HONR 46400 · Evidence-Driven Research

Studio 1 — Begin with your curiosity

Davi Moreira

What you can defend when you leave

Studio 1

Start from the one thing AI cannot supply: a curiosity you own. Commit it, stretch it with your tools, and choose the research problem worth months of your work.

Start without a tool

Research starts with something that will not leave you alone: a pattern you keep noticing, a contradiction, a decision you want to understand. This first move is yours alone. Write four lines by hand.

Start without a tool (cont.)

Start without a tool

1. One thing you genuinely want to understand, in one sentence.
2. Why an answer would matter, and to whom.
3. What you currently think the answer might be. A starting belief, not a claim.
4. One kind of evidence that could make you revise that belief.

Start without a tool (cont.)

Start without a tool

Keep what you wrote. Lesson 1 shows why this page is where research starts and why the jump on it is yours, then helps you stretch it, with AI welcome at the table, into the problem your project will chase.

The milestone ahead

Studio 1

This studio closes with Milestone 1: Your curiosity, committed, a short chapter of its own after the lessons. What it asks you to produce. Your committed curiosity with its starting belief and the evidence that would revise it, the brainstormed candidates with what you kept and rejected and why, and one chosen research problem, scored and red-teamed, with your expected answer on record.

Research begins with a jump that is yours to make. Einstein sketched discovery as a cycle: you start from experience, you leap to an idea that could explain it, and you reason down to predictions you can check. The leap has a name, abduction, and a Google DeepMind position paper by Tom Zahavy argues that today’s AI cannot make it: the tools have mastered patterns and derivations, but they have only text about the world, not the experience the jump starts from (Zahavy 2026). Take that as an argued position rather than a settled boundary. Either way the jump is yours, because you are the one who answers for the claim. This first milestone puts it on the record: the curiosity you committed by hand, the candidates you and your tools stretched it into, and the one research problem you chose to chase.

Milestone 2 sets the rules that protect this curiosity, opens the ledger that records every delegation, verifies your first AI assistance, and closes the arc by declaring your formal research question. The milestone chapter keeps the working details: the practice steps, the versioned record, and how the artifact is assessed.

The lessons in this studio

Studio 1 · Road map

• Lesson 1 — From Curiosity to a Research Problem: your committed curiosity, the brainstormed candidates with NEW elements marked and kept or rejected with reasons, your chosen research problem scored and red-teamed, and your expected answer.

From Curiosity to a Research Problem

Lesson 1 of this studio · Chapter 1

which one research problem, out of everything your curiosity could become, gets months of your work

The research decision

Chapter 1

Begin with an experience that will not leave you alone. Decide which curiosity you are willing to own, let AI widen the possibilities without choosing for you, and commit to one research problem with your reasons attached. The formal research question comes later, after your working rules and ledger exist.

The words this chapter uses

Chapter 1 · Key terms

Abduction

proposing what might explain a pattern in the first place, where deduction asks what must follow if an explanation is true and induction looks for patterns across observations.

Importance

asks who could use the answer, and for what.

Feasibility

asks whether you can reach suitable evidence with the time, skills, and permissions you actually have.

Possible contribution

asks what the project might add — and stays provisional, because you cannot know a gap is real until Studio 3 maps the existing evidence.

Each term is defined in the chapter on first use; these are the definitions as the book gives them.

A meal that cost eleven dollars in 2019 now costs sixteen

Chapter 1 · Why this decision matters

• You find a restaurant menu from 2019.
• Your first reaction may be immediate: prices near campus have exploded.
• That reaction is useful.
• It is also not yet a finding.

Start here because the first explanation arrives fast and feels complete. Ask the room what they would have concluded from those two numbers alone, and write it on the board before you move on. The next slide lists five other explanations for the same two numbers.

Research begins when the first explanation stops being the only one

Chapter 1 · Why this decision matters

• Perhaps restaurant prices rose faster than prices generally.
• Perhaps the old menu was unusually cheap.
• Perhaps portions changed.
• Perhaps fees moved out of the listed price.
• Perhaps the example is memorable precisely because it is unusual.

None of these has been checked, and that is the point: five rival readings of the same two numbers. Ask the room which one they could rule out fastest, and with what evidence. A curiosity that cannot survive this list is not ready to take months of work.

