I think that's a critical piece. [12:03] So we're constantly coming back to make sure that we're doing that. Um, of course, designing based [12:08] on what we know about learning. I think that gets mixed missed sometimes in the um, in the pursuit [12:15] of money. And I think we always are constantly bringing that back. So we cite our learning [12:19] science on our website. You go to voices.org you're going to read about the research. You're [12:23] going to see the publications that we have. And then we also encourage that we or we we try really [12:28] hard to be aware of the tech savior mentality. Um, I don't know how many colleges campuses you guys [12:33] have been on or middle schools you've been to, but right now there's a lot of resistance to AI. [12:38] So people who are creating products for these people have to remember that if it says AI in it, [12:43] they're much less likely to even try to use it, let alone borrow it. I have kids tell me all the [12:48] time, "Oh, I'm not going to use that. That's AI." Okay. The reason that we chose AI is because we [12:53] know it's going to make your life better. And I have to explain that. And I think we need to [12:57] be very explicit. So Pam Snyder was tell us and I came to her session in here yesterday [13:02] and she said, "Be loud about how you're ensuring it's responsible and human-centric." So we are. [13:08] We are we try to be very loud about how we've done it and I recommend the same. All right. [13:13] Co-design for kids. Great website if you work with kids or edtech at all. Sesame Workshop, they have [13:19] a whole bunch of techniques that they use to get at this co-design aspect. Um, one of them that I [13:25] love um, that's not I don't know if it's on the list, but we do storyboarding. So, we actually [13:29] have the kids storyboard a whole scenario. You know, what if we don't have what you need, [13:32] what do you think it should be? And they work in small groups to storyboard it. They also have [13:36] like a meet the Martian where they actually have somebody dress up as a Martian and come in and [13:40] the kids have to explain it like this person has never been to Earth before. So, they have to give [13:44] really, really specific and detailed. Just some of those techniques I think can be very helpful [13:49] even if you're not working with students, I think it could be helpful. So, the other thing is expert [13:53] and user-centered content or generated content. We do a lot of that. So, teachers and students help [13:58] build um our products cuz again, kids are experts at being kids now and as are educators. All right, [14:05] where AI speeds up the cycle, it doesn't replace it. And so, I heard several people throughout [14:09] this conference talking about that. And I think engendering trust is a critical piece and so, [14:13] we do that through the transparency and the pedagogy that we use as our foundation. But [14:18] we also talk about the rapid testing that we do. So, we use large language models, agents, whole [14:23] system to actually test ours to make sure that we don't have errors, that's accessible. You know, [14:28] we're getting at how do you break it and then recommend mitigation, all those things. So, that's [14:32] kind of where we use AI especially in that testing phase to do quick testing on multiple devices cuz [14:38] you can download it on an iPad, a Chromebook or an Oculus headset. We have a lot of kids with autism [14:43] who don't want something on their face. So, we make it accessible on an iPad or a Chromebook. [14:48] Um then on the 20% side, I think you've got to have humans testing it because [14:54] as was mentioned in the keynote this morning, that lived experience can't be replicated by AI. [14:58] Just that emotional resonance that happens when someone gets in front of something and they can [15:02] articulate that to you, AI can't do that. And so, for me, that's a critical piece to keep the humans [15:08] as a part of the design process. So, the business case for human user-centered design, [15:14] I think founders and investors are more um comfortable with it and they're going to have [15:18] more confidence. Remember that confidence is Scale Helix. I think schools, parents, and enterprises [15:23] are way more comfortable when you say the word research-based. It just It's a great way in. [15:28] And then um it's really good for the research rigor, and it's really good for the profit and [15:32] loss statement as well. So, I highly recommend um working with higher ed faculty who have this as [15:39] their specialty because we do this. Um I heard I think it was Feifei who's And I love her, by [15:44] the way. But she mentioned like there's not enough people talking to teachers about how to integrate [15:49] AI, and I was