The Innovation Lab Project Consultation

Home / AI Lab/ AI-Based Education & Workshops

TIL AI-Based Education & Workshops

Rather than learning how to use AI tools,
We learn how to think better with AI

Training in which AI produces the results for you may yield quick answers. But it does not create meaningful learning or transfer beyond knowing how to use AI tools. The Innovation Lab's programs are designed so that people define the problem first, examine it more deeply through AI cross-feedback, and retain final judgment and responsibility.

We connect individual efficiency to organizational problem-solving capability, and a one-time output to learning transfer that carries over to new challenges.

Ask about Training Tailored to Our Organization Explore Programs

Human First · Problem-Driven · Evidence-Based · Transfer-Oriented

01 · Why

Using More AI Tools Alone Will Not Change an Organization

Completing an individual task faster is not the same as helping an organization discover more important problems and make better decisions. Organizations must first define the problems they need to solve and the value they need to create, then position AI within the solution process.

SHIFT 01

From Individual Productivity to Organizational Value

Before asking which tool to use, we ask what needs to change. We define organizational bottlenecks, customer problems, and the decisions that must be made, then connect AI as a means of solving them.

SHIFT 02

From Producing Results to Building Capability

A strong result produced by AI and the capability a person retains can be two different things. Learning becomes an organizational capability only when people can explain, judge, and apply it to new problems without AI assistance.

The More AI Becomes a Shortcut, the More Carefully Learning Must Be Designed

Copying prompts to obtain results makes tasks easier, but it can lead to cognitive outsourcing, uncritical acceptance, diminished control over outputs, and less diversity in group thinking. AI should not be an answer provider, but a collaborative partner that helps explore counterarguments, blind spots, and alternatives.

Beyond people who perform well when AI is available,
We develop people and organizations that make better judgments even after learning with AI.

02 · How

The Innovation Lab's Approach to AI Education: HAMMA

Human-AI Mentoring Motif Approach

HAMMA structures the roles and sequence of people and AI so that AI does not provide the first answer or the final conclusion. It begins with human thinking and concludes with human judgment.

The five stages of the Human-AI Mentoring Motif Approach: Human First, AI Cross-feedback, Human Critical Loop, Human Decision, and AI Assist
Human First → AI Cross-feedback ↔ Human Critical Loop → Human Decision → AI Assist
01 · Getting Started
BEFOREAI creates the draft and a person refines it
AFTERA person creates the draft and hypotheses, and AI provides cross-feedback
02 · Goal
BEFOREFocused on saving time and increasing output
AFTERFocused on improving the quality of results, collaboration, and learning capability
03 · Thinking
BEFORECognitive outsourcing and uncritical acceptance
AFTERMaintaining critical thinking and human agency through learning friction
04 · Creativity
BEFORERaising the baseline quality of individual outputs
AFTERPreserving individual perspectives and the diversity of ideas within the group
05 · Transfer
BEFORELearning loss revealed when AI support is removed
AFTERConceptual transfer through independent explanation, judgment, and application

03 · Evidence

Designed on Evidence, Not Good Intentions Alone

Generative AI helps people complete tasks quickly, but independent thinking can weaken if it also takes over the planning, validation, and reflection that people should perform. The research does not conclude that AI use itself should be reduced. Rather, it points to designing the sequence and conditions of collaboration so that people attempt the work first, regulate AI assistance, and reclaim final judgment and explanation.

TIL's HAMMA applies empirical research on generative AI, learning, creativity, and independent performance, together with learning-science principles such as productive failure, self-explanation, and retrieval, to this process.

EVIDENCE 01

AI May Help Complete a Task, but It Does Not Necessarily Build Human Capability

Without guidance, learners using AI may skip the process of diagnosing the problem and evaluating the validity of assistance, instead seeking immediately actionable answers. Performance may improve while AI is available, yet self-regulated learning—including planning, monitoring, and reflection—can weaken. This is why we need AI that prompts learners to answer first, offers hints, and asks them to correct themselves instead of simply providing the answer.

