AlwaysAI

Forcing Correct AI Collaboration Through Assessment and Learning Design

A unified platform for AI-augmented assessment and multi-session learning activities implementing the O×A Framework

What Is AlwaysAI?

AlwaysAI is an AI-augmented assessment and learning platform that treats AI collaboration as a core educational competency. Instead of banning, ignoring, or detecting AI use, AlwaysAI makes AI collaboration mandatory, visible, and graded in every assessment and learning activity.

The platform implements the O×A Framework (Outcome × AI Collaboration)—a dual-scoring architecture that evaluates both what students produce and how they collaborate with AI to produce it. Every submission receives two scores: a traditional Outcome score and an AI Collaboration score, combined through configurable weights into a final grade.

Two Complementary Modes

📋

Structured Assessment

Evaluate task completion with AI collaboration in a single session. Focuses on outcome quality and AI collaboration quality at a point in time.

📚

Learning Activity

Support extended learning journeys across multiple sessions (days or weeks). Focuses on process and AI collaboration growth over time.

Learning activities span three pedagogical frameworks—Problem-Based Learning (PBL), Project-Based Learning, and Case-Based Learning—each with distinct scoring weights that prioritize learning process alongside AI collaboration.

AspectStructured AssessmentLearning Activity
PurposeEvaluate task completionSupport learning journeys
DurationSingle sessionMulti-session (days/weeks)
Primary FocusOutcome + AI collaborationProcess + AI collaboration growth
AI EvaluationPoint-in-time snapshotTrend analysis across sessions
FeedbackPost-submissionReal-time + post-completion
Pedagogical FitExams, assignments, projectsPBL, Project-Based, Case-Based

Why AlwaysAI?

The Problem: A Three-Way Failure

Educational institutions face a paradox. AI tools are ubiquitous, yet assessment systems were designed for a world without them. The three dominant responses have all failed:

🚫 Ban AI Entirely

Creates artificial conditions disconnected from real-world professional practice. Students who comply are disadvantaged; enforcement is unreliable.

🙈 Ignore AI

Evaluates only outcomes while remaining blind to how they were achieved. Students are incentivized to accept AI output uncritically.

🔍 Detect AI as Cheating

Treats AI use as academic dishonesty. Detection tools have unacceptable false positive rates, particularly for non-native speakers.

None of these approaches develop the AI collaboration competencies students need for modern professional work.

The Insight: Assessment Drives Behavior

Building on Biggs' (1996) principle that “students learn what they think they will be assessed on,” AlwaysAI applies a simple insight: if you want students to use AI correctly, assess their AI usage directly.

This transforms AI literacy from a curriculum topic into an emergent property of assessment design. Students don’t need a separate course on prompt engineering if their regular coursework requires and evaluates effective prompting, critical evaluation of AI outputs, and thoughtful human-AI task allocation.

The Philosophical Shift

AspectTraditional ApproachAlwaysAI Approach
AI useCheatingA skill to develop
ResponseDetect and penalizeMeasure and improve
AssessmentBinary (used / didn't use)Multi-dimensional quality
FocusOutput authenticityProcess quality
AssumptionAI diminishes learningAI can enhance learning

Student-Centered Education in the AI Era

True student-centeredness in the AI era means:

Withholding AI from students—or permitting it without guidance—is not student-centered. It either denies access to transformative learning tools or abandons students to discover usage patterns through trial and error.

The O×A Scoring Formula

Final Score = (w1 × Outcome) + (w2 × AI Collaboration) + (w3 × Interaction Quality)
Weights are configured by assessment philosophy to match educational goals

Four Assessment Philosophies

Different educational contexts call for different emphases:

Professional Readiness

Preparing students for AI-augmented workplaces

Outcome 50% AI 30% Int. 20%

Formative / Learning

Emphasizing AI skill development during learning

Out. 30% AI Collaboration 50% Int. 20%

High-Stakes Certification

When domain outcomes matter most

Outcome 60% AI 20% Int. 20%

AI Proficiency

Courses specifically targeting AI collaboration skills

Out. 20% AI Collaboration 60% Int. 20%

AI collaboration is always assessed—the question is how much weight it receives relative to task outcomes. Even in high-stakes certification (60% outcome), 40% of the grade still comes from how students collaborated with AI.

Six Dimensions of AI Collaboration

The AI Collaboration score is not a single number but a composite of six measurable dimensions:

20%

1. Prompt Precision

The clarity, specificity, and contextual richness of prompts. Students who provide clear context, specify constraints, and articulate actual needs receive more useful AI assistance.

Low: “Help me with my essay”
High: “Review my argument in paragraph 3 about market regulation. I’m concerned the causal chain from deregulation to increased competition isn’t fully supported. What evidence am I missing?”
20%

2. Iteration Depth

Systematic refinement through multi-turn dialogue. Expert AI users rarely accept first responses—they probe, refine, and build understanding iteratively.

Indicators: Follow-up questions, building on previous responses, progressive deepening of inquiry
20%

3. Critical Evaluation

Healthy skepticism and verification of AI outputs. AI systems can be confidently wrong; students must develop habits of verification rather than blind acceptance.

