A correct answer does not tell the whole story
A learner may find a worked solution clear yet struggle to begin a similar task. Another may submit an unfinished attempt while understanding why they chose a particular step and where they need help. Asking only whether a correct answer appeared on screen misses the difference between these experiences.
Answers are useful, and direct explanation may be exactly what a learner needs. A learning experience should also ask what changed: can the learner choose a step, explain its purpose, and recognise when an idea applies? These are design questions we take seriously, rather than outcomes we claim Warda has already demonstrated.
What does an experiment with AI assistance tell us?
A PNAS study compared general-purpose generative assistance with an educationally constrained tutor in high-school mathematics. Performance improved while tools were available, but assistance did not automatically produce better independent performance. The general-assistance group performed worse after tool access was removed. Guided assistance mitigated that harm without establishing a statistically significant independent advantage over the control group. Read the original study.
This study did not evaluate Warda. It invites us to examine independent work after assistance, within the context of each trial.
A hint that returns the problem to the learner
A useful hint must address the point where the learner is stuck. If the task itself is unclear, adding solution steps can deepen confusion. If the idea is understood but its application is uncertain, a small reminder may be more useful than repeating the entire lesson.
A conversation might begin with a light question about what the learner tried or which part is unclear. This must not become another examination before help is available. A route to explanation should remain open, including for someone who cannot begin. The aim is to make room for an attempt without requiring the learner to earn assistance through extra answers.
How does this connect to Warda?
Warda’s design starts from curriculum context: explanations connect to the learner’s material, and practice connects to questions with clear sources and criteria. Progressive hints help locate a difficulty without immediately revealing a complete solution. Practice and exam simulation have different assistance boundaries because they serve different educational purposes.
Our approach connects an attempt that reveals a need, assistance suited to that need, and an opportunity to work independently again. Direct explanation belongs in this vision; its timing and scope should serve the task. A convincing explanation opens a route to understanding, while independent capability requires a chance to apply the idea.
What makes assistance useful?
We interpret assistance in relation to the task: did it clarify the goal, leave room for an attempt and help the learner use the idea without the same support? A guiding question can become a burden; a hint can be vague or arrive too late. Adding conversation to a solution does not automatically make it suitable teaching.
Warda’s approach aims to support confident thinking with explanations learners can revisit. This value is not reduced to conversation length or answer speed: it concerns clarity and opportunities to understand and try an idea. Measuring impact requires evidence appropriate to that question.
Sources and context
- Bastani et al. (2025), Generative AI without guardrails can harm learning
A high-school mathematics experiment in Türkiye; not an evaluation of Warda.
Sources describe research, specifications or documented product behaviour, as identified above. They did not evaluate Warda or establish its effectiveness.