View More Research Evidence
  • High scores achieved with AI are different from the capability retained without it. In a field RCT involving high school mathematics, Bastani et al. found that the group using unrestricted GPT earned higher practice scores but scored 17% lower than the control group on an exam taken without AI. GPT Tutor, which provided step-by-step hints instead of answers, reduced the decline in independent performance to the control group's level. The finding is not that AI assistance should be removed, but that constraints should be placed on how assistance is provided.
  • When AI takes over a task, learners may scrutinize their own thinking less. In a randomized writing experiment by Fan et al., GPT-4 support produced greater essay improvement but did not increase measured knowledge acquisition or transfer to other domains. The researchers' observation of “metacognitive laziness” warns against interpreting high-quality outputs as evidence of learner growth.
  • The way people seek help from AI has also changed. In an experiment in which Chen et al. analyzed multimodal data from the writing and revision processes of 38 university students, the ChatGPT group showed nonlinear help-seeking that skipped the feedback-evaluation stage more often than the human-expert group and asked more operational questions that could be acted on immediately. The finding suggests that as AI becomes more accessible, diagnosing questions and evaluating feedback must be made explicit parts of the learning process.
  • When a prior answer and self-correction are required, AI becomes a learning mechanism rather than a “crutch.” In a study by Lee et al., the guided structure “my answer first → hints instead of the answer → self-correction” produced better results than general ChatGPT use across several measures related to behavioral engagement, knowledge, and self-regulation. This provides evidence that requiring an initial response and self-correction can change how learners engage.
  • TIL has people produce the first draft and hypotheses. AI provides questions, hints, and counterexamples instead of answers, and people explain the reasons for their revisions. The goal is not performance while using AI, but judgment that remains after AI is gone.
Key SourcesBastani et al., PNAS, 2025 ↗Fan et al., BJET, 2024 ↗Chen et al., Computers & Education, 2025 ↗Lee et al., 2024 ↗
EVIDENCE 02

AI Can Improve Individual Ideas, but It Can Also Make Everyone's Thinking More Alike

Generative AI can help make individual ideas more novel and useful. The benefit is especially large for people who struggle to generate ideas from scratch. But if everyone begins with suggestions from the same AI, the average individual level may rise while the range of perspectives across the group becomes more similar.

View More Research Evidence
  • Improvements in individual creativity and reductions in group diversity can occur at the same time. In an experiment by Doshi & Hauser, writers who received AI-generated ideas earned higher average ratings for novelty and usefulness, with greater benefits for writers whose initial creativity ratings were lower. However, stories written with AI support became more similar to one another. The finding suggests that AI can narrow performance gaps between individuals while also converging a group's ideas around a small number of frames.
  • When AI was introduced made a difference in independent creativity. In a randomized experiment by Wong & Qiu, the first-task advantage of the group allowed to use AI freely from the outset did not persist in a subsequent task completed without AI. By contrast, participants who first generated raw ideas themselves and then used AI as a partner for improvement and evaluation received higher creativity ratings even on the subsequent independent task.
  • Taken together, the two studies show that the timing of AI involvement matters. AI is more useful for refining distinct perspectives created by people than for replacing the starting point of ideation. TIL has individuals and teams create their own drafts and hypotheses before receiving AI feedback, improving individual creativity while preserving the group's diversity of perspectives.
Key SourcesDoshi & Hauser, Science Advances, 2024 ↗Wong & Qiu, 2026 ↗
EVIDENCE 03

Traces of an Initial Attempt—not a Finished AI Answer—Build Problem-Solving Ability

Seeing a finished AI answer first reduces opportunities to consider how to define the problem and which alternatives to compare. The process of first creating hypotheses and solutions—even if they are rough or wrong—reveals what learners know and what they have overlooked. Connecting explanation and AI feedback afterward creates a problem-solving structure that can be applied to new challenges.

View More Research Evidence
  • Productive-failure research does not say that failing first always results in better learning. It shows that conceptual learning and transfer may improve when learners create multiple representations and solution methods before instruction, then connect those attempts to correct concepts through explanation and feedback. An initial attempt activates the schema needed to engage more deeply with subsequent feedback.
  • What matters is not the number of solution methods, but the experience of representing and solving a problem in different ways. In a study by Kapur, Saba & Roll, the diversity of solution methods students generated during initial problem solving predicted subsequent learning. However, because all students in the study received productive-failure instruction, we do not expand this finding into a simple causal effect.
  • Early evidence also suggests that a brief period of productive struggle before asking AI may affect subsequent transfer. In a preprint study by Akgün & Toker, participants who first attempted a statistics problem on their own and then used ChatGPT scored higher in a similar new situation than those who asked ChatGPT immediately. Because this was a small-sample, non-peer-reviewed study, we use it only as supporting evidence.
  • TIL has people define the problem and generate multiple hypotheses and alternatives before using AI's counterexamples, omitted variables, and alternative interpretations to examine them. AI is not a shortcut that removes the need for prior thought and effort, but a feedback partner that turns an initial attempt into deeper problem solving.
Key SourcesKapur, Saba & Roll, 2023 ↗Sinha & Kapur, 2021 ↗Akgün & Toker, 2024 preprint ↗
EVIDENCE 04