You need a curiosity strong enough to survive alternatives

Chapter 1 · Why this decision matters

• You do not need a polished question yet.
• AI can propose explanations, comparisons, affected groups, or implications you missed.
• Use that help, mark what the tool introduced, and preserve the exchange.
• The final commitment stays yours.

You will be the one who must pursue, revise, and defend the problem, which is why the commitment cannot be handed over. Marking what the tool introduced is not bookkeeping for its own sake: it is how you can later say which parts of the framing were yours. Ask the room to name one curiosity they would still chase if every tool disappeared tomorrow.

The jump is yours to make; the cycle then returns to experience

Chapter 1 · The concept

• Experience: something in the world catches your attention.
• The jump to a possible explanation is abduction, and it is yours to make.
• Derivation and verification are delegable, then checked.
• The cycle is redrawn from Zahavy’s reconstruction of Einstein’s account (Zahavy 2026).

flowchart LR
  E["Experience<br/>something in the world catches"] -- "the jump: a possible explanation<br/>(abduction, yours to make)" --> J["Ideas and assumptions<br/>what could explain it"]
  J -- "derivation<br/>(delegable, then checked)" --> P["Predictions<br/>what else should be true"]
  P -- "verification against evidence<br/>(delegable, then checked)" --> E

The notes on the arrows, marking what you may delegate and what you may not, are this book’s rather than Zahavy’s. Point at the return arrow: verification sends you back to experience, so the cycle turns rather than ends. Ask the room which of the three arrows they have already delegated to a tool this week.

Abduction creates a candidate, not proof

Chapter 1 · The concept

• Restaurant prices seem to change every third week.
• “Payday timing moves menu updates” is an abductive possibility.
• It gives you something to check.
• It does not become evidence by sounding plausible.

Experience does not logically dictate one explanation; the researcher makes a jump toward an idea that could make sense of it. Deduction asks what must follow if an explanation is true, and induction looks for patterns across observations, while abduction proposes what might explain the pattern in the first place. Ask the room for a second abductive possibility for the same third-week pattern, then ask what evidence would separate the two.

Einstein’s jump did not begin with a loud error signal

Chapter 1 · The concept

• Newtonian physics already fit the data well.
• The reorganizing idea came from thought experiments about falling observers.
• The paper calls models strong at recombining patterns, and increasingly useful for derivation.
• Zahavy argues they lack the grounding in experience the jump begins from (Zahavy 2026).
• Treat that as an argued position, not a settled boundary.

The position paper was written at Google DeepMind and uses Einstein’s route to general relativity to make the case. Position papers exist to be tested, and tools keep improving, so nobody should carry this out of the room as a proven limit on what a model can do. The practical point holds either way: you answer for the claim, so the jump is yours to make and to defend.

Once your curiosity is on the page, AI joins as a divergence partner

Chapter 1 · The concept

• Ask for alternative explanations, overlooked comparisons, or questions challenging your belief.
• Mark every person, population, outcome, mechanism, or comparison it introduces as NEW.
• Accept no novelty claim and no citation without checking.
• Keep or reject each direction with a reason.
• What the tool cannot do is care.

The jump stops being a solo performance once its first line is written, which is why the order matters: your sentence first, the tool second. You choose what you would chase even if every tool disappeared tomorrow. Ask the room for one direction they would reject, and for the reason they would write down next to it.

Reason down four moves, then stop at the problem

Chapter 1 · The concept

• Experience. What happened, or what did you notice?
• Curiosity. What do you want to understand about it?
• Topic. What area of inquiry contains that curiosity?
• Research problem. What is not yet understood well enough for the decision, explanation, or audience you care about?
• The four moves follow the standard treatment of research problems (Booth et al. 2024).

Each move strips one layer of vagueness, and you can see where it leaves by reading the four lines in order. Stop at the problem. Units, outcomes, conditions, question kind and claim reach belong to the formal declaration that closes the next studio, so anyone reaching for them here is working ahead of the material.

