"HCI, AI, and Data-Driven Strategies in Screen Time Reduction and AI-Assisted Teaching Web Apps"
by Jai Joshi, Joe Fang, and Isaac Abell from the Intelligent Adaptive Interventions Lab, University of Toronto.
Speaking Point: “Hello everyone, and thank you for stopping by our poster. My name is [Your Name], and I’m here to introduce two exciting projects from our lab that blend Human-Computer Interaction (HCI), Artificial Intelligence (AI), and data-driven strategies to address two major challenges: personalized learning and screen time reduction.”
Key Details to Share:
Speaking Point: “First, I’ll highlight JoltEdMod, which explores how generative AI can help teachers and students create custom worksheets and lessons that cater to individual needs.”
Key Details to Share:
Why It Matters:
Speaking Point: “Next, I’ll discuss Focus Flow, an app designed to reduce impulsive screen time and digital distractions through real-time interventions and adaptive prompts.”
Key Details to Share:
Why It Matters:
Speaking Point: “Though focused on different challenges—personalized education vs. screen-time reduction—these projects share core principles of adaptive AI and a user-centric design approach.”
Key Details to Share:
Speaking Point: “We anticipate that both adaptive prompts in screen time apps and AI-generated, personalized teaching materials can improve outcomes—reducing user habituation to notifications and boosting student learning gains.”
Key Predictions:
Speaking Point: “We’d like to thank our mentors and collaborators whose support made these projects possible.”
Key Acknowledgements:
Below are a few references cited in the primary poster text (full list available on the poster):
References (See poster text for the complete references list.)
Speaking Point: “In conclusion, we believe these two projects—Focus Flow and JoltEdMod—demonstrate the power of adaptive AI systems and HCI principles in solving real-world problems, from helping students learn more efficiently to guiding us toward healthier digital habits.”
Final Takeaway:
Invite Questions and Discussion:
Opening Hook (0:00–0:45):
"Good [morning/afternoon], everyone. Let’s begin with a quick thought experiment. Imagine two scenarios:
These aren’t isolated issues. They’re symptoms of a larger problem: systems that don’t adapt to human needs. Today, we’re sharing research on how AI and Human-Computer Interaction (HCI) can bridge this gap. Our work tackles digital distraction and educational equity, two challenges where rigid, one-size-fits-all approaches fall short."
Background & Problem Statement (0:45–2:15):
"Let’s unpack these problems. First, digital distraction. Studies show the average person unlocks their phone 150 times daily [Duke & Montag, 2017], often impulsively. This isn’t just annoying—it fragments attention, reduces productivity, and correlates with anxiety [Keles et al., 2019]. Lockout apps like One Sec add friction, but they’re static. They don’t address why we reach for our phones or adapt to individual habits.
Now, education. Traditional worksheets assume all students learn the same way. But learning is deeply personal. A student struggling with loops in Python might need visuals, while another thrives on examples. Yet teachers, already stretched thin, lack tools to customize materials at scale [Celik, 2023]. Even ChatGPT-generated content risks irrelevance without pedagogical oversight.
At their core, both problems stem from a mismatch: humans are dynamic, but our tools are static. That’s where our research comes in."
Theoretical Foundations (2:15–3:30):
"We ground our work in two fields:
Human-Computer Interaction (HCI):
HCI teaches us that technology should work with human behavior, not against it. Take the Fogg Behavior Model [Fogg, 2009]. It states that behavior change requires three ingredients: motivation, ability, and a trigger. For example, reducing screen time isn’t just about blocking apps—it’s about designing triggers that align with a user’s goals. If someone opens Instagram when stressed, an effective intervention might offer a breathing exercise instead of just saying ‘Don’t do that.’
AI & Adaptive Systems:
AI isn’t just automation—it’s adaptation. In education, this aligns with Vygotsky’s Zone of Proximal Development [Vygotsky, 1978], the idea that students learn best when challenged just beyond their current ability. AI can personalize this ‘zone’ dynamically. For instance, if a student aces variables but struggles with loops, AI generates loop-focused exercises while adjusting difficulty based on real-time performance.
Together, these frameworks let us build systems that learn from users and adapt to their needs—not the other way around."
Research Questions (3:30–4:15):
"Our team asked two driving questions:
For Digital Habits:
Can AI-driven, real-time interventions reduce impulsive screen use more effectively than static reminders? And critically—can they sustain behavior change long-term?
For Education:
Can generative AI produce educational materials that are both personalized and pedagogically sound? How do these compare to human- or ChatGPT-generated content in engagement and comprehension?
These questions aren’t just technical—they’re deeply human. They’re about designing technology that respects how we learn, work, and live."
Significance & Transition (4:15–4:45):
"Why does this matter?
Now, let’s see these ideas in action. [Collaborator 1] will walk you through Focus Flow, our HCI-driven approach to mindful screen use, and [Collaborator 2] will demo JoltEdMod, an AI tool that personalizes programming education. Feel free to ask questions as we go!"
Summary (0:45–1:30):
"Let’s tie this together. Both projects show how merging AI with HCI can create systems that adapt to humans—not the reverse.
These aren’t just apps. They’re proof that technology can be human-first—responsive to our habits, needs, and potential."
Final Takeaway (1:30–2:00):
"As AI grows more powerful, the question isn’t what can it do? but how can it serve us better? Whether it’s helping students learn or helping us reclaim focus, the future lies in systems that learn, adapt, and respect human complexity. Thank you, and we’d love to hear your thoughts!"
Citations to Highlight:
Practice Flow:
Q&A Prep:
Q: Why combine HCI and AI in these projects?