People Must Remain Responsible for Planning, Monitoring, and Evaluation While Using AI

Simply declaring that people retain final judgment does not create agency. Learning must include procedures for setting goals, monitoring solution strategies, and evaluating AI responses and one's own understanding. Critical thinking likewise develops not through the number of interactions with AI, but through comparing evidence and explaining how and why a judgment was revised.

View More Research Evidence
  • Metacognitive supports must be deliberately designed even in generative AI environments. In a four-week quasi-experiment involving 68 university students, Xu et al. found that adding explicit support for task strategies and self-evaluation improved measures of self-regulated learning processes and learning experience. Because the difference in achievement between groups was not significant, this is not evidence that higher grades are guaranteed, but it does demonstrate the need for mechanisms that help learners regulate their own use of AI.
  • Planning, monitoring, and evaluation are not innate dispositions but teachable strategies. In a meta-analysis of 48 metacognitive-strategy interventions, de Boer et al. found that interventions teaching all three components produced academic-performance effects both immediately after the intervention and at follow-up. This general evidence from learning science suggests that preserving human judgment in the AI era requires training the decision-making process itself.
  • Critical thinking does not emerge automatically through conversation alone. In Abrami et al.'s meta-analysis, structured dialogue, real-world problems and cases, and mentoring were positive conditions for critical-thinking instruction. Rather than simply having learners converse more with AI, learning must be structured so that they apply AI's counterarguments to real challenges, compare evidence, and explain why they revised their judgments.
  • Adding more explanation alone does not necessarily reduce overreliance on AI. In an experiment with 199 participants, Buçinca, Malaya & Gajos found that a cognitive forcing function—which prevented participants from immediately following AI recommendations and required deliberation—reduced overreliance on incorrect AI recommendations more than a simple explanation condition. However, the design that reduced overreliance the most received the lowest subjective ratings, and participants with a higher propensity for reflection benefited more. This is evidence that desirable learning friction is necessary, but its usability costs and individual differences must also be considered in the design.
  • TIL has people first define goals and evaluation criteria, review the objections and alternatives proposed by AI, and record what they adopted, revised, or rejected and why. Before teaching techniques for using AI, we train the ability to evaluate and regulate its assistance.
Key SourcesXu et al., BJET, 2025 ↗de Boer et al., Educational Research Review, 2018 ↗Abrami et al., Review of Educational Research, 2015 ↗Buçinca, Malaya & Gajos, PACM HCI, 2021 ↗
EVIDENCE 05

Learning Lasts When You Retrieve, Explain, and Apply It After Stepping Away from AI

Familiarity when rereading AI-generated material can differ from genuine understanding. To know whether learning has lasted, learners must close the AI, retrieve key concepts from memory, explain them in their own words, and apply them to a different problem. That is why HAMMA ends not with submitting an output, but with unaided explanation and reapplication.