Indicators: Questioning responses, requesting verification, identifying errors, expressing appropriate doubt
20%

4. Integration / Synthesis

Combining AI assistance with personal knowledge and domain expertise. Effective AI collaboration is not outsourcing thinking—it is augmenting it.

Indicators: References to personal understanding, domain-specific contextualization, synthesis of AI output with other sources
10%

5. Metacognitive Awareness

Understanding when AI assistance is appropriate, what AI does well vs. poorly, and the respective contributions of human and AI in the collaboration.

Indicators: Explicit role allocation, AI limitation awareness, reflection on human-AI collaboration
10%

6. Efficiency

Achieving good outcomes with appropriate (neither excessive nor insufficient) AI interaction. The optimal range is typically 2–6 meaningful interactions per task.

Indicators: Avoiding redundant queries, strategic use of AI capabilities, appropriate cognitive offloading

Three-Layer Evaluation

Rather than relying on a single method, the platform evaluates through three complementary layers:

1

Pattern-Based Heuristics

Fast, consistent keyword and pattern matching against compiled regular expressions, detecting quality indicators across all six dimensions. Provides immediate, real-time scoring.

2

Linguistic Analysis

Deeper textual analysis including lexical density, speech act classification, prompt complexity scoring, and temporal pattern analysis (how prompts evolve over a session).

3

LLM-Powered Evaluation

Contextual, nuanced assessment that scores each rubric criterion with specific evidence quotes from student work, evaluates all six AI collaboration dimensions with examples, and provides actionable feedback.

Learning Activity Frameworks

For multi-session learning activities, the O×A framework adapts to three pedagogical models, each with distinct scoring weights:

FrameworkLearning ProcessAI CollaborationEngagementOutcome
Problem-Based Learning35%35%15%15%
Project-Based Learning25%30%20%25%
Case-Based Learning40%35%10%15%

Learning Process Sub-Dimensions

Learning process is itself evaluated across four equally-weighted sub-dimensions:

Exploration Depth

Breadth and depth of inquiry into the problem space. 25% weight.

Iterative Refinement

Quality of revision and improvement over time. 25% weight.

Conceptual Growth

Development of understanding across sessions. 25% weight.

Synthesis Quality

Integration of insights into coherent conclusions. 25% weight.

Diagnostic Profiles

Instead of just a score, students receive a diagnostic profile based on their performance pattern, with targeted recommendations:

Master Learner
High learning + High AI + High engagement
Ready for advanced challenges
Intensive Learner
High quality when engaged, inconsistent participation
Develop regular learning habits
🧭
Independent Learner
Good outcomes but underutilizing AI
Experiment with AI collaboration
🌱
Process Developing
Strong AI skills but shallow learning depth
Focus on deeper exploration
📈
Growth Trajectory
Clear improvement over time
Continue momentum
🏗
Building Foundations
Needs development across areas
Structured guidance needed

AI Interaction Modes & Roles

Three Interaction Modes

Students interact with AI through three modalities, all tracked and scored equally:

💬

Chat

Text-based conversation. Always available as the default interaction mode.

🎙

Voice

Real-time voice interaction via OpenAI Realtime API for hands-free exploration.

Note: Voice mode requires an advanced API (e.g., OpenAI Realtime API) and is not enabled by default. Please contact the application creator to enable this feature.

📎

Multimodal Upload

Images, PDFs, code, and data alongside text for complex analysis tasks.

Five AI Roles

The system classifies each AI interaction by the role the AI plays, helping instructors understand what students use AI for:

RoleDescriptionExample Use
ExecutorComputation, formatting, syntaxMath, code, formatting tasks
RetrieverFinding and summarizing informationResearch, literature review
InterlocutorSocratic dialogue and clarificationConcept exploration
CriticReviewing and providing feedbackContent review, editing
GeneratorProducing drafts, code, designsContent creation

Module Structure

Course content follows a competency-based progression organized across 14 weeks:

Three Phases

Destabilize
W00 – W04
Challenge assumptions, introduce new perspectives
Reconstruct
W05 – W09
Build new frameworks, apply concepts
Stabilize
W10 – W13
Integrate learning, develop mastery

Three Levels per Week

L1 — Acquire

Remember and understand. Foundational knowledge building.

L2 — Deepen

Apply and analyze. Intermediate skill development.

L3 — Create

Evaluate and create. Advanced mastery demonstration.

Data Persistence & Analytics

All student work is tracked through dual storage to ensure both session resilience and comprehensive analytics:

Google Cloud Storage

Full session snapshots for restore capability. Auto-saved every 10 seconds and on every AI interaction.

📊

xAPI Learning Record Store

Timestamped event logs for analytics and grading evidence. Complete audit trails of every learning journey.

Every student-AI interaction is captured with full prompt and response text, enabling complete audit trails of learning journeys.

Multi-Provider LLM Support

The platform supports 8 LLM providers to avoid vendor lock-in:

OpenAI Anthropic Google Gemini Mistral Cohere Together AI Groq Azure OpenAI

All providers are evaluated equally—the system does not bias scoring toward any specific model.

The Core Principle

Students who graduate without understanding how to collaborate effectively with AI are underprepared for modern professional work. By making AI collaboration mandatory and graded, AlwaysAI aligns assessment incentives with real-world demands.