A: HCI ensures solutions are human-centered, addressing why people behave or learn the way they do. AI adds adaptability, letting systems personalize interventions or content dynamically.
Q: How does the Fogg Behavior Model apply to Focus Flow?
A: Fogg’s model emphasizes motivation, ability, and triggers for behavior change. Focus Flow uses AI to tailor triggers (e.g., prompts for breathing exercises) to a user’s context (e.g., stress), making interventions more actionable than generic app blockers.
Q: What’s Vygotsky’s Zone of Proximal Development (ZPD), and how does JoltEdMod use it?
A: ZPD is the gap between what a learner can do alone vs. with guidance. JoltEdMod uses AI to identify this zone for each student—for example, generating loop exercises that are challenging but achievable based on their prior performance.
Q: How do you address ethical concerns with AI in education (e.g., bias)?
A: We involve teachers to review AI-generated content for accuracy and inclusivity. Additionally, JoltEdMod’s design includes transparency features, like explaining why a lesson was recommended, to mitigate “black box” risks.
Q: Why focus on screen time and education? Are these connected?
A: Both are about adaptation. Screen time tools need to adapt to individual habits, while education must adapt to learning styles. By tackling both, we explore how AI-HCI systems can address diverse human needs.
Q: Why compare JoltEdMod to ChatGPT instead of other AI tools?
A: ChatGPT is widely accessible but not designed for pedagogy. By benchmarking against it, we highlight the value of AI tailored for education—balancing personalization with structured learning outcomes.
Q: How did you choose metrics like “app reopen rates” or “quiz scores”?
A: These are validated proxies for deeper outcomes: reopen rates reflect impulse control, while quiz scores measure comprehension. We paired them with qualitative feedback to capture both behavior and perception.
Q: Why use Thompson Sampling in Focus Flow?
A: Thompson Sampling balances exploration (trying new prompts) and exploitation (using known effective ones). Unlike static systems, it adapts to user responses, which is critical for sustaining engagement over time.
Q: How do you ensure participants in your experiments represent diverse backgrounds?
A: We recruited students and teachers from varied socioeconomic and educational settings. For Focus Flow, we included both casual and heavy smartphone users to test robustness.
Q: What’s the biggest limitation of your methodology?
A: Short-term experiments may not capture long-term behavior change. For example, Focus Flow’s 4-week trial shows promise, but habits often regress over months. We’re planning longitudinal follow-ups.
Q: Could these tools worsen digital inequality if only some schools can access them?
A: Equity is central to our design. JoltEdMod is open-source, and Focus Flow’s code will be publicly available. We’re also partnering with NGOs to deploy these tools in underserved communities.
Q: How do you balance personalization with privacy in data-driven systems?
A: We anonymize user data and use on-device processing where possible (e.g., Focus Flow’s prompt logic runs locally). For JoltEdMod, schools control data storage to comply with FERPA/GDPR.
Q: Are you concerned about over-reliance on AI in education?
A: Absolutely. JoltEdMod isn’t a replacement for teachers—it’s a tool to augment their work. The AI handles content generation, freeing educators to focus on mentorship and critical thinking.
Q: How might your findings apply to workplaces, not just schools?
A: The principles are universal. For example, Focus Flow’s nudges could help reduce email-checking habits in offices, while JoltEdMod’s approach could train employees in technical skills.
Q: What’s the commercial potential of these projects?
A: While our focus is research, JoltEdMod could license its engine to EdTech platforms, and Focus Flow could adopt a freemium model. However, our priority is ethical deployment, not profit.
Q: What’s next for this research?
A: We’re expanding JoltEdMod to non-programming subjects (e.g., math) and testing Focus Flow on iOS. We’re also exploring LLM fine-tuning to reduce hallucinations in educational content.
Q: How did your team collaborate across HCI and AI specialties?
A: We held weekly cross-disciplinary workshops. HCI experts led user studies, while AI teams optimized algorithms. Regular feedback loops ensured alignment with human-centered goals.
Q: Would you partner with app companies (e.g., Instagram) to integrate Focus Flow?
A: Potentially, but cautiously. We’d need guarantees that integrations wouldn’t compromise user privacy or autonomy. Open-source partnerships are safer first steps.
Q: How can educators get involved in testing JoltEdMod?
A: We’re launching a beta program this fall—interested teachers can join via our lab’s website. Participants will co-design features and provide feedback on AI-generated content.
Q: What surprised you most during this research?
A: How much context matters. For example, Focus Flow’s breathing exercises worked best when prompted during stress peaks, not randomly. Personalization isn’t just about the user—it’s about timing.
Q: Isn’t this just another app that people will ignore after a week?
A: Unlike rigid tools, Focus Flow adapts. If a prompt isn’t working, the AI shifts strategies. Early data shows 40% of users sustained reduced screen time after 4 weeks—a strong sign.
Q: Can AI truly understand human learning better than teachers?
A: No—and it shouldn’t. JoltEdMod’s role is to handle repetitive tasks (e.g., worksheet variations) so teachers can focus on what humans do best: inspiring curiosity and critical thinking.
Q: How is Focus Flow different from Apple’s Screen Time?
A: Screen Time is retrospective (“You spent 3 hours here”), while Focus Flow is proactive. It intervenes in the moment with alternatives, addressing the root cause of distraction, not just the symptom.
Q: What if AI-generated worksheets contain errors?
A: JoltEdMod includes a teacher review step before materials are assigned. We also use constrained AI models that cross-check facts against vetted educational databases.
Q: Why not use existing AI like GPT-4 instead of building JoltEdMod?
A: Generic LLMs lack pedagogical guardrails. JoltEdMod fine-tunes models on curricula and student performance data, ensuring outputs align with learning objectives—not just statistical patterns.