AI-Driven Learning
At Enki, we wanted to explore conversational learning as a novel approach to teaching coding, with the aim of enhancing comprehension and engagement. This project investigates how leveraging GPT 4.0 can create more interactive, personalized learning experiences that better resonate with students.
Role: UX/UI Designer, Researcher
Tools: Figma, Illustrator, Photoshop
Timeframe: 1 week
Client: Enki
01 Research
-
The Map Is Not The Territory
-
What Are Chat Interfaces Good For?
-
What’s Wrong With AI Chat Interfaces?
02 Solution
-
How Do We Reduce The Learning Curve?
-
Prompt-AI Alignment & Designing For Contextual Awareness
-
Make Me Think
-
User flows
03 UI Design
-
Designing The MVP
04 Learnings & Conclusion
-
My Learnings
-
Conclusion
01 Research
The map is not the territory
Maps are visual representations of places, but they don’t fully capture the actual reality of being there, just as a menu is a description of food rather than the actual experience of the dish itself. It’s easy when researching to be disconnected from the reality learners face, so I simulated the learning experience firsthand using GPT 4.0 to move quickly.
What are chat interfaces good for?
-
Adaptive Learning: Can tailor the learning experience to your individual needs and preferences, adapting the pace, content, and style of instruction based on your learning style, prior knowledge, and progress, providing a personalized learning journey in real-time.
-
Interactivity: For the first time, you can interact with the textbook, and it can respond to your questions.
-
Efficiency: Streamlines the learning process by identifying areas where you need more practice or additional resources. It can prioritize content based on relevance and importance, optimizing your learning time and efforts.
-
Feedback: AI can provide instant feedback on your performance, highlighting strengths and areas for improvement. This real-time feedback mechanism enables you to track your progress more effectively and take corrective actions as needed.
What’s Wrong With AI Chat Interfaces?
-
Learning Curve: Chat interfaces lack affordances. When you first open ChatGPT, you come to a crossroad, a blank screen. How do you know how to use it? What can it do? What can't it do? How should you phrase prompts? The only clue you get is to type into the textbox, leaving the onus completely on the user to figure it out.
-
Prompt-AI Alignment: They don’t given enough feedback. Picture this, you laboriously write the perfect prompt, and GPT responds confidently incorrectly. Where did you go wrong? What context was missing? Again, users bear the burden of discerning this. So, how do we make it easier for learners to provide all the necessary context and understand how the AI has interpreted the prompt?
-
Make me think: To excel at something, you have to read, question, seek answers, and practice deliberately. But why bother writing when AI can do it for us? Why bother thinking when AI can generate code? The challenge lies in empowering individuals to cultivate critical thinking skills and retain knowledge amidst the convenience of AI in an engaging way.
02 Solution
How do we reduce the learning curve?
By embracing an adaptive AI-guided onboarding experience, (1) we can accelerate product familiarization while leveraging the flexible nature of chat interfaces to shape interactive dialogues, helping us (2) rapidly iterate towards 'aha' moments tailored to each persona. All the while, (3) reusing existing components to save development time.
Prompt-AI alignment and designing for contextual awareness
In the initial phase, leverage the AI to lead users through the subject matter, mirroring the approach of a teacher, step-by-step. Structure prompts with curated choices to ensure responses are contextually accurate and minimise writing fatigue.
Make me think
As users advance, they need tools that help them gain clarity in their thinking. Highlighting, real-time analysis, and offering feedback and suggestions on prompts will help refine their prompt engineering skills.
User flows
To better envision the flow, I outlined a high-level concept of the user interaction through sketches before creating an onboarding user flow diagram to visualize the entire sequence of steps holistically.
03 UI Design
Designing the MVP
To work efficiently, I used Figma to iterate quickly and explore various layout options, visual styles, and interaction patterns. This approach helped to identify and refine the most functional and user-friendly solutions, resulting in a final set of designs that struck an optimal balance between aesthetics, usability, and technical feasibility for development.
04 Learnings & Conclusion
My learnings
-
Adhoc Design: We need employ different design approaches that consider factors such as tone, personality, and conversational flow to create a seamless and natural interaction than just traditional GUIs.
-
Ethical Considerations: AI, if not designed ethically, risks perpetuating biases and manipulating user behaviour. Just as infinite scrolling led to decreased attention spans, AI’s could make users overly reliant, exploiting users' attention and decision-making. This underscores the critical need for ethical design in AI systems.
-
Cognitive Load & Feedback: The constant effort required to rephrase, clarify, and reconstruct prompts can quickly lead to fatigue and frustration. So it'll be crucial to provide clear and helpful feedback when users encounter misalignment.
Conclusion
Learning is like playing catch with the author. The author "throws" ideas that the reader must actively "catch" and comprehend. Through analysis, questioning, and interpretation, the reader then "throws back" their understanding. This back-and-forth is essential for deep comprehension and critical thinking. For the first time, AI can enable the learner to engage in this catch.
This means that we can now dynamically change the experience of the course. This has huge implications on learner outcomes. It allows even small teams to focus less on creating course content and more on being orchestrators, guided by empathy and creativity toward answering more meta questions:
-
How do we train critical thinking skills programmatically?
-
How do we improve knowledge acquisition?
-
How do we create more engaging learning experiences?
However, it’s important to understand that language models are not a panacea. The hallucination problem still needs to be tackled, AI is topically constrained, and the accuracy and quality of the models' knowledge determine the quality of interactions. These are the next set of challenges I aim to learn next.