View More Research Evidence
  • Retrieving information from memory is more beneficial for retention than rereading it. A meta-analysis by Adesope, Trevisan & Sundararajan found that retrieval activities such as practice tests generally helped learners retain material longer than rereading. This does not mean adding more tests; it means incorporating low-stakes activities in which learners hide the AI response and recall, compare, and apply the key content.
  • Explaining retrieved knowledge in one's own words helps reveal the connection between procedures and concepts. In Rittle-Johnson, Loehr & Durkin's meta-analysis of mathematical self-explanation, self-explanation prompts showed small to moderate immediate benefits for procedural knowledge, conceptual knowledge, and some procedural transfer. Questions and feedback that elicited good explanations were important, while evidence from classroom contexts and on delayed retention was limited.
  • Retrieval practice has repeatedly shown benefits not only in controlled laboratories, but also in real schools and classrooms. A systematic review by Agarwal, Nunes & Blunt included 50 experiments and 5,374 participants and found moderate or large benefits in 57% of 49 effect sizes. Not every experiment found a large effect, and studies from non-Western countries accounted for only 6% of the total, but the evidence for applying unaided retrieval in real educational settings continues to grow.
  • The three studies support a single learning flow: learners must go beyond rereading AI answers to retrieve information from memory, explain it in their own words, and apply it to a new problem. TIL includes this procedure at the end of each course to determine whether learning transfer—not merely an output—remains.
Key SourcesAgarwal, Nunes & Blunt, Educational Psychology Review, 2021 ↗Rittle-Johnson, Loehr & Durkin, ZDM Mathematics Education, 2017 ↗Adesope, Trevisan & Sundararajan, Review of Educational Research, 2017 ↗

Each study was conducted with a particular age group, task, duration, and set of AI-use conditions. TIL does not generalize individual findings directly to every context; we review multiple empirical studies together with learning-science principles and apply them as educational design principles.

04 · Programs

Learn through Real Organizational Problems and Create Outputs You Can Use Immediately

This is not a tool exercise built around predetermined examples. Participants bring their organization's operations, research and product, customer, market, and business challenges, analyze them using HAMMA, and turn them into outputs that can support real decisions and follow-up execution.

01 · HAMMA

AI-Powered Business Strategy Workshop

Read the landscape, make choices, and turn them into action.

Interpret the organization's business environment, identify key strategic issues and alternatives, and develop them into an action plan.

View Details → 02 · HAMMA

Trend-Sensing-Based Insight Development

Turn scattered signals of change into the next opportunity.

Use private- and public-sector data, APIs, and NextBM200 to build a system for detecting change signals and monitoring them continuously.

View Details → 03 · HAMMA

Customer Insight Discovery

Discover hidden problems and opportunities to create value.

Use deep research, synthetic users, and simulated interviews as exploration tools to design customer hypotheses for real-world validation.

View Details → 04 · HAMMA

Business Model and Go-to-Market Strategy Development

Design distinctive value and a path to growth.

Diagnose the business model and industry environment, then connect target markets, positioning, channels, and partners into an execution strategy.

View Details → 05 · HAMMA

AX Business Transformation Workshop

Turn a vague idea of AI adoption into a complete transformation strategy.

Connect customer value, human and AI roles, data, economics, risk management, and an execution roadmap into a unified strategy.

View Details →

05 · Curriculum

Detailed Program Structure

Expand each course to review its detailed structure. In actual delivery, the scope and sequence are adjusted to reflect the organization's goals, participants' roles, and real workplace challenges.

01

AI-Powered Business Strategy Workshop

Strategic-issues map · Strategic alternatives · Priority initiatives · Action plan

This is a hands-on strategy course for reading the landscape around an organization, defining core issues, and deciding which alternatives to pursue. People interpret the situation, while AI presents counterarguments, blind spots, and alternatives; people make the final decisions on strategy and execution criteria.

Common Learning Flow

  1. 1. Reading the Strategic Landscape
    • Connecting the external environment, competitive landscape, customer change, and internal capabilities
  2. 2. Defining Strategic Issues
    • Distinguishing symptoms from causes and designing core strategic questions
  3. 3. Designing Strategic Alternatives
    • Human-generated alternatives followed by AI red-team review
  4. 4. Selection and Execution
    • Structuring priorities, success conditions, risks, and execution initiatives

Optional Delivery Formats

  • AI-Powered Jeondopan Workshop— Total 6~8 sessions, 15~20 hours
  • Jeondopan Seminar— Half-day, 4 hours
  • AI-Powered One-Day Management and Business Planning Workshop— 8 hours
  • Supplementary materials: 《Strategy Is a Tool: The Power to Read the Strategic Landscape》, 《The Queen of Strategy》
02

AI-Powered Trend-Sensing and Insight Development Workshop

Change-signal map · Opportunity hypotheses · Priority insights · Monitoring board