An answer that confirms what everyone believes teaches almost nothing

Chapter 1 · The concept

• A 2026 essay in Science, “Advancing science by designing for surprise” (Evans et al. 2026).
• Elicit informed predictions first, then compare them with what the evidence shows.
• Before you look at evidence, write the answer you currently expect.
• Name one person for whom a different answer would be genuinely surprising.
• The more surprising the claim, the more evidence it takes.

The essay starts from a blunt fact and then makes it usable: the ordering is what does the work, predictions before evidence, so you can see whether a result would move anyone. The warning travels with the tool. Surprise is a place to aim your question, never a license to inflate your answer.

Your expected answer is a planning proxy, not a measurement

Chapter 1 · The concept

Audience

the informed readers whose expectations your result would move

Expected answer

the sentence you write, before any evidence, giving the answer you currently expect

The gap the essay names sits at the level of an audience, not at the level of any one researcher, so a result that surprises only you does not qualify. Treat the written sentence as a planning proxy for that gap rather than a measure of what anyone learned, which would take evidence about the audience you do not have yet. Keep the sentence: late in the book you hold your evidence up against it.

“Does moving more actually help you sleep?” cannot be studied as written

Chapter 1 · A worked example

• It is a real feeling. It names no one and measures nothing.
• Give it edges: physical activity and sleep quality in undergraduates. That is a topic.
• Research problem: we do not know whether more daily walking goes with better-rated sleep among residence-hall first-years.
• That points at one missing piece, not a whole subject.

Walk the funnel out loud, one line per stop, and ask the room where the vagueness leaves. The topic still contains no unknown; the problem does, because it names what we do not know. Stop at the problem, since the formal question that fixes units, outcome and reach is the close of the next studio, after your rules and ledger exist.

The sleep problem clears all three tests, with contribution still provisional

Chapter 1 · A worked example

Then test the problem three ways (Hulley et al. 2013).

• Importance: anyone advising sleep-deprived first-years wants the answer.
• Feasibility: step counts and a nightly rating are collectable in a few weeks.
• Contribution: studied broadly in adults, far less in this dorm-living first-year group.
• Contribution stays provisional until Studio 3 maps the existing evidence.
• The funnel and its tests follow the standard treatment of research problems (Booth et al. 2024).

Contribution stays provisional, because you cannot know a gap is real until Studio 3 maps the existing evidence, so score it honestly now and revisit it then. Feasibility means the access, data and time you actually have, not the ones you wish you had. The curiosity is now a problem you could defend in front of someone who wants to know why this one and not another.

The problem sentence names groups you can build out of the data

Chapter 1 · A worked example

• The block builds this example’s data and prints the numbers the section quotes.
• Watch the two edge groups: more than 8,000 steps, fewer than 4,000.
• The middle group is neither, and the question does not ask about it.
• Change one input and watch which sentence above stops being true.

import numpy as np, pandas as pd
SEED = 464
rng = np.random.default_rng(SEED)

# Sixty residence-hall first-years: average daily steps and a 1-10 sleep rating.
steps = rng.normal(6200, 2300, size=60).clip(1000, 15000)
sleep = np.clip(4.6 + 0.00035 * steps + rng.normal(0, 1.1, size=60), 1, 10)

group = np.where(steps > 8000, "more than 8,000 steps",
                 np.where(steps < 4000, "fewer than 4,000 steps", "in between"))
print(pd.DataFrame({"group": group, "sleep rating": sleep})
      .groupby("group")["sleep rating"].agg(["size", "mean"]).round(2).to_string())
print("\nthe question asks about the two edge groups; the middle is neither")

Sixty residence-hall first-years, average daily steps and a one-to-ten sleep rating. SEED is 464, so every run in the room produces the same numbers. What matters is not the size of the difference between the edge groups; it is that they exist in the data at all. Ask the room to raise the noise on the sleep rating and report which sentence from the previous slides breaks first.

An AI failure case

Chapter 1

Where the tool failed

You ask a tool to “turn my curiosity about sleep and health into a research question.” It replies, with total confidence: “Does blue-light exposure before bed reduce REM sleep in college undergraduates? Studies show blue light suppresses melatonin.” It sounds authoritative, and it is wrong for you.