like, "Come to KU because that's what we do." Um Lisa Dieker's here speaking. [15:54] Maggie Moser's here. We are all faculty at the University of Kansas in the School of [15:59] of Education. So, we are doing that, but we you know, we're only one. [laughter] And so, [16:04] we need more people out there doing that. So, I recommend finding higher ed folk who can do that [16:09] and working with them. This is just a summary of kind of the thought the top five ways that [16:13] I think you can spin up your own confidence to scale helix. Um I've mentioned all these already, [16:19] but that is there as a quick summary. And then I want to share um The Journal of Special Education [16:25] and Technology is getting ready to come out with a special edition that I'm the lead guest editor on, [16:30] and I have 11 different publications on this exact topic um in different areas. So, [16:35] no matter what area of ed tech you might be in, whether it's math, science, writing, reading, um [16:41] or in our case social skills, um the first article that's These are all out there and available at [16:46] that um QR code, but Fewer Clicks Matter, that's uh the story of how we created our first chatbot, [16:52] which I know is a little bit old, and that's the thing about research is again, we move a slower, [16:57] especially if we're doing peer-review journals. But it's giving you basically this whole talk [17:01] in a in a publication. So, if you just want to see the user-centered design that [17:05] we went through and the case user case that we used to create our first chatbot, you can. [17:10] Uh but then there's all kinds of other really good articles. I don't know if you saw Lisa Dieker, but [17:14] she was talking about biometric use with kids. Uh she published that article in here, too. So, the [17:19] official special issue should come out September, but you can get online first access at that URL. [17:26] A few other things I mentioned before I leave you. The Tools Competition is a group that we've [17:32] applied for and we got into the third round, I think, and didn't end up getting funded this last [17:37] round, but we're going to try again. But, they did come out, I think it was last week, with a really [17:42] good publication on user-centered design. And they talk about the fun the products that were funded, [17:48] the startups that were funded, and they summarize really well what it made them fundable. And so, [17:53] they're talking about user-centered design. They're talking about the learning science [17:57] and why Tools Competition was willing to fund them. So, highly recommend the Trends Report. [18:02] AI Advocates is a podcast that Maggie Moser and Alisa Dickerson have that helps [18:07] teachers integrate artificial intelligence. So, I felt like, based on this keynote this morning, [18:11] I'd tell you about them cuz it's 3 to 5 minute um podcast that walks you through a different tool [18:18] for each of their podcast. And then, the last one is the Co-design with Kids folks. They did a [18:22] podcast last week just on this topic. And so, they share about their co-design with kids.org, but [18:28] they also talk about how else they're implementing research-based and user-centered design. So, the [18:34] Confidence to Scale Helix, I created a scorecard that's one pager that your teams are welcome to [18:39] use. If you want access to it, just email me at amber.rollins@ku.edu and I'll send it to you. And [18:45] it basically lets your team a very quickly score how user-centered they are in each of these um [18:52] in each of these phases of the cycle. And then, you get yourself a score and see where you need [18:56] to go. So, you're welcome to email me, reach out if you have any questions, concerns, thoughts. If [19:01] you want me to recommend somebody for the topic that you happen to be working on, let me know. [19:06] And then, I'll leave you with Oliver. You know, I In that video, you noticed how happy he was to be [19:12] talking about it. And the teachers talk about how much Oliver has changed. His parents talk [19:17] about how much Oliver has changed as a result of the work that we did and the empowerment that we [19:21] offer through our product. And so, that's why I do what I do. And that's why I commercialize [19:26] the product because I want more kids to have access to this. And if I didn't, [19:30] it would have died with the grant. And so, you're going to I hope you're going to start to see a lot [19:34] more higher ed faculty coming up and and actually commercializing their products because it's a loss [19:41] if they don't. We have some really cool things happening. So, those are the slides. If you have [19:45] any questions, don't hesitate to reach out for me. And that is all I have for you today. [applause]