Day 1 — Identify Signals of Change and Interpret Them as Trends

  1. 1. A Trend-Sensing Perspective
    • Distinguishing fads, phenomena, and structural change
    • Setting criteria for signals that matter to the organization and its business
  2. 2. Data Sources and Exploration Design
    • Understanding private- and public-sector datasets and APIs
    • Collecting signals with NextBM200 and MCP
  3. 3. HAMMA-Based Signal Analysis
    • Reviewing human-generated change hypotheses against AI-identified counter-signals and blind spots
  4. 4. Trend-Sensing Canvas
    • Completing a map of promising signal clusters and initial trends

Day 2 — Turn Signals into Business Opportunities and Continuous Insight

  1. 1. From Signals to Opportunity Hypotheses
    • Connecting changes in customers, markets, technology, and regulation
  2. 2. Strengthening the Quality of Insights
    • Separating facts, interpretations, and hypotheses and assessing the strength of evidence
  3. 3. Prioritizing Opportunities
    • Assessing impact, urgency, uncertainty, and organizational fit
  4. 4. Opportunity Monitoring Board
    • Setting key metrics, update cycles, and threshold signals

Supplementary materials: 《Next Business Model 2026 — Overview & Industry Edition》 · Services used: NextBM200 Trend Explorer, MCP

03

AI-Powered Customer Insight Discovery Workshop

Customer problem hypotheses · Evidence-based personas · Customer journey map · Value proposition opportunities

Day 1 — Understand Customers through Questions and Evidence

  1. 1. Starting with Customer Insight
    • Distinguishing customer information from customer insight
  2. 2. Customer Problem Hypotheses
    • Structuring relationships with the Turtleback Framework and Flower Model
  3. 3. AI-Powered Customer Research
    • Using public sources, deep research, and synthetic-user data from 9 countries
  4. 4. HAMMA Cross-Validation
    • Reviewing opposing perspectives, extreme users, missing context, and bias

Day 2 — Turn the Voice of the Customer into a Value Proposition

  1. 1. Evidence-Based Personas
    • Defining behavior, purpose, context, pain points, and selection criteria
  2. 2. Customer Journey and Moments That Matter
    • Identifying drop-off points, unmet needs, and moments that matter
  3. 3. Opportunity Areas and Value Propositions
    • ERRC·Strategy Canvas and AI red-team validation
  4. 4. Customer Insight Action Plan
    • Designing questions and next actions for validation through real customer research

Synthetic users and simulated interviews are not definitive substitutes for real customers. We use them as exploration tools to broaden the range of hypotheses and uncover blind spots before real-world validation.

04

AI-Powered Business Model and Go-to-Market Strategy Development

Business model diagnostic · Improved business model · Go-to-market strategy · Growth roadmap

Day 1 — Diagnose the Business Model and Redesign the Value Structure

  1. 1. A Perspective on Business Models
    • Connecting customer value, delivery model, revenue structure, and core capabilities
  2. 2. Target Markets and Priority Customers
    • TPM Circle and market–organization fit assessment
  3. 3. Refinement Based on BM ZEN
    • Structuring customer problems, value propositions, the value chain, and revenue logic
  4. 4. Current Business Model Diagnosis
    • Using BM HealthCheck and IndustrySight

Day 2 — Turn the Plan into Market-Entry and Growth Execution Strategies

  1. 1. Exploring Alternatives and Red Teaming
    • Reviewing alternative models, competitive perspectives, failure conditions, and blind spots
  2. 2. PM Go-to-Market Strategy
    • Refining positioning, messaging, channels, partners, pricing, and the offer
  3. 3. Strategic Stress Testing
    • Reviewing competitive responses, failure scenarios, and omitted variables
  4. 4. Growth-Stage Roadmap
    • Setting goals and metrics for the exploration–validation–entry–scaling stages

Services used: BM HealthCheck, IndustrySight, AI ToolBox, BM Explorer (in development) · Frameworks: BM ZEN toolkit, TPM Circle, PM Go-to-Market Strategy

05

AI-Powered AX Business Transformation Workshop

AX transformation thesis · Human–AI collaboration structure · Data and performance flywheel · Execution roadmap

A two-day on-site workshop that moves beyond AI technology PoCs to connect customer value, human and AI roles, data, performance improvement, economics, risk management, and an adoption roadmap into a unified business transformation strategy.