How it failed

Chapter 1 · An AI failure case

• It fails in two ways.
• First, it silently changed your scope.
• You never chose blue light or REM sleep, yet the question now hangs on both.
• That is a silent scope change, and you catch it by reading the tool’s units, outcome, and population against the ones you committed.
• Yours were steps and a sleep rating, not screens and REM.

It fails in two ways. First, it silently changed your scope. You never chose blue light or REM sleep, yet the question now hangs on both. That is a silent scope change, and you catch it by reading the tool’s units, outcome, and population against the ones you committed. Yours were steps and a sleep rating, not screens and REM. Second, the melatonin line is stated as settled fact with no source. Before repeating it, you find and open a real study yourself, and if you cannot, you treat it as confident fabrication and drop it.

Do not delegate

Chapter 1

This stays yours

Three calls stay yours alone. Which problem you commit to. How you word the question, including its units, outcome, and population. And the judgment on whether it is important and genuinely new. A tool can widen your options and attack your choice. It cannot make the choice, because only you carry the reasons you will have to give when someone asks why this problem and not another.

It is your turn

Chapter 1 · Your move

1. Copy your opening move’s four lines here, date them, and label them version zero: the first entry of your project’s record.
2. Take the strongest curiosity you now hold and walk it down the funnel in writing: experience, curiosity, topic, research problem.
3. Run the ownership test on the problem you are converging on: would you still chase it if every AI tool disappeared tomorrow, and is the scope one you chose rather than one the tool drifted you into?
4. Score it on importance, feasibility, and contribution, one honest sentence each.
5. Write a second candidate problem from the same curiosity and scope it the same way.
6. Under the problem you kept, write your expected answer in one sentence, and name one real person or role for whom a different answer would be genuinely surprising.
7. Ask an AI to attack your chosen problem on all three tests, and keep only the objections that survive your own reasoning.
8. Finish by handing your problem sentence to someone who knows nothing about your project.

Work it in the companion notebook with Chapter 1 open beside it. Log every delegation in your AI Research Ledger.

The chapter’s numbered steps, in the book’s own order. The AI prompts each step carries live in the companion notebook.

Milestone 1: Your curiosity, committed

Studio 1 closes here

What the lessons handed you becomes one artifact you can defend.

What this milestone produces

Milestone 1

The artifact

What this milestone produces. Your committed curiosity with its starting belief and the evidence that would revise it, the brainstormed candidates with what you kept and rejected and why, and one chosen research problem, scored and red-teamed, with your expected answer on record.

What you bring

Milestone 1 · Check before you start

• Lesson 1 — From Curiosity to a Research Problem: your committed curiosity, the brainstormed candidates with NEW elements marked and kept or rejected with reasons, your chosen research problem scored and red-teamed, and your expected answer.

The practice

Milestone 1 · In the studio

1. Copy your opening move’s four lines as the artifact’s version zero, dated.
2. Widen the curiosity: write two candidate directions yourself, then run the structured brainstorm and mark everything it introduced as NEW.
3. Choose the research problem and score it honestly on importance, feasibility, and contribution, one sentence each.
4. Write your expected answer and name one real person or role for whom a different answer would be genuinely surprising.
5. Keep the trail: candidates kept and rejected, each with its reason, and the NEW elements you adopted as your own.

The four rails, here

Milestone 1 · Every studio, these four

Ethics, permissions, and data exposure

Note now whether chasing this curiosity would require people, and flag it before anything is designed.

Evidence, provenance, and reproducibility

Your starting belief is a claim to check, not to defend; write it so evidence can move it.

AI activity, verification, and human decisions

Brainstorm freely, and mark every AI stretch NEW; start a simple exchange log now, and Studio 2 turns it into your AI Research Ledger.

Uncertainty, claim boundary, and revision history

Write what would change your mind while nothing is yet at stake; that sentence is the seed of every later boundary.

A version, not a pass

Milestone 1

How the record works

Your milestone artifact is a dated, numbered version with the reason for the version attached. When later evidence changes it, you write the next version rather than editing the last one, because the sequence of changes is itself part of your research record.

The one rule

AI is your arm and your research assistant, not your brain.

AI can review AI, and a second model is a real auditor of the first. The last decision is always human.

Studio 1 in EDR|AI · Milestone 1 · Verification Guide