Day 1 — Turn AI Adoption Ideas into a Transformation Blueprint

  1. 1. The Perspective of AX Business Transformation
    • The difference between introducing automation or AI features and transforming the business
  2. 2. Transformation Initiatives and the Transformation Thesis — Why ①
    • Defining the core problem, transformation goals, and success criteria
  3. 3. Customer Value and Economic Opportunity — Why ②
    • Defining target customers, core value, and economic opportunities
  4. 4. Human and AI Roles — What ①
    • Distinguishing human judgment and accountability from AI-supported work

Day 2 — Build an Actionable Business Strategy and Roadmap

  1. 1. Data and Performance — What ②
    • Designing a data–learning–response–performance flywheel
  2. 2. AI Models, Operating Logic, and Governance — How ①
    • Define AI roles, quality standards, risks, and human approval points
  3. 3. Economic Engine and Execution Roadmap — How ②
    • Developing revenue, cost, adoption-impact, and phased execution plans
  4. 4. Sharing and Feedback
    • Team strategy presentations, cross-feedback, and confirmation of follow-up initiatives

Centering on the participating organization's real operations, research, product, and business challenges, the workshop brings departmental perspectives together on a single canvas and leaves participants with a draft AX transformation strategy for follow-up decisions at the end of the two days.

06 · Value

Beyond Knowing How to Use AI: Capabilities That Stay with the Organization

Build the capability to define problems, make better judgments, and transfer learning to new challenges—beyond short-term efficiency gains.

Corporate Employees

Think more deeply and make better decisions with AI.

Define the organization's core problems, critically validate AI's suggestions, and turn them into actionable strategies and solutions.

HRD Professionals

Design AI learning that transfers to the workplace.

Improve learning outcomes by focusing on human agency, collaboration, problem-solving ability, and learning transfer.

Startups

Explore uncertainty and validate growth hypotheses.

Read trends and customers, diagnose the business model, and complete a market-entry strategy and growth roadmap.

Public Institutions

Create responsible solutions grounded in evidence and citizens' perspectives.

Use public data and AI while keeping evidence validation and final judgment human-led.

07 · Operation

We Design the Course and Delivery Around Your Organization's Challenges

Delivery FormatOn-site training for businesses and public institutions, team workshops, seminars, and project-based learning
Learning DesignAdjusting the program to reflect the organization's goals, participants' roles, and real challenges
Delivery MethodConcept lectures, team discussions, practice with AI ToolBox and proprietary frameworks, and cross-feedback
Key DeliverablesCanvases, diagnostic results, strategic proposals, insight boards, and execution roadmaps
Schedule · Participants · LocationConfirmed through consultation after reviewing the organization's challenges and delivery requirements

Every organization faces different challenges. Rather than repeating a fixed list of tools, we build the most appropriate course and exercises around the problems and decisions that matter now.

Contact

Not training where AI produces the results for you,
Learning that helps people and organizations develop better questions and judgments

HAMMA is The Innovation Lab's approach to AI education, harnessing AI's speed while protecting human agency, creativity, problem-solving ability, and learning transfer. Design a program that starts with the problems your organization must solve and leaves you with real strategies and action plans.

Inquire about Education & Workshops Explore AI Lab

info@thelab.center · 02-3454-1108

ANSWER ENGINE

Frequently Asked Questions

Quickly understand The Innovation Lab's role and methodology from the perspective of your challenge.

What is HAMMA?

HAMMA is The Innovation Lab's approach to AI education: people first formulate the problem and hypotheses, critically review AI cross-feedback, make the final judgment themselves, and use AI to support synthesis and execution.

How is this different from conventional AI tool training?

We go beyond teaching the features of a particular generative AI tool or how to use prompts. We first define the organization's real problems, then examine strategy, trends, customers, business models, and AX transformation challenges with AI to develop both workplace-ready outputs and the judgment needed to use them.

What types of organizations can participate?

We work with corporate employees and HRD teams, startups, and public institutions. We review the organization's goals, participants' roles, and real-world challenges in advance, then design the engagement as on-site training, a team workshop, a seminar, or project-based learning.

What deliverables will we have after completing the program?

Depending on the selected course, deliverables include a strategic-issues map and action plan; a board for monitoring change signals and opportunities; customer problem hypotheses and a customer journey map; a business model diagnostic and go-to-market strategy; and an AX transformation blueprint and execution roadmap.